SOCIAL COGNITION, MOTIVATION, AND INTERACTION: HOW DO PEOPLE RESPOND TO THREATS IN SOCIAL INTERACTIONS?

EDITED BY: Eva Jonas and Christina Mühlberger PUBLISHED IN: Frontiers in Psychology

#### *Frontiers Copyright Statement*

*© Copyright 2007-2017 Frontiers Media SA. All rights reserved. All content included on this site, such as text, graphics, logos, button icons, images, video/audio clips, downloads, data compilations and software, is the property of or is licensed to Frontiers Media SA ("Frontiers") or its licensees and/or subcontractors. The copyright in the text of individual articles is the property of their respective authors, subject to a license granted to Frontiers.*

*The compilation of articles constituting this e-book, wherever published, as well as the compilation of all other content on this site, is the exclusive property of Frontiers. For the conditions for downloading and copying of e-books from Frontiers' website, please see the Terms for Website Use. If purchasing Frontiers e-books from other websites or sources, the conditions of the website concerned apply.*

*Images and graphics not forming part of user-contributed materials may not be downloaded or copied without permission.*

*Individual articles may be downloaded and reproduced in accordance with the principles of the CC-BY licence subject to any copyright or other notices. They may not be re-sold as an e-book.*

*As author or other contributor you grant a CC-BY licence to others to reproduce your articles, including any graphics and third-party materials supplied by you, in accordance with the Conditions for Website Use and subject to any copyright notices which you include in connection with your articles and materials.*

> *All copyright, and all rights therein, are protected by national and international copyright laws.*

*The above represents a summary only. For the full conditions see the Conditions for Authors and the Conditions for Website Use.*

ISSN 1664-8714 ISBN 978-2-88945-374-0 DOI 10.3389/978-2-88945-374-0

# About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

# Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

# Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

# What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

# **SOCIAL COGNITION, MOTIVATION, AND INTERACTION: HOW DO PEOPLE RESPOND TO THREATS IN SOCIAL INTERACTIONS?**

Topic Editors: **Eva Jonas,** University of Salzburg, Austria **Christina Mühlberger,** University of Salzburg, Austria

According to the Loop2Loop Model, threat results from a discrepancy between the reality (perception of the actual situation) and one's motives, the reality and one's cognitive focus, or one's motive and one's cognitive focus (1). These discrepancies can then set in motion a process by which people's motivational-affective states (2) and cognitions (3) are affected, which ultimately affects their behavior (4). Further, the model expresses how discrepancy can alter interactions (5) more broadly to the extent that discrepancies alter the effective situations of interaction partners and their behaviors toward each other (see Editorial by Jonas & Mühlberger, 2017, in this e-book).

Cover image: Schematic illustration of a social interaction. Image by Vicky König.

If we want to understand people's responses to threats in social interactions we can distinguish between three levels of analysis:

On a social level of analysis we can describe people's interpersonal behavior, on a cognitive level we can identify corresponding information processing mechanisms, and on a neural level we can specify neural systems, which underlie these processes.

In this Research Topic we want to present research connecting these three levels of analysis and propose their functional interconnection in social interaction.

We propose that threats in social interactions activate basic motivational processes, which manifest in neural processes related to behavioral inhibition vs. activation in a social situation. This shapes our attention to new information, and affects our cognitions about social identities, belief systems and worldviews. These changes in social cognition in turn affect people's behavior in social interactions and lead to corresponding reactions on behalf of the interaction partner. Thus, we assume that people's reactions to threat in interactions can be described as sequences of broader attentional processes resulting from basic motivational tendencies leading to specific social cognitions and subsequent behavior within social interactions. We can analyze this sequence in order to contribute to a better understanding of social interactions.

The three levels of analyses (social, cognitive, neural) shed light on social interactions from different angles:

On the social level we can analyze how the behaviors of the interaction partners mutually affect each other and how this is accompanied by specific cognitive, emotional and motivational processes. On the cognitive level we can analyze people's perception of a social situation leading to attentional and reasoning processes with regard to their interaction partner/s, which may be accompanied by certain emotional and motivational processes and determines the behavior towards the partner/s. Finally, we can focus on the neural mechanisms underlying cognitive, emotional, and motivational processes in social interactions.

**Citation:** Jonas, E., Mühlberger, C., eds. (2017). Social Cognition, Motivation, and Interaction: How Do People Respond to Threats in Social Interactions? Lausanne: Frontiers Media. doi: 10.3389/978-2-88945-374-0

# Table of Contents

*06 Editorial: Social Cognition, Motivation, and Interaction: How Do People Respond to Threats in Social Interactions?* Eva Jonas and Christina Mühlberger

# **I. Threat as discrepancy between motive and reality (perception of the actual situation)**

*10 Complaints as starting point for vicious cycles in customer–employeeinteractions*

Eva Traut-Mattausch, Sara Wagner, Olga Pollatos and Eva Jonas


Florian E. Klonek, Amelie V. Güntner, Nale Lehmann-Willenbrock and Simone Kauffeld

*53 Unstable power threatens the powerful and challenges the powerless: evidence from cardiovascular markers of motivation*

Daan Scheepers, Charlotte Röell and Naomi Ellemers


Johannes Keller, Ruth Mayo, Rainer Greifeneder and Stefan Pfattheicher

*92 Attracted to power: challenge/threat and promotion/prevention focus differentially predict the attractiveness of group power* Annika Scholl, Claudia Sassenrath and Kai Sassenberg

# **II. Threat as discrepancy between cognitive focus and reality (perception of the actual situation)**

*101 Threat to Freedom and the Detrimental Effect of Avoidance Goal Frames: Reactance as a Mediating Variable*

Daniela Niesta Kayser, Verena Graupmann, James W. Fryer and Dieter Frey

*114 Legitimate vs. illegitimate restrictions – a motivational and physiological approach investigating reactance processes*

Sandra Sittenthaler, Christina Steindl and Eva Jonas


# **III. Threat as discrepancy between motive and cognitive focus**


Tsachi Ein-Dor

*164 Inhibition Underlies the Effect of High Need for Closure on Cultural Closed-Mindedness under Mortality Salience*

Dmitrij Agroskin, Eva Jonas, Johannes Klackl and Mike Prentice


# Editorial: Social Cognition, Motivation, and Interaction: How Do People Respond to Threats in Social Interactions?

Eva Jonas\* and Christina Mühlberger

Division of Social Psychology, Department of Psychology, University of Salzburg, Salzburg, Austria

Keywords: threat, discrepancy, motivational-affective state, motivated cognition, motivated behavior, Loop2Loop model

**Editorial on the Research Topic**

### **Social Cognition, Motivation, and Interaction: How Do People Respond to Threats in Social Interactions?**

When people are in social interactions they often have to cope with different kinds of threats. Threats can result from interaction partners' concrete behaviors, like if they threaten our freedom to act or degrade our behavior. Further, threats can also be abstract or existential in nature, like when we sense a loss of control or meaning in a puzzling situation. These various forms of threats play important roles in our daily interactions and often change how we think and behave around others. In this special issue, we collected research approaching the topic of threat in social interactions from different perspectives. Now we step back and reflect on how these perspectives might be integrated into one broader framework.

#### Edited and reviewed by:

Snehlata Jaswal, L. M. Thapar School of Management, India

> \*Correspondence: Eva Jonas eva.jonas@sbg.ac.at

#### Specialty section:

This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology

Received: 03 August 2017 Accepted: 29 August 2017 Published: 27 September 2017

#### Citation:

Jonas E and Mühlberger C (2017) Editorial: Social Cognition, Motivation, and Interaction: How Do People Respond to Threats in Social Interactions? Front. Psychol. 8:1577. doi: 10.3389/fpsyg.2017.01577 SOURCES OF THREAT

Threat in social interactions can result from different sources: It can result from the behavior of our interaction partners and as such strongly influence our emotions, cognitions, and behavior, as when one feels ignored or excluded by the interaction partner (Wesselmann et al.), if the partner threatens one's freedom to act (Sittenthaler et al.; Steindl and Jonas; Niesta-Kayser et al.), uses unpleasant communication (Klonek et al.; Traut-Mattausch et al.), or appears to be deceptive or untrustworthy (Mackinger et al.).

However, threats can also result from situational aspects that are largely unrelated to the behavior of the interaction partner, like when the situation is socially uncertain (Gollwitzer et al.), puzzling (van den Bos et al.), involves vulnerability and the willingness to accept risks (Keller et al.) or poses a social-evaluative threat to the self (Frisch et al.). Moreover, one feels the pressure to improve one's performance (Scholl et al.) or one is in danger to lose a powerful position to the interaction partner (Scheepers et al.).

People can also enter an interaction situation by having previously been affected by a threat resulting from an unrelated event, in which they lost control (Stollberg et al.), had been reminded of their own mortality (Agroskin et al.), were exposed to goal conflicts (McGregor et al.), or feared to not fulfill one's duties and obligations (Keller et al.).

Finally, people can also feel threatened because of certain predispositions, like being in a prevention focus (Keller et al.; Scholl et al.), having high victim sensitivity (Gollwitzer et al.), or an anxious attachment style (Ein-Dor).

Despite the variation in circumstances, threatening situations always involve the experience of a discrepancy (Jonas et al., 2014), which means a discrepancy between the reality one is confronted with and what one would have otherwise wanted or expected. Broadly speaking one could sort these discrepancies as follows.

# THREATS AS THE EXPERIENCE OF DISCREPANCIES

# Discrepancy between Motive and Reality (Perception of the Actual Situation)

Employees want to fulfill their job duties well but they are confronted with customer complaints (Traut-Mattausch et al.). Bank customers search for advice and would like to trust their financial consultant (Mackinger et al.). People wish to stick to their habits but are confronted with a change request (Klonek et al.). They strive for power but experience that they lack stable power (Scheepers et al.). People generally want to reach their goals, but the actual situation carries with it the prospect of failure (Frisch et al.; Keller et al.; Scholl et al.).

# Discrepancy between Cognitive Focus and Reality (Perception of the Actual Situation)

People expect to possess a certain freedom but experience that this freedom is threatened (Steindl and Jonas; Niesta-Kayser et al.; Sittenthaler et al.). People expect to be included into a group but are excluded (Wesselmann et al.).

# Discrepancy between Motive and Cognitive Focus

Victim sensitive persons desire to trust but expect untrustworthinesss (Gollwitzer et al.) and insecurely attached people desire safety but expect threats (Ein-Dor). People strive for self-preservation, control, certainty, or meaning but have been reminded that they are mortal (Agroskin et al.), sometimes lack influence (Stollberg et al.), experience frustration and uncertainty (McGregor et al.), or experience confusion (van den Bos et al.).

In all these different forms, threats can be part of our daily interactions—and shape these interactions. When people are confronted with a threat it can have cascading consequences for their feelings, cognitions, and behaviors within a social interaction. In the different articles in this special issue, different sequences of central psychological variables (motivation, cognition, behavior depending on characteristics of the interaction situation, the behavior of the interaction partner, the motives and/or personalities of the acting person him-/herself) have been explored. We suggest connecting these different variables in a model of social interaction**,** which helps us to better understand the dynamic nature of social interactions (see Figure here).

# THREATS IN SOCIAL INTERACTIONS—THE LOOP2LOOP MODEL AS DYNAMIC PERSPECTIVE

According to Interdependence Theory (Kelley and Thibaut, 1978; Kelley et al., 2003), people transform the actual (objective) situation into an effective situation of idiosyncratic interpretations and meanings (see Figure 1 in Steindl and Jonas). To better understand this transformation process we suggest specifying people's motives, their cognitive focus in a specific situation, and their perception of the situation (see above and Figure here). The different papers in the special issue deal with a wide array of people's motives and needs, like their desire for control, certainty, meaning, belonging, freedom, and selfprotection. The cognitive focus might be activated by situational aspects (e.g., a focus on gains vs. losses, or a focus on certain norms) or by personality characteristics. This focus can interact with people's perception of the actual situation and/or their motives as well as the motives can interact with the situation. The Loop2Loop Model of social interactions (see Figure here; see also Steindl and Jonas and Jonas and Bierhoff, 2017) suggests that as a result, the actual situation is transformed into an effective situation, i.e., how people subjectively construe a situation, which induces a certain motivational-affective state that is accompanied by motivated cognitions and often transforms into behavior. In a social interaction this behavior is visible to the interaction partner. Whether interaction partners feel threatened or valued within the interaction has important implications for how the interaction will unfold further, and these processes are explored next.

# Loop of Motivational-Affective States, Motivated Cognition, and Motivated Behavior

The experience of discrepancies result in a motivationalaffective state, that can be related to the activation of two main motivational systems: the behavioral inhibition system (BIS, Gray and McNaughton, 2000), which is characterized by inhibition, anxious arousal, and attentional vigilance, and the (inversely related) behavioral approach system (BAS, Gray and McNaughton, 2000) characterized by an "impulse to go toward" (Harmon-Jones et al., 2013, p. 291). As described by McGregor et al., goal regulation can be described as interplay between the BIS and the BAS systems (see also Jonas et al., 2014). When the BAS is predominant people are in a process of unhindered goal-striving. They feel energized, free from anxious worries, concentrated and committed toward approaching their goal. When they are confronted with a discrepancy, BIS activity increases, ongoing goal-striving is inhibited, and people may feel greater anxiety. People want to leave this aversive state via motivational re-orientation. The activation of the BAS provides this orientation and helps to mute previous anxiety.

In support of this suggested process, Agroskin et al. demonstrates that participants high on trait need for closure reacted to mortality salience with increased BIS activity (measured i.a. by EEG), which in turn led to increased ethnocentrism and reluctance for cultural exploration. However, Scholl et al. and Scheepers et al. show that people can perceive a discrepancy as threat or challenge. When people think they lack sufficient personal resources (e.g., skills) to cope with the demands of the situation they experience threat because they evaluate their resources to be too low to overcome the discrepancy. In contrast, when people think they possess sufficient resources they experience challenge. This cognitive focus interacts with the features of the situation (e.g., stability of one's power) and together affects people's physiological responses and behavioral reactions to discrepancies. By making use of the Asch paradigm, van den Bos et al. demonstrate that situational circumstances can be so strong, for example when a certain experimental situation is very puzzling, that people can get stuck in the activation of the BIS. Similarly, Gollwitzer et al. argue that victim sensitive participants can get caught by anxiety, torn between anxiously showing attentional vigilance to threat related stimuli, and trying to avoid the threat. This can result in a motivational state of avoidance motivation. With regard to freedom threats, Niesta-Kayser et al. show that the activation of an avoidance goal can increase the experience of threat and in turn impair performance.

If people's goals and needs shape their thinking we call this motivated cognition. Motivated cognition can affect people's perception of the social situation, lead to biased attention to specific information, and guide decision-making (Hughes and Zaki, 2015). People can be engaged in motivated cognition in order to serve an avoidance goal. Such an avoidance orientation can lead to a defensive stance and help to explain, e.g., non-cooperative behavior. For example, victim-sensitive people who fear being exploited are more likely to interpret ambiguous cues as negatively foreboding in socially uncertain situations. This interpretive bias leads to mistrust which in turn increases uncooperative behavior (Gollwitzer et al.). In a similar vein, a prevention focus, characterized by feelings of worry, cautiousness, and wanting to prevent losses and failure (Keller et al.), is related to worry that others are dishonest or unreliable and to motivated behavior, such as transferring lower sums of money in a trust game. Similarly, Traut-Mattausch et al. found that when customer service representatives wanting to avoid talking to customers, they subsequently devalued the customer and his/her concerns/issues (motivated cognition). This devaluing by the representative increased degrading service reactions to the customer (motivated behavior).

As these aggressive responses could suggest, people might not only be motivated to further avoid the threatening stimulus but may also try to overcome the anxious state and activate the BAS and regain agency, which can be achieved by showing angry or aggressive responses directed toward the source of threat (Carver and Harmon-Jones, 2009) or by approaching rewarding stimuli. Jonas et al. (2014) proposed that if people think they are capable of changing the unpleasant situation (i.e., reduce the discrepancy to the desired end-state) they might attempt to resolve the threat through fairly direct efforts. Interestingly, in threat-related situations we often find an increase in both, anti- and pro-social responses. This is illustrated by Wesselmann et al. in the context of social exclusion. These authors suggest that in response to a threat a pro-social behavior is more likely if an inclusionary need (belonging and self-esteem) has been violated, and an anti-social response is more likely if a power/provocation need (control and meaningful existence) has been threatened. This threat-response mapping suggests that responses are aimed at restoration of a threatened need. Supporting this assumption, Stollberg et al. showed that loss of control threats increased identification with highly agentic groups, i.e., groups that served the threatened need, but not with low agentic groups. Furthermore, task-group identification mediated the increase in the feeling of collective control after a loss of personal control. Scholl et al. showed that depending on people's motivational-affective state, different interaction partners (high vs. low power groups) are preferred. For example, people often prefer interacting with those partners who fit one's regulatory focus, but threat leads people to prefer high power partners and thus may shift people's reliance away from dispositional preferences toward state deficits.

Yet, if a direct response to the threat is not available, people may respond in ways that do not resolve the threat but help to overcome the discrepancy in an indirect way. They may orient to a domain of pursuit that is separate from the source of discrepancy but that allows for renewed approach motivation and a restoration of a sense of agency. For example, in response to threat, people often increase their commitment to goals that seem unrelated to the threat. These goals can be categorized along the continua of personal to social and concrete to abstract in nature (see Jonas et al., 2014). Following existential threat people often affirm more abstract (personal or social) values, worldviews or group identifications—as illustrated by Agroskin et al. for increased ethnocentrism following mortality salience. Interestingly, McGregor et al. suggests that aggressive religious radicalization is often facilitated by concrete social defenses such as group aggression. Regardless of the domain, affirming the self, one's group and related goals can (at least partly) reduce the aversive feelings aroused by discrepancies.

To summarize, people's experience of threat depends on (a) their personality traits (e.g., some people are more sensitive toward experiencing discrepancies) and their motive strength, (b) situational circumstances and how they are perceived (e.g., situational uncertainties), and (c) what people focus on in a certain situation. People's reactions to threat seem to depend on similar variables, as McGregor et al. emphasize, like (a) their personality traits (e.g., some people are more prone toward experiencing anxiety and/or to show certain agentic responses, like for example aggression; furthermore people have different access to activate their internal resources to cope with a threat), (b) situational circumstances, and (c) what people perceive in a certain situation [affordances—what might be similar to people's cognitive focus (see Figure here)].

# Loop2Loop—Consequences for Social Interactions

People's responses to threat influence how the social interaction further develops. When an actor exhibits aggressive behavioral intentions and evaluations following a freedom threat, this is related to the arousal of approach motivation (Sittenthaler et al.), which will have downstream impacts on their cognition and behavior toward the interaction partner moving forward. However, the actor's behavior will also influence the partner's motivational-affective state and his/her motivated cognition and behavior (e.g., an aggressively phrased complaint leads to an increase in people's heart rate, induces closed-mindedness, and an ignorance of improvement suggestions included in the complaint, Traut-Mattausch et al.). People's responses to threat can thereby contribute to a conflict loop and downward spiral. As another example of this process, Frisch et al. review literature that suggests that social-evaluative threats trigger stress reactions, such as neurophysiological responses and negative self-related cognitions. These stress reactions can have dysfunctional effects on social memory retrieval and socially adaptive behavior. If chronic stress prevents people from developing or employing the ability to adjust their behavior to other's needs, then this provides a source of interaction dissatisfaction. This could in turn induce mistrust, devaluation, and uncooperative behavior from the partner, which in turn might increase the threat the actor feels. In this way one loop leads to the next and in the end a negative interaction overall results, which is characterized by long-term biased cognition and interpersonal conflict. Indeed Steindl and Jonas found mistrust to be a mediator for uncooperative behavior following threat.

An interaction loop can evolve in a positive way if trustful behavior such as honesty or reliability creates a trustworthy atmosphere (Mackinger et al.), which in turn arouses a feeling of good prospects for one's motives. Klonek et al. found that reflective listening by a change agent (using Motivational Interviewing) helped to prevent the perception of threat with regard to changing one's environmental behavior. The more the agent's behavior was perceived to be empathetic, the more they came to believe that changing their behaviors might be beneficial.

Finally, Ein-Dor outlines to think about threat reaction processes at the group level with regard to personality styles. Specifically, he suggests that people with different attachment styles react to threats in different ways, and that the distribution of different styles within the group may help the group work together effectively to overcome threats. Some people are more sensitive to threat-related cues and show higher accuracy in detecting threats. Others are better in understanding how to deal with threats, like employing self-protective actions. Others yet are better able to coordinate group actions and motivate

# REFERENCES


people to overcome the threat. Groups are effective if people who detect threats early alarm group members who are able to activate rapid responses. In this way social interaction between varied personality styles may help people to coordinate their activities and to discover ways to effectively behave together toward approaching common goals. The piece ultimately outlines some fascinating ways to consider how threats can be regulated effectively via social interaction for the benefit of each interaction partner.

# CONCLUSION

In this Research Topic, we presented research and theory connecting three levels of analysis (the social, the cognitive, and the motivational) and illustrated their functional interconnection in social interaction. The Loop2Loop model helps us to connect basic motivational processes, which manifest in neural processes related to behavioral inhibition vs. activation in a social situation, with social cognition, which shapes our attention and commitments toward belief systems, and people's behavior in social interactions, which can serve to motivate the reactions we receive from our interaction partners.

# AUTHOR CONTRIBUTIONS

EJ wrote the manuscript. CM contributed to and edited the manuscript. Both authors approved the final version of the manuscript for submission.

# ACKNOWLEDGMENTS

This research was financially supported by the Austrian Science Fund (project: "From anxiety to approach—Testing a unified model of threat and defense," P27457). In addition, CM was financially supported by the Doctoral College "Imaging the Mind" of the Austrian Science Fund (FWF-W1233). We are grateful to Mike Prentice for helpful comments on a previous version of the paper.

Experimental Social Psychology, Vol. 49, eds J. M. Olson and M. P. Zanna (San Diego, CA: Academic Press), 219–286.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Jonas and Mühlberger. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Complaints as starting point for vicious cycles in customer–employee-interactions

#### *Eva Traut-Mattausch1\*, Sara Wagner2, Olga Pollatos3 and Eva Jonas1*

*<sup>1</sup> Department of Psychology, University of Salzburg, Salzburg, Austria, <sup>2</sup> Department of Psychology, University of Munich, Munich, Germany, <sup>3</sup> Department of Psychology, University of Ulm, Ulm, Germany*

A ring-model of vicious cycles in customer–employee-interaction is proposed: service employees perceive complaints as a threat to their self-esteem resulting in defense responses such as an increased need for cognitive closure, a devaluation of the customer and their information and degrading service behavior. Confronted with such degrading service behavior, customers react defensively as well, by devaluing the employee for example with regard to his/her competence and by reducing repurchase and positive word-of-mouth (WOM). Three studies investigated each link in this ringmodel. In study 1, participants were confronted with an aggressive or neutral customer complaint. Results show that motivated closed-mindedness (one aspect of the need for cognitive closure) increases after an aggressive complaint leading to a devaluation of the customer and their information, and in turn to a degrading service reaction. In study 2, participants were confronted with a degrading or favorable service reaction. Results show that they devaluate the employees' competence after receiving a degrading service reaction and thus reduce their intention to repurchase. In study 3, we finally examined our predictions investigating real customer–employee-interactions: we analyzed data from an evaluation study in which mystery callers tested the service hotline of an airline. Results show that the employees' competence is devaluated after degrading behavior and thus reduces positive WOM.

Keywords: self-esteem threat, defense response, customer complaint, customer–employee-interaction, motivated cognition, motivated behavior

# Introduction

Several weeks ago, Betty went to the local office of a travel agency and booked holiday package for Paris at a favorable price. As a customer she was very satisfied with the service and therefore recommended the travel agency to friends and colleagues. Due to an urgent project at work, however, she now has to change her travel arrangements and leave a few days later than planned. Therefore, she called the hotline of the travel agency and had her tickets changed without any problems. Yesterday, however, she received an invoice charging an extra fee for changing the flights—something she had not been told by the employee of the hotline. In fact, she had even asked for a flexible ticket in the beginning in order to avoid extra charges in case of changes. Betty is surprised about the invoice and calls the hotline of the travel agency a second time. Having waited several minutes, she confronts the service employee, John, with her problem and complains about the extra charge. Furthermore, she says: "I do not want to pay it." John feels stressed because of the complaint. In fact, he is not willing to dispute about the problem

#### *Edited by:*

*Eddy J. Davelaar, Birkbeck, University of London, UK*

# *Reviewed by:*

*Annika Scholl, Knowledge Media Research Center, Germany Ann-Christin Posten, Harvard University, USA*

#### *\*Correspondence:*

*Eva Traut-Mattausch, Department of Psychology, University of Salzburg, Hellbrunnerstrasse 34, 5020 Salzburg, Austria eva.traut-mattausch@sbg.ac.at*

#### *Specialty section:*

*This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology*

*Received: 29 November 2014 Accepted: 11 September 2015 Published: 16 October 2015*

#### *Citation:*

*Traut-Mattausch E, Wagner S, Pollatos O and Jonas E (2015) Complaints as starting point for vicious cycles in customer–employee-interactions. Front. Psychol. 6:1454. doi: 10.3389/fpsyg.2015.01454* based on non-reliable information from an uninformed and incompetent customer like Betty and therefore replies: "Now don't you pretend you did not know that you could not change budget flights for free. Only because you do not want to pay for this, you cannot blame us for your mistake." Confronted with this degrading service reaction, Betty devaluates John's professional competence, decides to never book again with this travel agency and not to recommend their service anymore. She tells John: "I'm not responsible for this!" and complains again about her problem, the nasty business practices, and his degrading service reaction as well*...*

Research shows that this customer–employee-interaction might be quite realistic. For service employees, like John, the customers' behavior is quite challenging from time to time, especially in case of complaints (Richins, 1983). On the other side customers, like Betty, often remain dissatisfied after they have complained due to inadequate responsiveness (Naylor, 2003). For the hotel sector, Lewis and McCann (2004) report that almost 50% of all customers who had complained are dissatisfied with the service recovery. In their research about complaint management, Holloway and Beatty (2003) found that 58% of all complainants remain dissatisfied with the recovery, a majority of which blame it on poor customer service and poor interaction during the complaint encounter. So what exactly happens during the interaction when a service employee, like John, is confronted with a complaint and leaves the customer, like Betty, dissatisfied with the complaint management?

To answer this question it is important to focus on the *customer–employee-interaction*. Like all dyadic interactions, service encounters are a dynamic interactive process between customer and service employee (Rafaeli et al., 2008). Both interaction partners influence one another—the customer influences the service employee whose reaction in turn influences the customer and so forth. Based on the Loop2Loop model of social interactions (Jonas, 2015; Jonas and Steindl, 2015; Jonas and Bierhoff, in press), we investigated a ring-model of vicious cycles in customer–employee-interaction within the current research. The Loop2Loop model subdivides the interaction process between two persons into single steps. During the interaction between Person A and Person B, A's behavior affects B's emerging motivational-affective state which then leads to B's motivated cognition and motivated behavior toward A. Accordingly, B's behavior affects A's emerging motivational-affective state and thus the following motivated cognition and motivated behavior toward B and so forth. The relationship-loop can develop in a positive way. However, it can also result in a negative dynamic as soon as one's motive is threatened during the interaction process. Accordingly, a negative motivational-affective state arises, followed by negative cognitions and resulting in a negative motivated behavior which in turn possibly threatens the motive of the other interacting person. We transferred the steps described within the Loop2Loop model—motivation, motivated cognition, motivated behavior—to the ring-model of vicious cycles in customer–employee-interaction. The ring-model is illustrated in **Figure 1**.

#### The ring-model of Customer Complaints

Our proposed ring-model starts with the customer behavior when the complaint is brought forward by the customer (see step 1 illustrated in **Figure 1**). According to Kowalski (1996), a complaint is a behavioral expression of dissatisfaction and could be based on a disappointment with a product, service, or perceived unfair seller practices (Fornell and Westbrook, 1979). Complaining customers provide feedback that their expectations were not met which could be valuable information to improve products or service quality (Kowalski, 1996; Traut-Mattausch and Jonas, 2007). An example for this could be Betty, our customer, who complains about the extra charge. This is a negative feedback for her interaction partner John, the service employee, because Betty's complaint indicates that her expectations were not met during the service process. John could use this indication to improve the service process, if he reacted in an open-minded way: by asking for details in order to understand underlying reasons and by appreciating Betty's information and suggestions. In our short story, however, John reacts quite differently, in a closed-minded way. He questions Betty's competence, devalues her information, and shows a degrading service reaction. Why is this?

In order to predict John's reaction we have to look at the threatening effects of complaints. As already mentioned a complaint can be seen as *negative feedback* expressing a disappointment (Kowalski, 1996) and therefore different from other customer requests such as asking for product information and offers or negotiating about prices and interests (see dimension *request* illustrated in **Figure 2**). Only negative customer feedback challenges the positive view about oneself (e.g., complaining about incomplete information or a wrong recommendation) or one's group (e.g., complaining about a faulty product, about a delayed delivery process, or bad business practices) and therefore threatens the self-esteem of a service employee which in turn leads to defense responses (Traut-Mattausch and Jonas, 2007). Empirical evidence investigating the effect of positive compared to negative feedback supports this idea. Feedback is not only uncomfortable for the receiver but even a threat toward the receiver's own self-worth (Leung et al., 2001). Therefore, defensive reactions toward (negative) feedback can be shown: it is, for example, less accepted and perceived as less accurate than positive feedback (Snyder and Newburg, 1981; Fedor et al., 1989)—even if the feedback sender tries consciously not to offend the receiver (Argyris, 1985, 1991). Negative feedback can be seen as self-esteem threatening information about one's own perceived inadequacy. A devaluation of the feedback and/or its source is a useful defense for one's own self-worth (Leung et al., 2001). We therefore conclude that a complaint is a threat toward the self-esteem of an employee even if it is neutrally phrased.

Additionally, complaints (as well as other requests, "noncomplaints") can be phrased either neutral or aggressive by the customer (see dimension *phrasing* illustrated in **Figure 2**). For example Betty's complaint referring not only on one case of bad customer service but also to "nasty business practices in general." Let us now imagine that the situation escalates and Betty shouts furiously: "I won't even consider

of bonus miles and therefore expect a fair and favorable price. Otherwise I want to talk to your boss!").

paying this extra fee! I want to speak to your boss right now!" As research shows this is a quite realistic scenario and makes customer service challenging for employees like John, especially in case of complaints. Richins (1983) distinguished four types of customer interaction styles, which occur in case of complaints—non-assertive, assertive, resort-to-aggression, and aggressive. Unfortunately, from John's perspective, only nonassertive customers refrain from complaining, whereas all other types of customers complain rather frequently (Richins, 1983). Thus, complaint management contains a good deal of unpleasant interactions with rather aggressive customers. Grandey et al. (2004) even report that call center employees, like John, are verbally attacked by customers 10 times on an average working day. Not to be treated politely, and with dignity and respect by others (*interpersonal justice*, Greenberg, 1993) is harmful, because mistreatments—such as using abusive words and actions—are signals for disrespect (Bies, 2001). Furthermore, Ferris et al. (2012) pointed out that such interpersonal injustice conveys threatening self-relevant information and thus could cause a loss in self-esteem. Examples for such threatening self-relevant information are to be inferior, to occupy a low-status or lowvalue position (Lind and Tyler, 1988; Lind, 2001). Interpersonal injustice—such as accusation or insults directed against the employee—can act as a threat to self-esteem (e.g., Baumeister et al., 1996; Zdaniuk and Bobocel, 2012). Complaining customers tend to imply aggressive interaction styles (Richins, 1983) such as verbally attacking service employees (Grandey et al., 2004). Based on this interpersonal unjust treatment, we conclude that an aggressively phrased complaint should be even more threatening than a neutrally phrased complaint.

# Negative Feedback as Threatening Aspect of Complaints: Pilot Study 1

According to practice reports employees feel most uncomfortable when dealing with complaint situations and try to avoid them (e.g., Stauss and Seidel, 2007). This is no surprise under the assumption that a complaint should threaten the self-esteem of a service employee. We argue that complaints are threatening because the service employee is confronted with negative feedback. To test this assumption we conducted a *pilot study*. Participants (*N* = 28 students at the University of Munich) were asked to put themselves into the role of an employee (for order processing, public relations, and customer service) in an office furniture department in a big department store. All participants were introduced to the situation of receiving an e-mail from the customer Mr. Herbert and informed that the customer would make suggestions in order to improve the order processing. Therefore, all participants expected a customer feedback. However, in the *negative feedback condition* participants were additionally informed that the customer complains extensively about a problem with the order processing, whereas in the *neutral feedback condition* participants were informed that the customer purchases office furniture. According to the Kano-model a customer should be extremely dissatisfied as soon as "must-be" requirements are not fulfilled. Customers take these requirements for granted and do not ask for them beforehand (Matzler and Hinterhuber, 1998). As a consequence, complaints are expressed as negative feedback by customers. This negative feedback is one source to learn more about the "must-be" requirements of a product or service performance (Seder and Al Hazza, 2014). Within the Kano-model "mustbe" requirements are differentiated from "one-dimensional" requirements. Customer satisfaction is proportional to the level of fulfillment of these requirements. "One-dimensional" requirements are usually expressed by a customer beforehand (Matzler and Hinterhuber, 1998). For us this behavior would be expressed as neutral feedback and correspond to the type (A) neutrally phrased non-complaint (see **Figure 2**). In the negative feedback condition participants were informed that the customer complains extensively about a problem. Because participants did not receive the complaint they were not confronted with verbal attacks. Thus, for us the manipulation of negative feedback corresponds to the type (B) neutrally phrased complaint (see **Figure 2**). We therefore compared within the first pilot study type (A) neutrally phrased non-complaints (condition "neutral feedback") with type (B) neutrally phrased complaints (condition "negative feedback").

Participants were asked to indicate their actual feelings in order to measure psychological discomfort by using seven items from the psychological discomfort scale (Elliot and Devine, 1994): good (recoded), uncomfortable, friendly (recoded), anxious, bothered, optimistic (recoded), happy (recoded) (α = 0.84). All items were measured on a 7-point scale from 1 (*not at all*) to 7 (*very much*). Following Balcetis and Dunning (2007), who demonstrated that the manipulation of threat (walking across the courtyard wearing an embarrassing costume) increases psychological discomfort, we chose psychological discomfort as our indicator for threat response. Results revealed that participants felt more psychological discomfort when expecting negative customer feedback (*M* = 3.67, *SD* = 1.10) compared to a neutral customer feedback (*M* = 2.69, *SD* = 1.34), *<sup>F</sup>*(1,26) <sup>=</sup> 4.46, *<sup>p</sup>* <sup>=</sup> 0.044, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.15. This result demonstrates that negative feedback could be a threatening aspect of customer complaints.

# Aggressive Phrasing as Threatening Aspect of Complaints: Pilot Study 2

An aggressive phrasing should be another threatening aspect of customer complaints. Therefore, we conducted a *second pilot study* to test this proposition and investigated the effect of neutrally vs. aggressively phrased complaints on physical stress measured by heart rate (HR) and skin conductance level (SCL)1 . Therefore, we compare within the second pilot study type (C) aggressively phrased complaints with type (B) neutrally phrased complaints (see **Figure 2**). We furthermore chose HR and SCL as additional indicators for threat response, because Schmid and Schmid Mast (2013) demonstrated that the manipulation of threat (social evaluation situation) increases HR and Wood et al. (2014) demonstrated that the manipulation of threat (loud whitenoise) increases SCL. We hypothesized that the mean HR as well

<sup>1</sup>Psychosocial stress is widely known to induce various adaptive responses of physiologic systems with particular increasing activities in the hypothalamuspituitary-adrenal axis (HPA) as well as in the sympathetic-adrenal-medullary (SAM) system. Schommer et al. (2003) emphasize that rapid habituation of HPA responses after repeated exposure to stressful stimulation is a frequently reported characteristic of the HPA axis. The authors presented evidence suggesting that with repeated psychosocial stress, a dissociation of HPA and SAM response patterns can be observed, characterized by a different temporal profile of habituation of catecholamine responses as compared to the HPA axis (Schommer et al., 2003). Suzuki et al. (2003) investigated the interaction between coping skills and psychophysiological reactions on stress using cardiovascular responses and SCL level. They reported that effort-distress coping intensified both cardiovascular responses as well as SCL. Therefore, we measured HR and SCL to investigate the stress response of our participants.

as SCL should increase to a higher extent after an aggressively phrased complaint compared to a neutrally phrased complaint. Similar to the first pilot study, participants (*N* = 14 students at the University of Munich) were asked to put themselves into the role of an employee in an office furniture department. All participants were confronted with an aggressively phrased complaint as well as with a neutrally phrased complaint after a baseline measure of HR and SCL. In support of our assumption, results revealed that HR increased significantly more after an aggressively phrased complaint (*M* = 76.65 beats per minutes, *SD* = 9.13) compared to a neutrally phrased complaint (*M* = 74.31 beats per minutes, *SD* <sup>=</sup> 9.88), *<sup>F</sup>*(1,13) <sup>=</sup> 5.26, *<sup>p</sup>* <sup>=</sup> 0.039, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.29, and that SCL tended to increase more after an aggressively phrased complaint (*M* = 3.08 μS, *SD* = 1.14) compared to a neutrally phrased complaint (*M* = 2.69 μS, *SD* = 1.06), *F*(1,13) = 4.48, *p* = 0.054, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.26. These results indicate that an aggressive phrasing could be another threatening aspect of customer complaints.

#### Employees' Defense Responses

The theory of lay person epistemology predicts a high need for cognitive closure when the situation is difficult, exhausting, or uncomfortable (Kruglanski, 1989, 2004). This need is defined as an "individual's desire for a firm answer to a question" (Kruglanski, 2004, p. 6). A high need for cognitive closure evokes a closed-minded-attitude toward new information (Kruglanski, 2004). Consequently, social judgments were made on readily accessible person schemas (Pierro and Kruglanski, 2008). An employee's schema of a complaining customer could be "a short-tempered nagger who wants something for free" leading to biased judgment. Recent research demonstrated that the manipulation of threat increases aspects of the need for cognitive closure leading to defense responses. More precisely, Thórisdóttir and Jost (2011) showed that the manipulation of threat (Studies 1a and 1b: recall of a high amount of threatening life experience) motivated closed-mindedness (one aspect of the need for cognitive closure). Moreover, Agroskin and Jonas (2013) demonstrated that the manipulation of threat (Study 3: mortality salience and control deprivation) increases the need for structure (another aspect of the need for cognitive closure) followed by ingroup defense for people with low self-esteem. We therefore predict within our ring-model an increased *need for cognitive closure* as *motivation* after a self-esteem threatening customer complaint (see step 2 illustrated in **Figure 1**). However, as need for cognitive closure and avoidance motivation are related (Agroskin, unpublished doctoral dissertation) avoidance motivation could play a further role during the customer– employee-interaction. Avoidance motivation is defined "as the energization of behavior by, or the direction of behavior away from, negative stimuli (objects, events, possibilities)" (Elliot, 2006, p. 112). Both pilot studies demonstrated that a complaint situation could be seen as a negative stimulus resulting in an increased psychological discomfort or psychosocial stress response. Therefore, it could be assumed that complaints can elicit an avoidance motivation as an additional motivation. However, in our research we focused on the need for cognitive closure as motivation within the ring-model (see **Figure 1**).

Given that employees are motivated to maintain a positive view of themselves (Tesser, 2000), they should strive to restore their threatened self-esteem (Zdaniuk and Bobocel, 2012). For example by using biased judgments, such as the devaluation of the customer and his/her information. Consistent with this idea, several social psychological theories (e.g., motivated reasoning approach, Kunda, 1990; multiple motive approach of the heuristic systematic model, Chaiken et al., 1989; biased hypothesis testing model, Pyszczynski and Greenberg, 1987) have illustrated that people defend their self-esteem relevant positions and perceptions against attacks and criticism by processing information selectively. This process makes it possible to protect one's self-esteem and to sustain one's perspective of the world. Therefore, a self-esteem-beneficial hypothesis will be confirmed (e.g., the products of the own company are of high quality and do not need improvements) and the self-esteem-threatening hypothesis will be avoided (e.g., complaint and complaining behavior are legitimate). Accordingly, the customer as well as the customer's feedback should be evaluated in a biased way. Negative feedback should be devalued in order to defend the employee's threatened self-esteem. Therefore, our ring-model predicts that the employee's increased need for cognitive closure, as a desire for a firm solution with regard to the complaint situation, should be followed by a *devaluation of the customer and his/her information* as a result of the employee's *motivated cognition* (see step 3 illustrated in **Figure 1**).

For Davidow (2000, 2003) a favorable complaint handling by employees includes embracing attentiveness (i.e., listening carefully to the complainant) and credibility (i.e., explaining the problem). Based on our assumption that employees should devaluate the complaining customer and his/her information in order to boost their threatened self-esteem, we thus predict that the employee reacts in an unfavorable way, i.e., with degrading service behavior in a closed-minded manner (e.g., not listening to the complaining customer, not-accepting any inconvenience, making the customer responsible for the problem). This degrading service reaction could be a form of self-affirmation, a common response to self-esteem threats (Steele, 1988), because it symbolizes the employee's dominance and therefore affirms the employee's self-esteem to the extent of being superior to the victim (Baumeister et al., 1996)—in case of complaints being superior to the complaining customer. To sum it all up, we predict within our ring-model that the devaluation of the customer and his/her information should be followed by *increased degrading service reaction* toward the customer as a result of the employee's *motivated behavior* (see step 4 illustrated in **Figure 1**).

## Customers' Defense Responses

As already mentioned we consider the customer–employeeinteraction as a dynamic process of interaction. Complaining customers expect to be treated politely and with dignity and respect by employees (interpersonal justice, Greenberg, 1993). A favorable complaint handling (Davidow, 2000, 2003) would meet these expectations. A degrading service reaction, however, should challenge the customer's interpersonal justice motive. Furthermore, the complaining customer receives threatening self-relevant feedback (Lind and Tyler, 1988; Lind, 2001), for instance "you are not worth being listened to carefully, we do not accept your inconvenience and you are the problem not our service," and might perceive a loss in self-esteem (Ferris et al., 2012). We therefore conclude that degrading service behavior poses a threat toward the customer's self-esteem and leads to defense responses by the customer. Within our proposed ringmodel of vicious cycles in customer–employee-interaction (see **Figure 1**) we differentiate between a customer's cognitive and a behavioral response.

With regard to the *motivated cognition* we predict an increased *devaluation of employees' competence* after degrading service behavior (see step 5 illustrated in **Figure 1**). This prediction could be derived from the above mentioned motivation to maintain a positive view of oneself (Tesser, 2000) resulting in an attempt to restore the threatened self-esteem (Zdaniuk and Bobocel, 2012), for example by devaluating the employee's competence. This prediction is in line with "motivated cognition"-approaches (e.g., Pyszczynski and Greenberg, 1987; Kruglanski, 1989; Kunda, 1990) pointing out that people defend important self-esteem relevant perspectives against attacks and critiques, and distort information in respect of this defense, for example by devaluating the criticism itself or its source.

The last step of our ring-model predicts the devaluation of the employee's competence followed by *reduced positive word-of-mouth* (WOM) and a *reduced repurchase intention* as a result of the customer's *motivated behavior* (see step 6 illustrated in **Figure 1**). Positive WOM is defined as the likelihood of customers spreading favorable information about an organization, which includes recommending the organization and its products and services (Maxham and Netemeyer, 2003) whereas repurchase intention is defined as the intention to continue to do business with an organization (Blodgett et al., 1997). Positive WOM usually occurs after positive and satisfying experiences (Parasuraman et al., 1988; for an overview see de Matos and Rossi, 2008), when customers' expectations were met or even exceeded (Maxham and Netemeyer, 2002). These expectations are competent and helpful service behavior and the feeling that employees take them seriously (Zeithaml and Bitner, 2000; Gruber et al., 2006). In this line competence could be identified as one key driver for positive WOM in laboratory experiments (Johnson and Zinkhan, 1991; Johnson et al., 1998). For the financial service sector, Rajaobelina and Bergeron (2009) report expertise and customer orientation as antecedents for a good relation between employee and customer which in turn positively influence repurchase intentions and positive WOM. Furthermore, negative interpersonal experiences reduce the intention to spread positive WOM (for an overview see Richins, 1983; Gremler and Gwinner, 2000; Gremler et al., 2001). Therefore, we predict a decreased positive WOM and decreased repurchase intention triggered through the devaluation of employees' competence during the service interaction within our ring-model.

### Overview of Studies

We conducted two laboratory studies and a field study to test each link of our proposed ring-model of vicious cycles in customer–employee-interaction. At first, study 1 investigated the motivation (step 2), the motivated cognition (step 3), and the motivated behavior (step 4) provoked by an aggressively phrased complaint compared to a neutrally phrased complaint (step 1). Study 2 then investigated the motivated cognition (step 5) and the motivated behavior (step 6) provoked by degrading compared to favorable service behavior (step 4). Furthermore, we conducted a field study investigating real customer–employee-interactions (study 3) to show evidence for our proposed ring-model. Whereas we compared (C) aggressively phrased complaints to (B) neutrally phrased complaints in study 1, we compared (C) aggressively phrased complaints to (A) neutrally phrased non-complaints in study 3 (see **Figure 2**) 2 .

# Study 1

The aim of study 1 was to test the hypothesized steps 1–4 assumed within the described ring-model (see **Figure 1**). More precisely, we wanted to show that participants respond with an increased need for cognitive closure (motivation), with an increased devaluation of the customer and his/her information (motivated cognition), and with an increased degrading service reaction (motivated behavior) when confronted with an aggressively phrased complaint compared to a neutrally phrased complaint. Furthermore, we hypothesized that the effect on increased degrading service behavior intention should be mediated by the need for cognitive closure (mediator 1) and the devaluation of the customer and his/her information (mediator 2).

# Method

#### Participants and Design

Fifty-eight students (42 female and 16 male) with an average age of 23.33 years (*SD* = 3.28) of the University of Munich participated in this study. The design was a 2 (*complaint phrasing*: neutral vs. aggressive) × 2 (*service reaction*: favorable vs. degrading) factorial design with repeated measure on the second factor. Participants were randomly assigned to the experimental conditions aggressive vs. neutral phrasing. We balanced the order in which both service reactions—favorable and degrading—were presented3 .

<sup>2</sup>All laboratory studies (pilot studies 1–2, main studies 1–2, study reported within the discussion part of study 1) were conducted before the ethical board at the University of Salzburg was established in 2011. Nevertheless, treatments of the participants were in accordance with APA ethical standards. All laboratory studies were undertaken with informed consent of the participants. Within the field study 3 we analyzed data from a mystery calls-evaluation survey. This survey was conducted by a professional service provider for mystery calls on behalf of a German airline in order to evaluate the service hotline of the airline. This is a common procedure for research purposes in marketing sciences (e.g., Wilson, 1998, 2002; Finn, 2001; Van der Wiele et al., 2005). Anonymous customers, the mystery callers, pretend to be real customers in order to evaluate the service quality of a service hotline. When they start their position, employees consent to the evaluation of their service behavior in general. At the given moment, however, it was not evident for employees whether the call was "real" or a mystery call.

<sup>3</sup>We tested whether the order of presentation of the within-factor has an effect. Results revealed no effect, *F*s *<* 1.

#### Procedure and Material

Participants were asked to put themselves in the place of service employees of an airline. In this role, they read the statement of a customer, who called their service hotline and refused paying an extra charge for the change of a booking. For one half of the participants the customer complaint was phrased aggressively; for the other half the customer complaint was phrased in a neutral way.

(*neutrally* phrased complaint) I do not want to pay an extra charge for the change of my booking. I called beforehand and asked whether it is possible to fly 2 days later. This would have been the perfect time to tell me that I cannot change my bookings for free. All in all, the booking process was too fast, so I did not have time to read through your terms of business. But this is exactly why I called your hotline: to get detailed information. It is not fair to inform me about this extra charge now.

(*aggressively* phrased complaint) I won't even consider paying an extra charge for the change of my booking! I called beforehand and asked whether it is possible to fly 2 days later. You should have told me that I cannot change my bookings for free. But you were so keen to come to an end that I did not have enough time to read through your terms of business. If I take extra time to call your hotline I expect excellent advice, but you are obviously not capable. To inform me about this extra charge now is monkey business!

Then, participants were asked the following two items as a manipulation check: (a) how much do you feel the customer offends you and (b) how polite is the customer (recoded). Next, participants were asked a single item in order to measure the aspect "closed-mindedness" of the need for cognitive closure as mediator one—how much would you like to talk to the customer (recoded)4 —and the following eight items in order to measure the devaluation of the customer and his/her information as mediator two: (a) how competent is the customer (recoded), (b) how reliable is the customer (recoded), (c) how intelligent is the customer (recoded), (d) how appropriate is the customer's behavior (recoded), (e) how credible is the customer (recoded), (f) how informative is the customer's call (recoded), (g) how seriously do you take the information (recoded), and (h) how reliable is the information (recoded).

Furthermore, participants were presented a favorable and a degrading service reaction.

(*favorable* service reaction) Unfortunately, this was not going very well. You are right, we should pay more attention to explain our terms of business before our customers make their bookings. It has to be transparent for our customers and give them enough time to reconsider their decision.

(*degrading* service reaction) Now do not you pretend you did not know that you could not change budget flights for free. Only because you do not want to pay for this, you cannot blame us for your mistake.

After each service reaction participants were asked how much they would like to react in this way. All items were measured on a 6-point scale from 1 (*not at all*) to 6 (*very much*).

# Results

#### Manipulation Check

The intercorrelation of the two manipulation check items was high enough to allow both items to be compiled to an overall measure (*r* = 0.51, *p <* 0.001). The mean for this measure was significantly higher in the aggressive phrasing condition (*M* = 4.54, *SD* = 0.94) compared to the neutral phrasing condition (*M* = 3.10, *SD* = 0.98), *F*(1,56) = 32.36, *p <* 0.001, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.37. This result indicates that the manipulation of the complaint phrasing was successful.

#### Closed-mindedness

To test whether closed-mindedness was higher after an aggressively phrased complaint compared to a neutrally phrased complaint we ran an ANOVA. Results revealed more closed mindedness after an aggressively phrased complaint (*M* = 4.86, *SD* = 1.24) compared to a neutrally phrased complaint (*M* = 3.83, *SD* <sup>=</sup> 1.51), *<sup>F</sup>*(1,56) <sup>=</sup> 7.90, *<sup>p</sup>* <sup>=</sup> 0.007, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.12.

#### Devaluation of the Customer and his/her Information

We first created a measure by aggregating the eight items (α = 0.87). We then analyzed the influence of complaint phrasing on the devaluation of the customer and his/her information. Results revealed that the devaluation was higher after an aggressively phrased complaint (*M* = 3.86, *SD* = 0.69) compared to a neutrally phrased complaint (*M* = 3.30, *SD* = 1.00), *<sup>F</sup>*(1,56) <sup>=</sup> 6.12, *<sup>p</sup>* <sup>=</sup> 0.016, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.10.

#### Service Reaction

Next, we ran a 2 (*complaint phrasing*: neutral vs. aggressive) × 2 (*service reaction*: favorable vs. degrading) ANOVA with repeated measures on the last factor. The ANOVA showed a significant main effect for the "service reaction" factor, *F*(1,55) = 16.26, *<sup>p</sup> <sup>&</sup>lt;* 0.001, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.23 (favorable: *<sup>M</sup>* <sup>=</sup> 4.02, *SD* <sup>=</sup> 1.42; degrading: *M* = 2.67, *SD* = 1.71). However, and even more importantly, we found an interaction between "complaint phrasing" and "service reaction," *F*(1,55) = 5.84, *p* = 0.019, η<sup>2</sup> = 0.10. Simple effects analyses showed that confronted with an aggressively phrased complaint participants preferred to show degrading service reaction (*M* = 3.36, *SD* = 1.73) compared to a neutrally phrased complaint (*M* = 1.97, *SD* = 1.43), *p* = 0.001. However, regarding the preference of the favorable service reaction, simple effects analyses showed no difference between an aggressively phrased complaint (*M* = 3.89, *SD* = 1.31) and a neutrally phrased complaint (*M* = 4.14, *SD* = 1.53), *p* = 0.520. This result indicates that after an aggressively phrased complaint participants preferred to show a degrading service reaction to

<sup>4</sup>Research conducted by Thórisdóttir and Jost (2011)showed that the manipulation of threat motivated the aspect "closed-mindedness" of the need for cognitive closure. We therefore decided to measure this aspect as well. However, the need for cognitive closure is defined as an "individual's desire for a firm answer to a question" (Kruglanski, 2004, p. 6). Transferred to a complaint situation a high need for cognitive closure would imply the desire to come to the self-esteem beneficial conclusion that the customer is responsible for the problem which represents the firm answer to the question "who is responsible for the complaint." Consequently, the service employee should not be motivated to dispute about the complaint (Traut-Mattausch and Jonas, 2007). Thus, we think that the item "how much would you like to talk to the customer (recoded)" reflects closed-mindedness as one aspect of the need for cognitive closure adapted for a customer–employee-interaction.

a greater extent compared to a neutrally phrased complaint, whereas the favorable reaction did not differ depending on the complaint phrasing. Looking at it from the other way, simple effects analyses showed, that after a neutrally phrased complaint, participants preferred to show a favorable service reaction over a degrading service reaction, *p <* 0.001, whereas after an aggressively phrased complaint both service reactions were preferred equally, *p* = 0.275.

In addition, we created a difference measure by subtracting the preference for the degrading service reaction from the preference for the favorable service reaction as dependent variable.

> Difference measure *service reaction* = preference to show favorable service reaction − preference to show degrading service reaction*.*

Then, we investigated the effect of complaint phrasing on the difference measure service reaction. Results revealed a greater difference after the neutrally phrased complaint (*M* = +2.14, *SD* = 2.46) compared to the aggressively phrased complaint (*<sup>M</sup>* = +0.54, *SD* <sup>=</sup> 2.54), *<sup>F</sup>*(1,55) <sup>=</sup> 5.84, *<sup>p</sup>* <sup>=</sup> 0.019, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.10.

#### Mediation Analysis

Moreover, we analyzed whether the effect of complaint phrasing (independent variable) on the difference measure service reaction (dependent variable) could be explained through closedmindedness (mediator one) and devaluation of the customer and his/her information (mediator two) using PROCESS (model 6, Hayes, 2013, p. 446). The results are displayed in **Figure 3**.

The analyses revealed a significant influence of complaint phrasing on the difference measure service reaction, *B* = −1.60, *SE* = 0.66, *p* = 0.019, and the mediator one, *B* = 1.03, *SE* = 0.37, *p* = 0.007. Subsequent analyses of the mediator one on the mediator two, *B* = 0.21, *SE* = 0.08, *p* = 0.012, as well as of the mediator two on the difference measure service reaction, *B* = −1.50, *SE* = 0.37, *p* = 0.001, showed significant regression weights indicating considerable influence of mediator one on mediator two as well as of mediator two on the difference measure service reaction. When finally examining the influence of the complaint phrasing of mediator one and mediator two on the difference measure service reaction concurrently, the effect of the complaint phrasing was considerably reduced, *B* = −1.04, SE = 0.64, *p* = 0.109. The indirect effect of the complaint phrasing on the difference measure service reaction through mediator one and two was highly significant as indicated by the 95% CI (−0.92, −0.04) using 1,000 bootstrap estimations.

#### Discussion

Results of study 1 support all hypothesized steps 1–4 assumed within our ring-model. An aggressively phrased complaint increased the closed-mindedness motivation compared to a neutrally phrased complaint (motivation). Consequently, the customer and the customer's information were devalued in case of an aggressively phrased complaint compared to a neutrally phrased complaint (motivated cognition). Concerning the preferred service reactions (motivated behavior), our expectations were also confirmed. A neutrally phrased complaint resulted in increased preferences for the favorable service reaction toward the customer and reduced preferences for the degrading service reaction. This difference, however, disappeared in case of an aggressively phrased complaint, when the preference for the degrading service reaction increased to an equal level with the favorable service reaction. Finally, the mediation analysis provides evidence for the assumed role of the need for cognitive closure and the devaluation of the customer and his/her information as mediators in predicting the service reaction.

However, study 1 is limited by the fact that closed-mindedness was the only aspect of the need for cognitive closure that was measured. Therefore, we conducted a further study measuring another aspect of the need for cognitive closure—the personal fear of invalidity (Thompson et al., 2001). Fear of invalidity (avoidance of early closure) can motivate people to be accurate by a desire to avoid the dullness associated with relying on whatever thoughts come to mind first (Moskowitz, 2005). In contrast, people with *no* fear of invalidity (achievement of early closure) should be motivated to evaluate information on readily accessible person schemas. An employee's schema of an aggressively complaining customer could be "a short-tempered nagger who wants something for free" leading to a negative evaluation of the customer and his/her information. To test the effect of fear of invalidity on the devaluation of customers' information we asked participants (*N* = 75 students of the University of Munich) to put themselves into the role of an employee in the student counseling service of the University of Munich5 . All participants were informed that students can evaluate the service of the counseling department through an online form. Next, all participants were asked to read several web evaluations. Each evaluation was a complaint phrased either neutrally (two evaluations) or aggressively (two evaluations) and included suggestions for the improvements of the service. In order to measure the devaluation of the suggestions we asked all participants to evaluate the suggestions by using the following three items: (a) the student provides suggestions in order to optimize the service of the counseling department (recoded), (b) the suggestions of the student are helpful (recoded), and (c) the suggestions of the student are important (recoded) (α = from 0.73 to 0.84). All items were measured on a 5-point scale from 1 (*not at all applicable*) to 5 (*very much applicable*). In the absence of a well-established *state* measure we decided to use the *trait* measure "personal fear of invalidity" (German short version, Hänze, 2002). This trait measure consists of six items such as "I tend to continue to evaluate recently made decisions" (α = 0.86). We recoded the items to measure *no* fear of invalidity which represents an achievement of early closure. Within our ring-model we assume an effect of complaints on the devaluation of the customers' information mediated by the need for cognitive closure as a state variable. However, we measured (no) fear of invalidity as a trait. Therefore, we predicted that the effect of aggressively vs. neutrally phrased complaints on the devaluation of customers' suggestions is moderated by

<sup>5</sup>Moreover, we tried to manipulate the identification of the participants within this study. Due to this purpose further variables were measured. However, results of the manipulation check indicate that the manipulation was not successful.

the trait (no) fear of invalidity. To test this assumption we z-standardized the moderator (no fear of invalidity) and ran a repeated measure ANOVA with the within-factor devaluation (aggressively vs. neutrally) and no fear of invalidity as covariate. The ANOVA showed a significant main effect for the within factor, *F*(1,73) = 71.15, *p <* 0.001, η<sup>2</sup> = 0.49 (devaluation of the aggressively phrased suggestions: *M* = 2.21, *SD* = 0.79; devaluation of the neutrally phrased suggestions: *M* = 1.49, *SD* = 0.54). However, and more important, we found an interaction between "no fear of invalidity" and the withinfactor, *<sup>F</sup>*(1,73) <sup>=</sup> 4.49, *<sup>p</sup>* <sup>=</sup> 0.038, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.06. To illustrate the nature of this interaction we calculated the correlation between the moderator (no fear of invalidity) and the devaluation of aggressively phrased suggestions and between the moderator (no fear of invalidity) and the devaluation of neutrally phrased suggestions, respectively. Results show a significant correlation between no fear of invalidity (achievement of early closure) and aggressively phrased complaints, *r* = 0.29, *p* = 0.013; whereas the correlation between no fear of invalidity (achievement of early closure) and neutrally phrased complaints was ns, *r* = 0.08, *p* = 0.481. Taken in conjunction, results of study 1 and the recently reported study show that different aspects of the need for cognitive closure (open-mindedness, fear of invalidity) are relevant for predicting a devaluation of the customer and his/her information depending on how this customer phrases the complaint.

The results of study 1 show that after an aggressively phrased complaint both the degrading and the favorable service reaction are equally preferred. However, it is unclear which reaction a service employee would choose in reality. As already mentioned, between approximately 50 to 60% of all customers who had complained are dissatisfied with the service recovery, a majority of which blame it on poor customer service and poor interaction during the complaint encounter (Holloway and Beatty, 2003; Lewis and McCann, 2004). Based on these facts, one can speculate that one half of service employees show degrading behavior in reality whereas the other half would not and that this difference could depend on the applied organizational culture, pre-defined display rules, and control possibilities of service

quality (e.g., presence of others, recoding of phone conversation). However, future research may imply an either favorable or degrading behavior response measure to answer the question which behavioral option is actually chosen by employees.

All in all, study 1 revealed evidence for the assumed steps 1–4 within our ring-model. Now we shift our focus in the interaction from the employee to the customer. Study 2 therefore examined how complaining customers react when confronted with degrading service reaction by employees.

# Study 2

The aim of study 2 was to test all hypothesized steps 4–6 assumed within the described ring-model (see **Figure 1**). More precisely, we wanted to show that participants respond with an increased devaluation of employees' competence (motivated cognition) and with a reduced repurchase intention (motivated behavior) when confronted with a degrading service reaction compared to a favorable service reaction. Furthermore, we hypothesized that reduced repurchase intention in case of degrading service reaction should be mediated by the devaluation of employees' competence.

# Method

#### Participants and Design

Fifty-six consumers (39 female and 17 male) with an average age of 22.22 years (*SD* = 3.59) participated in this study based on a one factorial between-subject design with two experimental conditions (*service reaction*: favorable vs. degrading). Participants were recruited at the University of Munich and randomly assigned to one of the two experimental conditions.

#### Procedure and Material

Participants were asked to put themselves in the role of customers of an airline. In this role, they called the service hotline and refused to pay an extra charge for the change of their booking. After reading their own complaint, half of the participants were confronted with a favorable service reaction, and half of the participants were confronted with a degrading service reaction (the same service reactions as depended variable used in study 1).

(*scenario*) Please imagine you booked a flight online after having consulted a travel agency. You would now like to change your booking and are being informed that you will be charged an extra fee. However, you had not been informed about this extra fee before booking. You call the service hotline of your travel agency. The service employee responds:

(*favorable* service reaction) Unfortunately, this did not go very well. You are right we should pay more attention to explaining our terms of business before our customers make their bookings. This has to be transparent for our customers and they should be given enough time to reconsider their decision.

(*degrading* service reaction) Now do not you pretend you did not know that you could not change budget flights for free. Only because you do not want to pay for this, you cannot blame us for your mistake.

All participants were then asked the following two items as a manipulation check: (a) how open-minded do you consider this reaction and (b) how repulsive do you consider this reaction. Next, participants were asked the following two items in order to measure their devaluation of employees' competence as a mediator: (a) the agent is competent (recoded) and (b) the agent acts in a professional manner (recoded). Furthermore, participants were asked the following three items in order to measure their repurchase intention as dependent variable: (a) would you recommend the service hotline, (b) would you like to call the hotline again, and (c) would you like to book another flight with the agency. All items were measured on a 6-point scale from 1 (*not at all*) to 6 (*very much*).

# Results

#### Manipulation Check

First we checked the manipulation of the "service reaction" factor. Therefore, we ran a 2 (*service reaction*: favorable vs. degrading) × 2 (*evaluation of the service reaction as*: openminded vs. repulsive) ANOVA with repeated measures on the last factor. Results revealed a significant interaction between "service reaction" and "evaluation," *F*(1,54) = 62.36, *p <* 0.001, η<sup>2</sup> = 0.54. Simple-effect analyses showed that participants regarded the favorable service reaction as more open-minded (*M* = 3.74, *SD* = 1.38) compared to the degrading service reaction (*M* = 1.66, *SD* = 1.01), *p <* 0.001. Furthermore, simpleeffect analyses revealed that participants evaluated the degrading service reaction as more repulsive (*M* = 5.66, *SD* = 0.86) compared to the favorable service reaction (*M* = 3.56, *SD* = 1.48), *p <* 0.001. These results indicated that the manipulation of the "service reaction" factor was successful.

#### Devaluation of Employees' Competence

We first created a measure by aggregating the two items (*r* = 0.59, *p >* 0.001). Next, we analyzed how the service reaction influenced the devaluation of employees' competence. Results revealed that the employees' competence was more devaluated after the degrading service reaction (*M* = 5.48, *SD* = 0.65) compared to the favorable service reaction (*M* = 3.96, *SD* = 1.32), *<sup>F</sup>*(1,54) <sup>=</sup> 30.49, *<sup>p</sup> <sup>&</sup>lt;* 0.001, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.36.

#### Repurchase Intention

We created a measure by aggregating the three items (α = 0.72) and analyzed how the service reaction influenced the repurchase intention. Results revealed that the repurchase intention was lower after the degrading service reaction (M = 1.54, SD = 0.77) then after the favorable service reaction (M = 2.31, SD = 1.04), F(1,54) <sup>=</sup> 9.96, p <sup>=</sup> 0.003, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.16.

#### Mediation Analysis

Finally, we analyzed whether the effect of service reaction (independent variable) on repurchase intention (dependent variable) could be explained through the devaluation of employees' competence (mediator) using PROCESS (model 4, Hayes, 2013, p. 445). The results are displayed in **Figure 4**. The analyses revealed a significant influence of the service reaction on the repurchase intention, *B* = −0.77, *SE* = 0.24, *p* = 0.003, and the mediator, *B* = 1.52, *SE* = 0.28, *p >* 0.001. Subsequent analyses of the influence of the mediator on the repurchase intention showed a significant regression weight, *B* = −0.37, *SE* = 0.11, *p* = 0.002, indicating considerable influence of the mediator on the repurchase intention. When finally examining the influence of the service reaction and the mediator on the repurchase intention concurrently, the effect of the service reaction was considerably reduced, *B* = −0.21, *SE* = 0.28, *p* = 0.451. The indirect effect of the service reaction on the repurchase intention through the mediator was highly significant as indicated by the 95% CI [−1.09, −0.14] using 1,000 bootstrap estimations.

#### Discussion

The two hypothesized defense responses of the customer—steps 5 and 6 within the ring-model (see **Figure 1**)—were confirmed. A degrading service reaction increased the devaluation of employees' competence (motivated cognition) and lowered the repurchase intention (motivated behavior) compared to a favorable service reaction. Furthermore, the mediation analysis provides evidence for the assumed role of the devaluation of employees' competence as a mediator predicting the repurchase intention.

All in all, results of the first two studies revealed evidence for the assumed steps within our ring-model: Study 1 demonstrated the assumed defense responses by the employee and study 2 demonstrated the assumed defense responses by the customer. However, both studies were conducted in the lab using scenarios where participants were asked to put themselves in the role of an employee or in the role of a customer. Even though using scenarios is a usual procedure in psychological research, our studies do not demonstrate predictions derived from the proposed ring-model investigating real customer–employeeinteractions. We addressed this issue in study 3.

# Study 3

For this purpose, we analyzed data from a mystery callsevaluation study. These mystery calls were conducted in order to evaluate the service hotline of a German airline. This is a common procedure for research purposes in marketing sciences

(e.g., Wilson, 1998, 2002; Finn, 2001; Van der Wiele et al., 2005). Anonymous customers, the mystery callers, pretend to be real customers in order to evaluate the service quality of a service hotline. In this evaluation study, customers i.e., the mystery callers—brought forward a complaint or a non-complaint (neutral request) and then evaluated employees' competence. Furthermore, they also indicated whether they would engage in positive WOM or not6 . The nature of complaints was specified by the German airline responsible for the service quality of the service hotline (e.g., the golf equipment did not turn up during a golf holiday, changes in flight times, inconsistent cost information provided online, wrong information provided online, lost seat reservations). In case of complaints, mystery callers were asked to act aggressively and in a provocative manner toward the service employee7 . Based on this setting we investigated whether the devaluation of service employees' competence (as motivated cognition, step 5) is higher in case of (aggressive) complaints compared to non-complaints leading to a lowered positive WOM (as motivated behavior, step 6). We hypothesized less positive WOM in case of (aggressively phrased) complaints compared to (neutrally phrased) non-complaints. This effect should be mediated by the devaluation of employees' competence.

More importantly, in case of complaints mystery callers should evaluate the degrading behavior of the service employee. Therefore, it is possible to investigate the hypothesized steps 4–6 within the described ring-model (see **Figure 1**) for this subsample. We expected to replicate the findings of study 2 and hypothesized that more degrading behavior should lead to less positive WOM and that this effect should be mediated by the devaluation of employees' competence.

#### Method

#### Sample and Design

All in all, 32 professional mystery callers conducted 160 mystery calls in order to evaluate the service hotline of a German airline. All mystery callers were blind with regard to the tested hypotheses. The design was a one factorial between-subject

design with two experimental conditions (*request*: non-complaint vs. complaint). Before each call, the mystery callers randomly chose their behavior in the upcoming service encounter on request—non-complaints or complaints.

#### Procedure and Material

The content of each request (personal details, reasons for the call, etc.) were set by default in order to guarantee for standardized conditions. Mystery callers received training on the requests and the evaluation of the service employee during the call. The requests included complaints (60 calls) and non-complaints (100 calls) such as questions concerning different departments (50 calls) and flight bookings (50 calls). Two further (noncomplaint) requests (specific questions concerning products and pricing in order to test selected competences of the agents) were too specific to compare it with the complaining request and therefore excluded from the data analysis. During the call, the mystery callers evaluated the competence of the service employee by answering the following three items in order to measure the devaluation of employees' competence as mediator: (a) the agent knows the product very well (recoded), (b) the agent provides reliable information (recoded), and (c) the agent works independently and addresses problems (recoded). In case of a complaint, the following three items were used in order to measure the perceived degrading behavior of the service employee: (a) the agent recognizes the dissent (recoded), (b) the agent expresses understanding (recoded), and (c) the agent offers a solution (recoded). All items were measured on a 6-point scale from 1 (*very good*) to 6 (*very bad*). Furthermore, the mystery callers indicated whether they would recommend the service hotline or not by answering "yes" (coded 1) or "no" (coded 0) in order to measure positive WOM.

#### Results

#### Devaluation of Employees' Competence

We created a measure by aggregating the three items (α = 0.72). Next, we analyzed the influence of the request factor on the devaluation of employees' competence. Results revealed that the employees' competence was more devaluated after a complaint (*M* = 1.92, *SD* = 1.34) then after a non-complaint (*M* = 1.43, *SD* <sup>=</sup> 0.67), *<sup>F</sup>*(1,158) <sup>=</sup> 9.69, *<sup>p</sup>* <sup>=</sup> 0.002, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.06.

<sup>6</sup>Due to the purpose of the evaluation study further variables were measured. 7It was not possible to record the calls. Therefore, it was not possible to check whether the complaints were more aggressive than the non-complaints.

#### Positive WOM

A2(*request:* non-complaint vs. complaint) × 2 (*recommendation:* no vs. yes) chi-square analysis was conducted. Results revealed that only 60% of the complaining customers would recommend the hotline compared to 78% of the noncomplaining customers, <sup>χ</sup>2(1, *<sup>N</sup>* <sup>=</sup> 160) <sup>=</sup> 5.93, *<sup>p</sup>* <sup>=</sup> 0.015, eta = 0.19.

#### Mediation Analysis

We then analyzed whether the effect of request (independent variable) on positive WOM (dependent variable) could be explained through the devaluation of employees' competence (mediator) using PROCESS (model 4, Hayes, 2013, p. 445). The results are displayed in **Figure 5**. The analysis revealed a significant influence of request on positive WOM, *B* = −0.86, *SE* = 0.36, *p* = 0.016, and on the mediator, *B* = 0.49, *SE* = 0.16, *p* = 0.002. Subsequent analyses of the influence of the mediator on positive WOM showed a significant regression weight, *B* = −0.81, *SE* = 0.23, *p <* 0.001, indicating considerable influence of the mediator on positive WOM. When finally examining the influence of the request and the mediator on positive WOM concurrently, the effect of the request was considerably reduced, *B* = −0.56, *SE* = 0.39, *p* = 0.150. The indirect effect of the request on positive WOM through the mediator "employees' competence" was highly significant as indicated by the 95% CI (−0.89, −0.11) using 1,000 bootstrap estimations.

Mediation Analysis (Sub-sample Complaints, *n* **=** 588) We created a scale by aggregating the three items measuring degrading behavior (α = 0.66). Degrading behavior was significantly related to positive WOM, *r* = −0.27, *p* = 0.043, indicating that the more degrading behavior participants perceived the less positive WOM they would spread.

Finally, we analyzed whether the effect of degrading behavior (independent variable) on positive WOM (dependent variable) could be explained through the devaluation of employees' competence (mediator) using PROCESS (model 4, Hayes, 2013, p. 445). The results are displayed in **Figure 6**. The analysis revealed a significant influence of the degrading behavior on positive WOM, *B* = −0.72, *SE* = 0.36, *p* = 0.046, and the mediator "devaluation of employees' competence," *B* = 0.69, *SE* = 0.20, *p <* 0.001. Subsequent analyses of the influence of the mediator on positive WOM showed a significant regression weight, *B* = −0.73, *SE* = 0.32, *p* = 0.022, indicating considerable influence of the mediator on positive WOM. When finally examining the influence of the degrading behavior and the mediator devaluation of employees' competence on positive WOM concurrently, the effect of the degrading behavior was considerably reduced, *B* = −0.31, *SE* = 0.38, *p* = 0.407. The indirect effect of the degrading behavior on positive WOM through the mediator was highly significant as indicated by the 95% CI (−1.72, −0.07) using 1,000 bootstrap estimations.

#### Discussion

Results of study 3 confirmed our expectations. The two hypothesized defense responses of the customer were confirmed. The devaluation of employees' competence (motivated cognition) was higher in case of (aggressive) complaints compared to non-complaints. In addition, following (aggressive) complaints, customers reduced their intention to engage in positive WOM, whereas following non-complaints, customers were more willing to engage in positive WOM. Finally, the mediation analysis provides evidence for the assumed role of the devaluation of employees' competence as a mediator in predicting positive WOM. More importantly, the results of study 2 were replicated investigating real customer–employeeinteraction. More degrading behavior leads to less positive WOM and this effect was mediated by the devaluation of employees' competence.

# General Discussion

Building on the assumption that an employee, when confronted with a self-esteem threat in form of a complaint, shows defense responses we proposed a ring-model of vicious circles in customer–employee-interaction, in which we differentiate between three defense responses by the employee: motivation, motivated cognition, and motivated behavior. At first, the service employee is confronted with a probable aggressive complaint that elicits a need for cognitive closure as a motivational response. This motivation then leads to a devaluation of the customer and his/her information as a cognitive response followed by the behavioral response of degrading service reaction. The customer confronted with degrading service reaction in turn devaluates the service employees' competence as a cognitive response followed by reduced repurchase intention and positive WOM as behavioral responses.

At first, the aim of study 1 was to test all hypothesized defense responses of the employee assumed within our ringmodel (see **Figure 1**). In this study, participants were asked to put themselves in the role of a service employee and were confronted with either an aggressively phrased complaint (step 1) or a neutrally phrased complaint. Results revealed, as expected, that closed-mindedness (one aspect of the need for cognitive closure) increased (motivation, step 2) after an aggressively phrased complaint leading to a devaluation of the customer and his/her information (motivated cognition, step 3) and in turn to more preferred degrading service reaction (motivated behavior, step 4). Furthermore, study 2 investigated all hypothesized defense responses of the customer assumed within the ring-model. For this purpose we asked our participants to put themselves in the role of a complaining customer and confronted them either with a degrading service reaction or a favorable service reaction. Results revealed, as expected, that a degrading service reaction (step 4) increased the devaluation of employees' competence (motivated cognition, step 5) leading to less repurchase intention (motivated behavior, step 6) compared to a favorable service reaction. All in all, results of the first two studies revealed evidence for the single links in our ring-model. However, both

<sup>8</sup>In two cases there were missing data.

FIGURE 5 | Devaluation of employees' competence mediated the effect of request (non-complaint vs. complaint) on positive WOM in study 3. The mediation analysis was conducted using PROCESS (model 4, Hayes, 2013, p. 445). Coding of the independent variable "request" was non-complaint = 0 and complaint = 1. The mystery callers indicated whether they would recommend the service hotline or not by answering "yes" (coded 1) or "no" (coded 0) in order to measure positive WOM. *a* is an unstandardized regression coefficient; *b*, *c*, and *c* are unstandardized logistic regression coefficients. ∗*p <* 0.05, ∗∗*p <* 0.01, ∗∗∗*p <* 0.001.

studies were conducted in the lab using scenarios for which participants were asked to put themselves in the role of an employee or in the role of a customer. To address this issue, data from a mystery call-evaluation study were analyzed (study 3). The mystery callers evaluated the service hotline of a German airline. They brought forward an aggressively phrased complaint or a non-complaint (e.g., flight booking), evaluated employees' competence, and indicated whether they would engage in positive WOM. As predicted, the results revealed more devaluation of employees' competence (motivated cognition) in case of complaints then in case of non-complaints. This devaluation then resulted in decreased engagement in positive WOM (motivated behavior). More importantly, the mystery callers also evaluated the degrading behavior of the service employees which made it possible to investigate whether the results of study 2 were replicable testing real customer–employee-interactions. Results demonstrated, as assumed, that more degrading behavior leads to less positive WOM and this effect was mediated by the devaluation of employees' competence.

#### Theoretical Implications

The presented results replicate several findings with regards to complaint management and related outcomes such as WOM behavior and repurchase intentions. Corresponding with prior reports we found that customer complaints indeed evoke negative evaluations of the customer and in turn lead to rather degrading service reactions (Traut-Mattausch and Jonas, 2007). Furthermore, we could confirm that customers are more willing to spread positive WOM and have higher repurchase intentions after they have received competent service (e.g., Johnson et al., 1998; Brady and Cronin, 2001).

Despite the amount of literature concerning antecedents for positive WOM up to this date the main focus of research has been either on employees' behavior, on customers' satisfaction, loyalty or trust, or on organizational conditions (for an overview see de Matos and Rossi, 2008). On the one hand, our studies therefore add to the existing knowledge in that they take into consideration and combine the perspectives of both customers and service employees. On the other hand, the presented studies explain existing results by providing new insights in the interaction between customer and service employee. Our ringmodel of the interactive process combines customers' and service employees' responses and thereby could show that and how both are interrelated.

One might argue that the interdependence of interaction partners is not new. Indeed, examples like the interpersonal circle show that behaviors of interaction partners depend on each other, e.g., critical behavior of one person encourages distrust of the other (e.g., Freedman et al., 1951; Leary, 1957; Strong and Hills, 1986; Wiggins, 2003). The prediction that complaints negatively affect service reactions could also be just another proof that non-cooperative behavior elicits low intention to cooperate in different contexts (Kelley and Stahelski, 1970; Van Lange and Visser, 1999). Or our results could be simply explained by the fact that unpleasant interaction partners evoke matching behavior, i.e., people lower their own level of pleasantness in return (Burgoon et al., 1995) and respond in a hostile and aggressive way (Menon and Dubé, 2000).

However, our analyses went one step further. We could show that the reasons for service employees' degrading service reactions following complaints are more than only adapting to customers' rude behavior. The reason is rather an underlying psychological process: Complaints are perceived as unpleasant and threatening events and evoke service employees' need for cognitive closure. Consequently, the feedback is processed in a closed-minded and biased manner resulting in a devaluation of the customer and his/her information and thus in degrading service reaction. The proposed ring-model of the interactive process of complaint management hence not only describes, but also explains how customers and service employees react upon each other and, eventually, both determine the outcomes of their encounter. Based on that knowledge it should be possible to create effective intervention strategies to improve the customer– employee-interaction.

#### Limitations and Future Research

In our research we investigated the ring-model assuming that during the interaction between a complaining customer and a service employee, customer's behavior (step 1) affects employee's emerging motivation (step 2) which then leads to employee's motivated cognition (step 3) and motivated behavior toward A (step 4). Accordingly, employee's behavior affects customer's emerging motivated cognition (step 5) and motivated behavior toward the employee (step 6). The motivation "need for cognitive closure" should be responsible for the motivated cognition and for the motivated behavior shown by the employee. However, no motivation is specified for the motivated cognition and the motivated behavior shown by the customer within the ringmodel. Results of study 2 demonstrated that degrading service behavior (step 4) elicit a devaluation of employees' competence (step 5) which could be a cognitive defense response. We therefore suggest that this effect could be explained through *defense motivation* (Chaiken et al., 1989) and thus assume that the effect of degrading service behavior on devaluation of employees' competence should be mediated by defense motivation. In further research this assumption should be tested and the ring-model based on empirical evidence—expanded.

We investigated the ring-model of vicious cycles in customer– employee-interaction by building on the Loop2Loop model, which is a dynamic model of social interaction (Jonas, 2015; Jonas and Steindl, 2015; Jonas and Bierhoff, in press). However, not all steps of the Loop2Loop model were transferred to the ringmodel such as the motivational-affective state (e.g., physiological arousal) or the specific cognitive focus (e.g., specific goal) of both interaction partners. It would be interesting in future research to also include these additional variables of the Loop2Loop model into the investigation of customer–employee-interactions as well as to further explore the dynamic nature of the resulting social interaction to receive a broader picture of the underlying process of vicious cycles of customer complaints.

As illustrated in **Figure 2** we distinguish four types of behavioral expressions (see **Figure 2**): (A) neutrally phrased noncomplaints, (B) neutrally phrased complaints, (C) aggressively phrased complaints, and (D) aggressively phrased noncomplaints. In our research we compared the crucial type (C) aggressively phrased complaints to the type (B) neutrally phrased complaints in study 1, as well as to the type (A) neutrally phrased non-complaints in study 3. In future research type (C) aggressively phrased complaints should be compared to type (D) aggressively phrased non-complaints. Based on our argument that a complaint can be seen as negative feedback and therefore threatens the self-esteem of a service employee we propose that service employees respond with an increased need for cognitive closure (motivation), with an increased devaluation of the customer and his/her information (motivated cognition), and with an increased degrading service behavior when confronted with an aggressively phrased complaint compared to an aggressively phrased non-complaint. Moreover, our assumptions would ideally be addressed in future research by comparing all types of behavioral expressions in one study.

A further limitation of our research is the fact that the data in all studies rely on self-report from the same source (except for the second pilot study). Therefore, we cannot rule out that parts of our effects could be driven by participants' consistent responses across the used self-report measures. In further research it would be important to address this limitation by using psychophysiology measures (threat specific peripheral neurophysiological responses, cf. Blascovich and Mendes, 2010) in combination with real behavior. For this purpose, the psychophysiology response from the service employee as well as from the customer should be measured during an interaction whereas the interaction itself should be videotaped and analyzed afterward by trained raters based on a coding system. This procedure would ensure in our view to investigate customer– employee interaction on genuine interaction date and provide the possibility to test the proposed underlying process—the threatening effect of customer complaints.

## Practical Implications

Positive WOM is spread to five other people whereas in case of dissatisfying events negative WOM warns nine other people, i.e., potential new customers (Knauer, 1992). This fact demonstrates how important it is to foster recommendations. Taking into account that people trust most in personal recommendations (Murray, 1991) and considering the enormous social communication networks online positive WOM even gains in relevance. As our results show, positive WOM decreases in case of complaints due to perceived incompetent service and employees' degrading behavior toward customers. Successful complaint management therefore needs to focus not only on

formal guidelines for service employees or skills training, but also on the fundamental mindset that leads to the defense service reaction. The perception of complaints as visible sign of errors that are to be avoided needs to be replaced by the belief in complaints as valuable chance to satisfy and retain a customer. Thus, trainings should support service employees in reducing the need for cognitive closure and degrading reactions in order to deliver high quality complaint management. In this context Moskowitz et al. (1999) proposed the possibility of inhibiting an automatic motivation in case of complaints by another automatically activated goal. Employees' need for cognitive closure might thus be inhibited by another equal

# References


motivation such as to be curious in order to collect customer ideas within complaints for improvement.

Failures and failing recoveries decrease satisfaction (e.g., Bitner et al., 1990) and are the key drivers for complainants' switching behavior (Keaveney, 1995). However, successful complaint management offers the chance to restore satisfaction: Successful recovery can even achieve higher satisfaction than prefailure satisfaction levels (for an overview see de Matos et al., 2007) and in the aftermath increase the complainants' willingness to talk about this positive experience, i.e., spread positive WOM (Swanson and Kelley, 2001).


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2015 Traut-Mattausch, Wagner, Pollatos and Jonas. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# When Advisors' True Intentions Are in Question. How Do Bank Customers Cope with Uncertainty in Financial Consultancies?

Barbara Mackinger\*, Eva Jonas and Christina Mühlberger

Department of Psychology, Social Psychology, University of Salzburg, Salzburg, Austria

When making financial decisions bank customers are confronted with two types of uncertainty: first, return on investments is uncertain and there is a risk of losing money. Second, customers cannot be certain about their financial advisor's true intentions. This might decrease customers' willingness to cooperate with advisors. However, the uncertainty management model and fairness heuristic theory predict that in uncertain situations customers are willing to cooperate with financial advisors when they perceive fairness. In the current study, we investigated how perceived fairness in the twofold uncertain situations increased people's intended future cooperation with an advisor. We asked customers of financial consultancies about their experienced uncertainty regarding both the investment decision and the advisor's intentions. Moreover, we asked them about their perceived fairness, as well as their intention to cooperate with the advisor in the future. A three-way moderation analysis showed that customers who faced high uncertainty regarding the investment decision and high uncertainty regarding the advisor's true intentions indicated the lowest intended cooperation with the advisor but high fairness increased their cooperation. Interestingly, when people were only uncertain about the advisor's intentions (but certain about the decision) they indicated less cooperation than when they were only uncertain about the decision (but certain about the advisor's intentions). A mediated moderation analysis revealed that this relationship was explained by customers' lower trust in their advisors.

#### Keywords: uncertainty, self-interest, fairness, trust, advice taking

# INTRODUCTION

Ninety-four-year-old Mrs X, whose cognitive and physical abilities are impaired by old age, received a sizeable amount of money from her disbursed life insurance. To reduce her uncertainty on how to reinvest the money, she consulted a financial advisor. The advisor recommended that the "optimal" solution was to put her money in a long-term investment—obviously not considering Mrs X's age (WirtschaftsWoche, 2009). Thus, one may doubt the advisor's trustworthiness. What was the advisor's motive and has s/he acted in the client's best interest or rather in her/his own interest? In 2008 the global financial crisis had swamped the Eurozone. During this time the crisis dramatically expanded in the European public sphere. It was not only centered in the attention of political leaders, public opinion, and media. Also private persons like Mrs X were involved

#### Edited by:

Eddy J. Davelaar, Birkbeck University of London, United Kingdom

#### Reviewed by:

Mario Gollwitzer, Philipps University of Marburg, Germany Snehlata Jaswal, L. M. Thapar School of Management, India

> \*Correspondence: Barbara Mackinger barbara.mackinger@sbg.ac.at

#### Specialty section:

This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology

Received: 12 July 2016 Accepted: 16 June 2017 Published: 30 June 2017

#### Citation:

Mackinger B, Jonas E and Mühlberger C (2017) When Advisors' True Intentions Are in Question. How Do Bank Customers Cope with Uncertainty in Financial Consultancies? Front. Psychol. 8:1112. doi: 10.3389/fpsyg.2017.01112

and harmed. Especially, those who had been working with a financial advisor were now facing a dilemma: on the one hand they wanted and needed advice, but on the other hand they were not sure if they could rely on their financial advisor. In recent years the media has reported many incidents where experts did not have the knowledge about the optimal investment or at any rate did not use their knowledge to best help their customers. As a result, the questions remain—how can customers regain confidence and how can banks and financial consultants win back their customers' trust in uncertain situations?

# Uncertainty

Uncertainty is experienced when people feel unable to predict future events or when they feel inconsistencies between important cognitions, experiences, or behaviors (Van den Bos and Lind, 2002). Uncertainty can be provoked by various situations. For example, thinking about insecure aspects of one's self or one's relationships to others, or losing control over one's own life can lead to aversive and threatening feelings (Van den Bos and Miedema, 2000; Van den Bos, 2001) which people strive to reduce (Van den Bos and Lind, 2002; Heine et al., 2006; Hogg, 2007). Such uncertainty is also relevant in investment decisions where people are dependent on the financial market. Even if they choose a low risk product such as a long-term fixed capital saving, they face the risk of a decreased interest rate. Therefore, investing money is a risky decision which entails uncontrollable consequences, making one's future unpredictable and thus, uncertain.

# Uncertainty about the Investment Decision

One way to cope with uncertainty about one's decisions is to look for advice (for overview see Bonaccio and Dalal, 2006). Advice helps people to optimize their choice and gain confidence in their decisions (Heath and Gonzalez, 1995; Budescu et al., 2003). This is especially true for financial investment decisions, which contain high risks, in particular the risk of losing money. However, an advisor's knowledge reduces such uncertainty and improves accuracy only if people can trust the advice they receive (Sniezek et al., 2004; Yaniv, 2004). This means that even if people decide to receive advice, they cannot be sure if the advice is of high quality.

To estimate the quality of the advice, customers use cues. One such cue is an advisor's expertise. Harvey and Fischer (1997) showed that people are twice as likely to accept advice from experts as to accept advice from novices. In particular they found that perceived expertise predicts subsequent advice utilization, suggesting that a person's impression of whether or not an advisor is an expert is very important to clients (de Vries and Wilke, 1995; Harvey and Fischer, 1997; Sniezek and Van Swol, 2001; Jungermann and Fischer, 2005). Thus, the perceived expertise of the advisor is a cue, which can reduce people's uncertainty regarding their decisions. With regard to high-risk decisions, also a second cue is important – the advisor's expressed confidence (for an overview, see Bonaccio and Dalal, 2006). If an advisor is perceived as highly confident, customers gain confidence themselves (Sniezek and Buckley, 1995; Sniezek and Van Swol, 2001; Price and Stone, 2004). As a consequence, recommendations of highly confident advisors are followed more often than those of less confident peers (Sniezek and Buckley, 1995; Sniezek and Van Swol, 2001). However, advisors often overestimate their confidence (Price and Stone, 2004; McKenzie et al., 2008) and use this overconfidence as a strategy to persuade customers (Van Swol, 2009). This phenomenon is also present in the context of financial consulting: participants confronted with a fictitious financial advisor preferred an overconfident advisor over a moderately confident advisor and even assumed the overconfident advisor to be more knowledgeable (Price and Stone, 2004). Thus, the expressed confidence of the advisor is another cue, which can reduce people's uncertainty regarding their decisions.

Accordingly, in situations where people are uncertain regarding their decisions, customers focus on cues allowing them to gain the needed confidence for their decisions. However, in financial consultancies, people do not only have to cope with uncertainty regarding the decision but also regarding the advisor's true intentions either to support the client or to pursue self-interest. Therefore, in addition to having to cope with uncertainty about their investment decision (UnD), clients also have to cope with uncertainty regarding the advisor (UnA).

# Uncertainty about the Advisor's Intention

Uncertainty regarding the advisor refers to the difficulty to identify his/her true intentions. Usually, the financial advisor possesses information and knowledge that the client lacks. The lack of knowledge makes it difficult for the client to assess whether the advisor is acting in the best interest of the client. According to Principal–Agent Theory (Ross, 1973; for an overview, see Eisenhardt, 1989), agents (advisors) who pursue their own goals instead of acting in the principal's (client's) best interest use their scope of action to behave in a strategic way (conflicting goals). In financial consulting, the advisor might neglect to carry out all necessary actions ("hidden action") such as searching for less risky investments, might withhold information about potential risks from the client ("hidden information"), or might hide their true intentions ("hidden intentions") and thus, unbalanced relationships follow. The client is dependent on the advisor and thus, susceptible to deception. The less the client knows about the advisor's actions, information, and intentions, the more the client's uncertainty increases.

Knowing the advisor's intentions is crucial to the client's continued reliance on the advisor (Bonaccio and Dalal, 2009). A study by Jodlbauer and Jonas (2011) showed that customers who did not know their advisor's true intentions but assumed their advisor to pursue self-interested intentions evaluated him/her less trustworthy. As a consequence, people were less likely to utilize the advisor's recommendations. So how can people cope with such uncertainty?

To find answers to this question, we build on two theoretical models from justice research – the Uncertainty Management Model (UMM) and the Fairness Heuristic Theory (FHT, Van den Bos et al., 1998; Lind, 2001; Van den Bos, 2001) which might

help to reduce uncertainty regarding the investment decision and uncertainty regarding the advisor.

# The Uncertainty Management Model

The Uncertainty Management Model (Van den Bos and Lind, 2002; see also Lind and Van den Bos, 2002) describes how people cope with general uncertainty. According to the UMM, uncertainty is a general and abstract concept that can also be induced in an abstract way (e.g., by thinking about uncertainty in terms of emotions or losing control or thinking about insecure aspects of one's own life; Van den Bos and Miedema, 2000; Van den Bos, 2001). Such general uncertainty, which makes people's future unpredictable increases people's sensitivity to fairness cues, e.g., whether one has voice in a given situation (procedural fairness; Van den Bos and Miedema, 2000; Van den Bos, 2001). People also react with more positive affect toward fair and more negative affect toward unfair treatment (Van den Bos, 2001). Thus, applied to investment decisions, fairness may be a valuable cue when people experience general uncertainty about their decisions.

# The Fairness Heuristic Theory

The Fairness Heuristic Theory (Van den Bos et al., 1998; Lind, 2001; Van den Bos, 2001) explains how people cope with uncertainty regarding a person's trustworthiness in an interdependent relationship, i.e., a relationship in which one is dependent on another person (e.g., employee and authority). Uncertainty regarding trustworthiness means that one does not know whether the other person will keep one's best interests in mind (Barber, 1983). Thus, fear of exploitation and exclusion is present (Van den Bos et al., 1998; Lind, 2001; Van den Bos, 2001). In such situations, people use fairness cues to decide whether to cooperate with the other person. For example, when participants were uncertain about the trustworthiness of an authority, they showed higher commitment to a decision when they were given voice (procedural fairness) compared to no voice (Van den Bos et al., 1998). This means that fairness can compensate for people's uncertainty and the resulting unwillingness to cooperate because of the interaction partner's trustworthiness. However, the authors also emphasize that fairness is not the same as trust. While fairness is an evaluation whether a person acts or decides morally correct, trust is a person's willingness to be vulnerable to the actions of another party. Therefore, trust always involves uncertainty regarding the risk of being exploited (Coleman, 1990; Moorman et al., 1992; Mayer et al., 1995). Based on this definition trust is something highly fragile (Mayer et al., 1995). The perception of trust in interdependent relationships is highly relevant when people decide whether to cooperate with a person or not (Van den Bos et al., 1998; Lind, 2001; Van den Bos, 2001). Research showed that clients who perceived self-interest on behalf of their advisors, mistrusted their advisors and were consequently less likely to accept advice (Jodlbauer and Jonas, 2011) and even showed aggressive intentions and negative attitudes toward the interaction partner (Steindl and Jonas, 2015).

Bank customers are in uncertain situations and do not know if they can trust their advisor. To investigate how people cope with such uncertainty we tested – based on the UMM and FHT – whether uncertainty increases people's sensitivity to fairness cues and increase their willingness to cooperate with financial advisors regarding their investment decisions.

# Dimensions of Fairness

When deciding whether or not a situation is fair people seem to distinguish between different aspects of fairness (or justice<sup>1</sup> ): the distribution of resources (distributive fairness; Adams, 1965), the provision and transparency of information (informational fairness; Bies and Moag, 1986), the procedures on which the decision is based (Thibaut and Walker, 1975; Leventhal, 1976, 1980), and whether they have been treated with respect and dignity (interpersonal fairness; Bies and Moag, 1986; for an overview, see Colquitt, 2001). These four aspects of fairness perceptions are used as cues to guide evaluation in uncertain situations (e.g., Van den Bos and Miedema, 2000; Lind, 2001; Van den Bos, 2001). Thus, the advisor's expressed fairness may also play an important role in financial consultancies characterized by high uncertainty. Here, fairness may compensate for uncertainty.

# The Current Research

In financial consultancies people are facing those two kinds of uncertainty—uncertainty regarding the financial investment decision in general and uncertainty regarding the advisor's true intentions. In particular in the course of the financial crisis starting in 2008, people were confronted with both uncertainties. Thus, we assume that they were facing general uncertainty about their financial decisions and were especially vigilant to financial advisors who acted in self-interested manners which increased their uncertainty about the advisors' trustworthiness. To investigate how both uncertainties worked together during the financial crisis, we integrated the UMM and the FHT. Both theories state that people need fairness cues to regain certainty. Therefore we predicted that perceived fairness is important when any uncertainty is high and especially important when both uncertainties are high. In the current study, we investigated how the two types of uncertainty influenced people's intended future cooperation with an advisor and whether fairness cues can promote trust and thus help to increase cooperation.

# MATERIALS AND METHODS

# Procedure

Participants were invited to take part in a survey on financial consulting<sup>2</sup> and asked if they had contact with a financial advisor in the last year. The data were collected in November 2009, shortly after the beginning of the financial crisis in Europe. In a questionnaire participants were asked to think about their experiences with "saving and investing money" over the last year. To answer the questions they were instructed to think about their financial service provider. If they were customers of different

<sup>1</sup> In the literature, primarily known as distributive, informational, interpersonal, and procedural justice (e.g., Colquitt, 2001).

<sup>2</sup>Our survey took place in November 2009; therefore uncertainty regarding investment should have been especially salient to participants.

service providers, they were instructed to concentrate on the provider with whom they had spent the most time in the last year and they were also asked how many consultancy meetings they had.

# Participants and Procedure

Two-hundred and forty-two participants answered our questionnaire. Due to missing data on the scales cooperation intention, UnA, and trust, we had to exclude nine participants, leaving us with a final sample of 233 (16–81<sup>3</sup> years old; Mage = 36.86, SD = 15.84; 138 women and 95 men). Our participants were either approached in a shopping mall when the University of Salzburg was having a public event there or via the experimenter's social networks. All participants were asked to fill out a questionnaire in which they had to think about their financial service providers and indicated their agreement to several items on a scale from 1 ( = not at all) to 6 ( = completely).

# Measures

#### Uncertainty

We assessed two aspects of customers' uncertainty about their financial service providers: uncertainty about their investment decisions and uncertainty about their advisor (the advisor's strategic behavior)<sup>4</sup> .

#### Uncertainty about Investment Decisions (UnD)

This measure focused on the uncertainty about investment decisions (whether they had made the right decision, e.g., regarding the choice of service provider or product). Five items asked about the success of the participants' investment (e.g., "Thinking about my investment in the last year raises the feeling that I reached the "right" decision regarding my financial service provider" inverted; α = 0.81, see complete questionnaire in the appendix).

# Uncertainty about the Advisor (UnA)

Participants were asked to think about the behavior of their financial advisor and describe his/her behavior with the help of the following items: the strategic behavior scale covered the three agency problems described above [hidden intention: e.g., "In situations where our interests were in conflict the advisor focused on his/her interests," seven items; hidden information: e.g., "I got the impression that my advisor did not communicate essential information about the protection of the money (e.g., a capitalback guarantee)," five items; hidden action: e.g., "My advisor handed over written information about my investment product after the contract conclusion (e.g., information with risk details)," seven items]. For further analyses we combined the subscales to produce a general uncertainty scale about the advisor (α = 0.93; see complete questionnaire in the appendix).

#### Fairness

Participants evaluated their advisor's fairness behavior with items from the German version (Maier et al., 2007) of the Organizational Fairness Scale (Colquitt, 2001). The items were adapted to the financial advisor situation. Two items were additionally developed and one original item<sup>5</sup> was excluded. Procedural fairness was assessed with eight items (e.g., "Have you been able to express your views and feelings during consultations with your advisor?"). Distributive fairness was represented with three items (e.g., "How appropriate are your returns considering the amount of money you invested?"). Informational fairness was measured with seven items (e.g., "Has your advisor explained the procedures thoroughly?") and interpersonal fairness with four items (e.g., "Have you been treated with dignity by your advisor?"). In the following calculations we combined these subscales to a general fairness scale (α = 0.94; see complete questionnaire in the appendix).

## Trust

Trust in the financial advisor was measured with items from a questionnaire designed by Schoorman and Ballinger (2006) with the main focus on the willingness to yield control to the advisor and the customers' accepted vulnerability. Five items were translated and adapted to the financial service sector (e.g., "I would be willing to let my advisor have complete control over my future investment decisions"; α = 0.56<sup>6</sup> ; see complete questionnaire in the appendix).

#### Dependent Variable—Cooperation Intention

This scale measured the intention of the participants to further use the consulting services of the current financial advisor (e.g., "I will still use the services of my financial advisor in the future"; five items, α = 0.82; see complete questionnaire in the appendix). For the means, standard deviations, and intercorrelations of all measured variables, see **Table 1**. 7

<sup>3</sup>Because of the financial practices in Austria, we also included 16- and 17-year old participants in the analyses. In Austria it is absolutely common for kids to have their own saving account (typically at the same financial institution as their parents) and when they turn to become teenagers, for example at the age of 16, they get a phone call once a year where they get invited to a financial consultancy. Teenagers under 18 years are already an important group of customers and get in contact with financial institutes regularly (in legal terms teenagers aged 14 and older become already contractually capable to the maximum content of their income). Additionally one of the most popular forms of saving is a so called "Bausparer" which is a long-term saving (6 years) where the government also pays a bonus. A lot of young people use this to afford, e.g., a driver's license, motorbike or car.

<sup>4</sup> In a confirmatory factor analysis including the two aspects of uncertainty (UnD: 5 items; UnA: 19 items), we compared a one-factor to a two-factor model. This analysis yielded a significant difference [one-factor: χ 2 (252) = 723.78, two-factor model: χ 2 (251) = 620.71), p < 0.001], indicating that the two-factor model is more tenable than the one-factor model.

<sup>5</sup> "To what extent does this result reflect the effort you carried out for the company?" This original item was excluded because the transfer in the consultancy context did not work. We added two items extending the subscale informational justice. They are marked in the questionnaire in the appendix as IFG2a and IFG3a. <sup>6</sup>We used the trust scale introduced by Schoorman and Ballinger (2006). As two items from the original questionnaire could not be adapted to the context of the financial consultancy ("I feel comfortable being creative because my advisor understands that sometimes creative solutions do not work"; "Increasing my vulnerability to criticism by my supervisor would be a mistake"), we deleted those two items. Unfortunately, Cronbach's alpha turned out to be very low (α = 0.55). However, this is according to some authors (e.g., Peterson, 1994) still in the acceptable range.

<sup>7</sup>We also measured their emotions with the help of the PANAS (Watson and Clark, 1988) to which we added some additional items (e.g., uncertain, skeptical etc.) and individual differences in emotional uncertainty with the Uncertainty Response

#### TABLE 1 | Intercorrelations between all measured variables.


Ratings were made on a six-point scale. ∗∗p < 0.01, two-tailed.

# RESULTS

According to the two theoretical models dealing with uncertainty—UMM (Lind and Van den Bos, 2002; Van den Bos and Lind, 2002) and FHT (Van den Bos et al., 1998; Lind, 2001; Van den Bos, 2001)—fairness plays a particularly important role when uncertainty is high. Therefore, we tested how the different types of uncertainty in combination with fairness influenced people's decision regarding future cooperation with the advisor.

# Multiple Regression Analysis

We conducted a multiple regression including three continuous predictors (UnA, UnD, and fairness as moderator; all standardized), and all of their interactions regarding the dependent variable intention to cooperate (Hayes, 2013; Model 3). We used a 95% bias corrected bootstrap confidence interval (95% BCCI) and 5,000 bootstrap samples. High values refer to one standard deviation above and low values to one standard deviation below the standardized values for the respective variable.

The analysis revealed significant main effects of UnD (β = −0.17, SE = 0.06, t(225) = −2.65, p = 0.009, [95% CI: −0.29 to −0.04]), UnA (β = −0.58, SE = 0.07, t(225) = −8.37, p < 0.001, [95% CI: −0.72 to −0.45]), and perceived fairness (β = 0.25, SE = 0.08, t(225) = 3.36, p = 0.001, [95% CI: 0.11 to 0.40]) indicating that all three variables (high UnD, high UnA, and low fairness) decreased people's cooperation intentions.

In addition and most importantly, the three-way interaction between UnA, UnD, and fairness was significant, β = −0.06, SE = 0.03, t(225) = −2.19, p = 0.030, [95% CI: −0.11 to −0.01]. Simple slopes indicated that when UnA and UnD were low, high and low fairness did not make a difference regarding people's intentions to cooperate (p = 0.773). However, when any kind of uncertainty was involved (high UnA or/and high UnD), high fairness led to more cooperation intentions than low fairness (all ps ≤ 0.041). Furthermore, when both uncertainties were high, people had the lowest intentions to cooperate in both the low and high fairness conditions (see **Figure 1**).

Moreover, the regression weights show that the weight for UnA was higher than for UnD. This suggests that UnA had a larger impact on client's cooperation intentions than UnD. In all conditions, whether fairness and/or UnD were high or low, high UnA always led to lower cooperation intentions than low UnA (all ps < 0.001). In contrast, high UnD (compared to low UnD) led to lower cooperation when UnA was high and fairness low (p = 0.023). See **Table 2** for a summary.

Additionally, we performed separate moderation analyses for the four fairness dimensions and found significant threeway interactions for the procedural fairness dimension, and marginal significant three-way interactions for the distributive and interpersonal fairness dimensions, all pointing into the same direction (procedural fairness × UnA × UnD, β = −0.06, SE = 0.03, t(225) = −2.19, p = 0.030; distributive fairness × UnA × UnD, β = −0.05, SE = 0.03, t(225) = −1.78, p = 0.076; interpersonal fairness × UnA × UnD, β = −0.06, SE = 0.03, t(225) = −1.67, p = 0.096; see **Tables 3–6**). This suggests that especially procedural fairness is important to compensate for the uncertainties.

To test whether a loss of trust in the advisor was responsible for the stronger influence of UnA compared to UnD on cooperation intentions we explored the role of trust as a mediating variable between the two types of uncertainty and cooperation intentions.

# Multiple Regression Analysis

We conducted a mediated moderation analysis using PROCESS SPSS macro (Hayes, 2013; Model 11) with the three standardized predictors UnA (as independent variable), UnD, and fairness (as moderator), trust (as mediator), and the dependent variable intention to cooperate. We used a 95% bias corrected bootstrap confidence interval (95% BCCI) and 5,000 bootstrap samples. The three-way interaction between UnA, UnD, and fairness on trust was marginally significant, β = 0.06, SE = 0.03, t(225) = 1.97, p = 0.050, [95% CI: 0.00 to 0.12]. Trust had a significant effect on cooperation intentions, β = 0.25, SE = 0.06, t(230) = 4.12, p < 0.001, [95% CI: 0.13 to 0.37]. The direct effect of UnA on cooperation intentions was significant as well, β = −0.67, SE = 0.06, t(225) = −11.09, p < 0.001, [95% CI: −0.79 to −0.55].

For the slopes, we found that trust mediated the conditional indirect effect of UnA on cooperation when UnD was high and fairness was low, β = −0.13, SE = 0.05, [95% CI: −0.25 to −0.06], as well when UnD was high and fairness was high, β = −0.16, SE = 0.06, [95% CI: −0.30 to −0.07], and when UnD was low and fairness was high β = −0.13, SE = 0.04, [95% CI: −0.23 to −0.07]. Thus, the analyses revealed a negative association with the customers' intention to cooperate in future in three conditions. However we did not observe that trust mediated the indirect effect of UnA on cooperation when UnD

Scale (Greco and Roger, 2001). Individual emotional uncertainty did not relate significantly to the intention to cooperate, r(234) = −0.020; p = 0.761.

FIGURE 1 | The effects of uncertainty about the advisor (UnA), uncertainty regarding the investment decision (UnD), and fairness on people's intention to cooperate with the advisor. Plotted values reflect intention to cooperate at one standard deviation below and above the standardized values for fairness. Significant slopes are marked with asterisks. Higher values reflect higher intention to cooperate.



TABLE 3 | Effects of the moderated regression analysis including procedural fairness (Hayes, 2013, Model 3).


and fairness were both low, β = −0.05, SE = 0.04, [95% CI: −0.15 to 0.03]. Please see **Table 7** for all slopes. This moderated mediation analysis shows that the negative relation between UnA and intention to cooperate can be explained via the mediator trust, which means that UnA leads to loss of trust which can better explain the reduced cooperation intention. Only when customers are confident in their decision (low UnD) and perceive low fairness, trust in the advisor does no longer seem to be involved in the process between UnA and intention to cooperate.

# DISCUSSION

We opened this paper by referring to the dilemma faced by financial consulting customers. On the one hand, customers want and need advice to reduce uncertainty regarding their risky decisions. On the other hand, they have to deal with uncertainty regarding the advisor's true intentions, i.e., whether they pursue their own interests instead of the clients' best interests. This dilemma describes two different forms of uncertainty: uncertainty regarding investment decisions and uncertainty

TABLE 4 | Effects of the moderated regression analysis including distributive fairness (Hayes, 2013, Model 3).


TABLE 5 | Effects of the moderated regression analysis including interpersonal fairness (Hayes, 2013, Model 3).


TABLE 6 | Effects of the moderated regression analysis including informational fairness (Hayes, 2013, Model 3).


TABLE 7 | Trust mediating the conditional indirect effects of X (UnA, Fairness, UnD) on Y (Cooperation Intention) at values of the variables (Hayes, 2013, Model 11).


regarding the advisor's intentions. The current study examined customers' intended future cooperation with an advisor when they were facing these two forms of uncertainty. Furthermore, as research has shown that in uncertain situations people need cues to regain their confidence, we introduced fairness as an important predictor for cooperation in consultancies.

We based our study on the UMM and the FHT predicting that people react more strongly to fair treatment when uncertainty salience is high. Therefore, we suggested that perceived fairness is important when any uncertainty is high and especially important when both uncertainties are high. In the present study we asked participants to think about their experiences with "saving and investing money," assessed their uncertainty regarding the decision and regarding the advisor's intention, and their perception of fairness in the consultancy. Our findings indicated that whenever people were uncertain in a consultancy, whether uncertain about the decision or/and advisor, their intentions to cooperate with the advisor increased with perceived fairness. Moreover, both uncertainties together led to the lowest intentions to cooperate.

Interestingly, a higher regression weight for UnA than for UnD indicated that in our study, UnA had a larger impact on client's cooperation intentions than UnD. Thus, high UnA seems to be most costly for financial advisors regarding the customers' intention to cooperate in the future. High experienced UnA seems to be especially detrimental for the relationship between the financial advisor and his/her customer. What changes in the relationship between customers and their advisors when they are confronted with uncertainty regarding the advisor's intention, i.e., when people face the risk of being exploited by their advisor? The mediation analysis suggests that trust can help to explain this process: because clients mistrust that their advisors act in the client's best interest and instead pursue their own goals, they refrain from cooperating with the advisor. However, if people want to work together in interdependent relationships, they need to trust each other (Van den Bos et al., 1998; Lind, 2001; Van den Bos, 2001). If customers, like Mrs X, face the dilemma that they need advice, but at same time are uncertain about their advisor's intention, they have to ask themselves if they can trust the advisor. When people are uncertain about another person's trustworthiness, trusting this person makes them even more vulnerable and thus, even more uncertain (Deutsch, 1962; Coleman, 1990; Moorman et al., 1992; Mayer et al., 1995).

In the current article, we have introduced fairness as a crucial cue in uncertain consultancies. Fairness is a cue displayed in the social interaction itself and thus differs from variables such as the advisor's expertise or confidence which stem from the classic Judge-Advisor Research (for overview Bonaccio and Dalal, 2006). According to the input-process-output model proposed by Bonaccio and Dalal (2006), expertise and expressed confidence of an advisor are both individual level inputs. Fairness is a process variable happening in the interaction between people. It consists of different dimensions which should be taken into account in consultancies because customers pay attention to more than just fair returns on the investment of their money (distributional fairness). They also wish to be treated with dignity and respect (interpersonal fairness), to receive sufficient information (informational fairness), and to understand how decisions are made (procedural fairness; for an overview, see Colquitt, 2001). According to our results, people need fairness cues if they perceive high uncertainty. In our study procedural fairness seemed to be especially important. Thus, customers who had the feeling that they could participate in the investment decision process increased their intention to cooperate with the advisor.

# Theoretical Implications

Prior research has established FHT and UMM to explain how persons cope with these two types of uncertainty. Studies have supported each of these theoretical frameworks and the two kinds of uncertainty have been investigated separately. Thus, previous studies found fairness cues as especially helpful in highly general uncertain situations (UMM) or authority-oriented based situations (FHT). The current results expanded this line of reasoning by taking into account both uncertainties together and found that both are important in explaining cooperation intentions but that uncertainty regarding the advisor was more influential. However, they did not reinforce each other. This might be different in other consultancy situations. For example in health-related contexts – when people ask themselves what the best decision is and whether they can trust their advisors but they are anxiously aroused because they worry about their health – cues that induce mistrust might lead to overreactions. A potential mediator in this situation might be perceived loss of control. In future research, it would be fruitful to investigate the influence of different kinds of uncertainties and their relationship in various advisor–client interactions and to shed light on relevant mediators.

# Practical Implications

Resulting from the economic crisis, people seem to be highly aware of both types of uncertainty (UnA and UnD) and even more vigilant to advisor's trustworthiness. Therefore, since then, bank customers may evaluate the financial consultancy through the eyes of uncertainty. Consequently, it is important for banks to develop strategies that help their customers overcome this uncertainty. Our findings suggest that fair treatment might be such a strategy. Fairness increased people's intentions to cooperate even when uncertainty was high. Nevertheless, it is essential to not understand fairness as a substitute which can easily eliminate people's uncertainty. Rather, our findings indicated the lowest willingness to cooperate with the advisor when UnA and UnD was high. Fairness slightly improved the willingness but did not set people's willingness back to baseline (when there was no uncertainty present). Thus, the way how advisors treat their customers and explain their products is critical and should not be neglected in the day-to-day running of a financial business. Fair treatment during consultancy is one way to help customers to reduce uncertainty and regain trust in their financial advisors.

# Limitations and Future Directions

An advantage and disadvantage at the same time is that we collected our data in the field. On the one hand, a limitation of our research is that we could not control for further variables (loss of money, risk taking, etc.), on the other hand and at the same time it may be an advantage because we were able to directly examine how people cope with the insecure situation after the economic crisis. Moreover, manipulating a general uncertainty (UMM) and personal uncertainty (FHT) independently of each other in the

laboratory could have helped to understand the interplay between different types of uncertainty and the role of fairness in this interplay. It would be important for future research to investigate different types of uncertainty in combination with the risk of losing real money (e.g., provide investment products with high vs. low risk) in the controlled environment of the laboratory. For future research it would also be interesting to investigate the long-term process of uncertainty and the reconstruction of trust in the advisor. Uncertainty regarding the advisor's intentions might lead to mistrust, and fair treatment might be a way to regain trust. We are aware that we only measured people's intention to cooperate in the future with the advisor, which is not the same as real behavior. Nevertheless, we believe that future studies might benefit from our findings and we hope that researchers further investigate the positive effect of fairness on cooperation in uncertain financial conditions.

# CONCLUSION

Uncertainties are part of our daily lives and especially part of our social interactions. In particular in financial consultancies, where uncertainty regarding an investment decision and uncertainty regarding the advisor's intentions are high, we need cues helping us to deal with those uncertainties. We identified fairness as one cue helping people to compensate for uncertainty. Furthermore, our results indicate that customers' uncertainty about the advisor

# REFERENCES


is most costly for the customers' intention to cooperate in future. However, our further analysis could identify loss of trust as a mediator, which can explain why customers facing high uncertainty about the advisor are not willing anymore to cooperate with their financial advisor.

We are grateful to Barbara Grall for helping in the data collection and to Thomas Scherndl for his statistical advice.

# ETHICS STATEMENT

Although this study used human subjects, the approval of an ethical committee does not seemed necessary. The study was only about non-intrusive issues.

# AUTHOR CONTRIBUTIONS

BM and EJ developed together the research question, hypothesis and the research design. BM and CM did mainly the data analysis; together we did the writing of the manuscript.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpsyg. 2017.01112/full#supplementary-material


eds J. Greenberg and R. Cropanzano (Palo Alto, CA: Stanford University Press), 56–88.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Mackinger, Jonas and Mühlberger. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Using Motivational Interviewing to reduce threats in conversations about environmental behavior

#### *Florian E. Klonek1\*, Amelie V. Güntner2, Nale Lehmann-Willenbrock2 and Simone Kauffeld1*

#### *Edited by:*

*Christina Steindl, University of Salzburg, Austria*

#### *Reviewed by:*

*Isabella Uhl, University of Salzburg, Austria Janet Kleber, Alpen-Adria-Universität Klagenfurt, Austria*

#### *\*Correspondence:*

*Florian E. Klonek, Department of Industrial/Organizational and Social Psychology, Institute of Psychology, Technische Universität Braunschweig, Spielmannstraße 19, Braunschweig 38106, Germany f.klonek@tu-bs.de*

#### *Specialty section:*

*This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology*

*Received: 24 November 2014 Accepted: 06 July 2015 Published: 21 July 2015*

#### *Citation:*

*Klonek FE, Güntner AV, Lehmann-Willenbrock N and Kauffeld S (2015) Using Motivational Interviewing to reduce threats in conversations about environmental behavior. Front. Psychol. 6:1015. doi: 10.3389/fpsyg.2015.01015* *<sup>1</sup> Department of Industrial/Organizational and Social Psychology, Institute of Psychology, Technische Universität Braunschweig, Braunschweig, Germany, <sup>2</sup> Department of Experimental and Applied Psychology, Faculty of Psychology and Education, VU University Amsterdam, Amsterdam, Netherlands*

Human behavior contributes to a waste of environmental resources and our society is looking for ways to reduce this problem. However, humans may perceive feedback about their environmental behavior as threatening. According to self-determination theory (SDT), threats decrease intrinsic motivation for behavior change. According to self-affirmation theory (SAT), threats can harm individuals' self-integrity. Therefore, individuals should show self-defensive biases, e.g., in terms of presenting counterarguments when presented with environmental behavior change. The current study examines how change recipients respond to threats from change agents in interactions about environmental behavior change. Moreover, we investigate how Motivational Interviewing (MI) — an intervention aimed at increasing intrinsic motivation — can reduce threats at both the social and cognitive level. We videotaped 68 dyadic interactions with change agents who either did or did not use MI (control group). We coded agents verbal threats and recipients' verbal expressions of motivation. Recipients also rated agents' level of confrontation and empathy (i.e., cognitive reactions). As hypothesized, threats were significantly lower when change agents used MI. Perceived confrontations converged with observable social behavior of change agents in both groups. Moreover, behavioral threats showed a negative association with change recipients' expressed motivation (i.e., reasons to change). Contrary to our expectations, we found no relation between change agents' verbal threats and change recipients' verbally expressed selfdefenses (i.e., sustain talk). Our results imply that MI reduces the adverse impact of threats in conversations about environmental behavior change on both the social and cognitive level. We discuss theoretical implications of our study in the context of SAT and SDT and suggest practical implications for environmental change agents in organizations.

Keywords: Motivational Interviewing, environmental behavior, intervention study, interaction analysis, self-determination theory

# Introduction

Most of us have an understanding that natural and energy resources are finite and have considered our own carbon footprint. However, even if we assume that people today are generally aware of the importance of pro-environmental behavior, there is still a considerable discrepancy between committing to pro-environmental norms and the actual behavior that contributes to environmental protection (Séguin et al., 1998; Vining and Ebreo, 2002; Castro et al., 2009; Fritsche et al., 2010).

A possible explanation for this discrepancy can be derived from the notion that confrontation with one's own poor environmental behavior and its expected consequences can elicit individual perceptions of threat. The intention behind many pro-environmental initiatives is that individuals will respond to negative feedback about their environmental behavior by changing their behavior and becoming more environmentally friendly. However, people often resist such feedback and therefore lack positive behavior change in terms of more sustainable activities (Sherman and Cohen, 2002, 2006). Information about climate change is potentially threatening to individuals because it implies changes and constraints in human living conditions (Fritsche et al., 2012). More generally speaking, "threats result from some experience of discrepancy between an expectation or desire and the current circumstance" (Jonas et al., 2014, p. 229).

In order to alleviate individual responses to threats in social interactions, we introduce Motivational Interviewing (MI) as a communication approach. We explicate how MI can prevent potential threats in social interactions about behavior change by reducing the actual amount of threats (social effect) as well as alleviating the negative perception of threats (cognitive effect). Furthermore, we argue for an interplay between change agents'1 verbal behavior and the verbally expressed motivation of their change recipients.

# Confrontational Behavior: Threats in Social Interactions

A substantial body of research has focused on various reasons why individuals may experience threats in social interactions (e.g., Festinger, 1957; Brehm, 1966, 1993; Steele, 1988; Jost and Banaji, 1994; Blascovich and Mendes, 2000; Gawronski, 2012; Jonas et al., 2014). Means by which people feel threatened are, for example, verbal confrontations with a certain discrepancy at present. Previous research argues that the experience of discrepancy between a desired state and the actual situation entails challenges to the fulfillment of individuals' psychological needs (Jonas et al., 2014).

To understand sources of threat in social interactions, one research stream highlights individuals' need for self-relevant clarity and cognitive consistency (Hogg, 2007; Swann, 2011), while other studies have focused on individuals' need for selfworth or self-integrity (e.g., Tajfel and Turner, 1979; Steele, 1988). Other studies have focused on the overall need for order and stability. For example, system justification theory (Jost and Banaji, 1994) states that individuals have the need to create and maintain a favorable self-image, also labeled as "ego justification." This justification tends to occur at the expense of others (e.g., group interests) and undermines individuals' motivation to change their behavior or attitude (Jost and Banaji, 1994; Jost et al., 2010).

A second stream of research emphasizes individuals' need for personal control (Kay et al., 2009; Leotti et al., 2010), behavioral autonomy (Deci and Ryan, 2000, 2002), and freedom in general (Brehm, 1966, 1993). For example, Brehm's (1966, 1993) theory of psychological reactance illustrates that individuals believe that they hold the freedom to engage in behaviors as they please — based on the satisfaction of their needs. When this perceived behavioral freedom is threatened by a persuasive message, individuals are motivated to reinstate that particular freedom. Consequently, they engage in defensive behaviors that are directed away from the behavior that the persuasive message targets on. This effect has also been labeled as the "boomerang effect" (Brehm, 1966, 1993).

Based on these previous findings and in line with Jonas et al. (2014), we conclude that the experience of discrepancy and unfulfilled personal needs can be considered a primary source of threat. We apply this assumption to the case of feedback about environmental behavior: That is, individuals who are confronted with a discrepancy between their current environmental behavior and a desired pro-environmental behavior, experience this as a threat. Seen through a social psychology lens, we offer the following possible explanations underlying subsequent defensive responses.

# Self-Defense and Motivational Responses from a Social Psychology Perspective

There are several reasons why negative feedback about environmental behavior may elicit perceptions of threat. First, such negative feedback can threaten an individual's need for self-integrity, because the suggestion that one has harmed the environment (e.g., by littering or by leaving the lights on) conflicts with the belief about being an environmentally conscious person. According to self-affirmation theory (SAT; Steele, 1988), individuals have a fundamental motivation to protect their personal image of self-integrity, in terms of seeing themselves as moral, adaptive, and capable of controlling important outcomes. When confronted with self-threatening information, individuals show self-defense responses such as denying threatening information, presenting counterarguments, or expressing resistance to change in order to restore their self-integrity (Sherman and Cohen, 2002, 2006).

Second, individuals may perceive negative verbal feedback that asks for environmental behavior change as a threat to their autonomous decision-making ability (Osbaldiston and Sheldon, 2003). According to self-determination theory (SDT; Deci and Ryan, 2000, 2002; Ryan and Deci, 2000, 2002), individuals can experience motivation and self-determination in their behaviors

<sup>1</sup>We will use the more generic terms 'change agent' throughout this paper for individuals that promote behavior change (e.g., therapists, physicians, advisors, environmental inspectors, block leaders) and 'change recipients' for individuals that carry out behavior change (e.g., clients, patients, conversational partners, environmental inspectors, employees).

only when the need for autonomy is fulfilled (Ryan et al., 1995; Deci and Ryan, 2000, 2002; Ryan and Deci, 2000, 2002). SDT argues that humans have a natural interest in pursuing selfdetermined goals and behaviors, rather than pursuing goals directed by external forces. In line with SDT, we assume that the development of motivation is strongly dependent on social context, such as the relevant interaction partners (Deci and Ryan, 1987). In particular, we argue that individuals may experience feedback about their environmental behavior as a potential threat to their autonomy and self-determination. As a result, individuals are less likely to develop the necessary motivation to respond with environmental behavior change. There is preliminary support for this general idea in environmental psychology research. For example, Osbaldiston and Sheldon (2003) asked their participants to commit to pro-environmental goals. The authors reported participants' perceptions of the experimenters' autonomy-supportive behavior were positively linked to greater internalized motivation.

In sum, we conclude that individuals may react with defensive responses in order to restore their psychological needs when they experience a discrepancy between a desired and their current behavior (Jonas et al., 2014). However, in the face of necessary environmental behavior change, such defensive responses are problematic. When individuals resist threatening environment messages, positive behavior change in terms of more sustainable activities becomes unlikely (Sherman and Cohen, 2002, 2006). This highlights the challenge of bringing about environmentally friendly behavior change. What is needed to address this challenge is a means to create circumstances under which individuals do not perceive negative feedback about their environmental behavior as a threat to their self-integrity as well as to their perceived autonomy.

#### Reducing Threats: The Case for MI

As one such means, we introduce the intervention method of MI. When considering conversations about environmental behavior change, MI can provide a practical skill that helps preserve the self-integrity and autonomy of the conversational partner (e.g., Markland et al., 2005; Vansteenkiste and Sheldon, 2006; Leffingwell et al., 2007; Vansteenkiste et al., 2012). MI is defined as a "collaborative conversation style for strengthening a person's own motivation and commitment to change" (Miller and Rollnick, 2013, p. 12). Although it remains to be seen how MI fares in the context of environmental behavior change, several meta-analytic studies have established MI as an evidence-based intervention method in facilitating behavior change in the clinical context (e.g., Hettema et al., 2005; Rubak et al., 2005; Carroll et al., 2006; Lundahl et al., 2010; Magill et al., 2014).

We have outlined that individuals may perceive feedback about their environmental behavior as threatening. MI may alleviate this tension by asking change agents to abstain from confrontations or from trying to impose strategies for behavior change (Miller and Rollnick, 2013). In other words, we argue that MI prevents change recipients from threats to their selfintegrity. MI provides methods to deal with client resistance and to support clients' self-efficacy by validating negative client statements (Werner et al., 2009) and by selectively attending

to clients' verbal expressions in favor of change (i.e., change talk, Miller and Rose, 2009; Glynn and Moyers, 2010). Change talk might include statements such as "There are certainly some options to save energy" or "I am going to implement this right away." By contrast, sustain talk captures clients' verbal expressions against change, such as "It is just not so simple, not while I have all my work to do" or "But that's the way we always did it."

Furthermore, individuals may experience feedback about (poor) environmental behavior as a potential threat to their perceived ability in making autonomous decisions. MI addresses this need for autonomy through the belief that the recipient, rather than the change agent, is the primary source of ideas and solutions for accomplishing behavior change. Accordingly, change agents should serve the need for autonomy in deciding about solutions for reaching these goals. Moreover, more autonomously regulated behaviors are not only executed more persistently, but also with greater care and quality (Ryan and Deci, 2000).

In sum, we argue that MI as a communication method serves the fulfillment of psychological needs, particularly the need for self-integrity and autonomy (e.g., Steele, 1988; Deci and Ryan, 2000, 2002; Vansteenkiste and Sheldon, 2006; Leffingwell et al., 2007). MI can create the very conditions that help reduce perceived threat to self-integrity and autonomy when individuals are confronted with negative feedback. As a result, MI is likely to foster individuals' motivation to engage in pro-environmental behavior.

Motivational Interviewing as a communication method has been applied in numerous behavior change settings, such as reducing risk behavior (e.g., Colby et al., 1998; McCambridge and Strang, 2004), treating psychological problems (Burke, 2011), or promoting healthy behavior (cf., Lundahl et al., 2010; Perry and Butterworth, 2011). Whereas MI was traditionally taught to practitioners in the helping professions, it is also highly suitable for the context of environmental behavior (Tribble, 2008; Forsberg et al., 2014; Klonek and Kauffeld, 2015). For example, Klonek and Kauffeld (2015) suggest that human resources departments could provide MI training for energy managers in order to reduce organizational energy consumption. In support of this argument, Forsberg et al. (2014) showed that MI training increased the conversational skills of Swedish environmental inspectors. Furthermore, the authors showed that MI training positively affected inspectors' empathy (rated by external observers).

However, several gaps in the literature remain. First, whereas the assessment of empathy by means of independent observers aligns with most previous research in the context of MI applications (e.g., Catley et al., 2006; Tollison et al., 2008; Forsberg et al., 2014), this approach tends to neglect the actual perspective of the change recipients. In other words, change recipients should be best suited to evaluate whether their interactional partner has confronted them or demonstrated empathic understanding. We aim to address this gap in the literature by considering both social and cognitive reactions by change recipients. A second gap concerns the lack of human interaction in some previous work on MI in the context of environmental behavior. Tribble (2008) conducted a laboratory computer experiment in which participants performed a decisional balance task – which is typical of a MI intervention – that is, they listed pros and cons about changing their environmental behavior. The intervention had no effect on the environmental outcome scores. Tribble (2008) attributed this to the fact that the MI task was realized within a computer environment, that is, it lacked the expression of empathy by a real human being. A third gap in the literature (e.g., considering the study by Forsberg et al., 2014) is the lack of data on client verbal responses (in terms of change talk and sustain talk). Taken together, we aim to address these gaps in the literature (1) by examining a human interactionbased MI intervention that focuses on environmental behavior change, (2) by considering both agent and recipient behavior within the conversation process, and (3) by including cognitive reactions to the change conversation, in terms of change recipients' perceptions about confrontation and empathy of change agents.

# Hypothesis Building Social Effects of MI

Avoiding autonomy-restrictive or verbal threats is one of the core principles in the application of MI (e.g., Miller et al., 2004). MI advises not to confront change recipients with direct argumentation or suggestions for change, but instead to encourage them to develop their solutions (Miller and Rollnick, 2013). Hence, we expect that MI should become expressed in terms of the actual verbal behavior of change agents. Specifically, change agents using MI should show fewer verbal behavioral threats such as suggestions, confrontations, or argumentations about behavior change toward more environmentally conscious behavioral conduct. In other words, we expect that feedback by change agents who use MI involves a lower amount of actual behavioral threats than feedback by change agents who do not use MI. Concerning the social effects of MI skills, we hypothesize the following:

H1: Using MI in environmental feedback conversations reduces the amount of verbal threats by change agents.

#### Cognitive Effects of MI

In addition to avoiding autonomy-restrictive behavior, showing empathic behavior has been considered just as important in the use of MI and for effective communication in general (Miller and Mount, 2001; Miller et al., 2004; Miller and Rose, 2009; Moyers and Miller, 2013). Previous research has demonstrated the relevance of empathy for perceptions of safety and trust in physician–patient relationships and, in turn, for the quality of conversational outcomes (e.g., Miller et al., 1980; Hojat et al., 2002; Moyers et al., 2005; Nicolai et al., 2007; Gaume et al., 2009). Although these previous findings were obtained in clinical settings, we assume that the core value of empathic communication can apply in the context of environmental feedback as well.

Empathy can be considered a multidimensional construct that can function on either a cognitive or an affective level (Davis, 1983; Galinsky et al., 2008; Reniers et al., 2011). The affective component of empathy refers to the capacity to show appropriate emotional reactions toward other people and to the experiences they articulate. The cognitive component of empathy describes the ability of an individual to perceive the world from another person's perspective by putting oneself in that person's position (Davis, 1983; Reniers et al., 2011).

In the context of MI, empathy is served by means of verbally communicating empathic understanding through reflective listening. Reflective listening means that change agents paraphrase verbal statements of their conversational partners (also termed empathic back-channeling; Miller and Rollnick, 2013). That is, change agents need to put themselves in the position of the recipient. Previous studies suggest that putting oneself in the position of another person, i.e., perspective taking, improves the relationship that is necessary for collaboratively reaching solutions (e.g., Parker and Axtell, 2001; Galinsky et al., 2008; Steindl and Jonas, 2012). Thus, change recipients should experience a conversation with MI change agents as highly empathic.

Further, we can assume that reflective listening in social interactions helps change recipients pay attention to their own argumentation (for behavior change) instead of being pushed toward pro-environmental behavior by a change agent (cf., Hettema et al., 2005). For example, instead of saying "You should realize that your behavior harms the environment," change agents using MI might reflect a previous recipient statement as "You mentioned that you realized that your behavior harms the environment." Even though, the content of both sentences in this example might impose a threat (i.e., a discrepancy between recipients' current behavior and desired behavior), the MI conversational style should alleviate change recipients' perception of threats during a conversation. Consequently, change recipients' cognitions about being confronted about their environmental feedback should be reduced through the use of the MI conversational style.

H2: Using MI in environmental feedback affects change recipients' cognitions about the social interaction in terms of (a) increasing perceived empathy and (b) decreasing perceived confrontation.

# The Interplay between Social Behaviors and Cognitions

Further, the present study aims at gaining insight into the interrelation between verbal behaviors of change agents and cognitions of change recipients in social interactions about behavior change. The underlying mechanism is that change agents interpersonal behaviors within the social interaction shape clients' cognitions in a way that also affects their motivation to change a specific behavior. Specifically, we expect that change agents who use MI demonstrate fewer threats when giving feedback about environmental behavior. As a result, change recipients should perceive MI-feedback as less confrontational. The link between change agents' social behavior and change recipients' social cognitions can prevent negative conversational

dynamics that are typically evoked by confrontational change agents (Miller and Rollnick, 2002; Klonek et al., 2014). In sum, we argue for an interplay between change agents' verbal threats and clients' corresponding cognitions. Put formally:

H3: Change agents' social behavior and recipients' perceptions of agents' behavior are intertwined, such that the amount of observed threats is positively related to perceived confrontation and negatively related to perceived empathy.

## SAT and SDT based Predictions of Threats in Conversations about Environmental Behavior

We have outlined in a previous section how SAT and SDT can help explain interpersonal dynamics in social interactions about behavior change. Next, we derive predictions regarding agent– recipient dynamics from both theories.

Based on SDT (Deci and Ryan, 2000, 2002; Ryan and Deci, 2000, 2002), change agents should address recipients' need for autonomy in order to evoke intrinsic motivation. The less change recipients are threatened with behavior change, the more likely they will make self-determined decisions about behavior change. The interpersonal dynamics derived from SDT are depicted in **Figure 1**. Verbal threats negatively affect self-determination of participants, that is, verbal threats that feedback a gap between actual and desired environmental behavior will harm the intrinsic motivation of change recipients. As illustrated in **Figure 1**, on the cognitive level, change recipients might have thoughts during the interaction such as "I'm an adult and I can make my own decisions." As a result, change recipients should be less likely to show change talk on the observable interpersonal behavior level. In other words, change recipients should be less likely to provide change-directed utterances such as "I think I should travel less" that indicate that they have their own reasons for change (Miller

and Rose, 2009; Glynn and Moyers, 2010). Therefore, we expect that:

H4: Change agents' verbal threats will be negatively related to change recipients' motivation (i.e., change talk).

Different from SDT, SAT (Steele, 1988) posits that threats about behavior change are negatively affecting change recipients' self-integrity or self-esteem (see **Figure 1**). In order to protect their self-esteem, change recipients can use self-defense strategies such as counter-arguments, denying threatening information, or expressing resistance to change (Sherman and Cohen, 2002, 2006). That is, based on SAT, we assume that change recipients who are threatened with behavior change are more likely to protect themselves using counter-change language (i.e., selfdefenses). In MI, counter-change language is captured in terms of sustain talk (Miller and Rose, 2009; Glynn and Moyers, 2010). An example of counter-change language that change recipients could use to defend their self-integrity would be "I see this differently! I do not see a way to change this." We hypothesize:

H5: Change agents' verbal threats will be positively related to change recipients' sustain talk.

Meta-analytic findings on the use of MI in clinical settings provide initial evidence for these predictions derived from SDT on the one hand and SAT on the other hand. Magill et al. (2014) found that verbal threats (i.e., MI non-adherent behavior) were linked to an increase of sustain talk and a decrease of clients' motivation in terms of change talk. Although this interplay between change agents' and recipients' verbal behaviors has not yet been examined in the context of environmental behavior

SDT, self-determination theory.

change to date, we assume similar behavioral linkages in this domain.

#### Current Research

In order to test our hypotheses, we compared interpersonal (i.e., verbal behavior) and cognitive variables (i.e., change recipients' perceptions of confrontation and empathy) between two groups of dyadic change-related conversations. In each dyad, change agents discussed discrepancies between current environmental behavior and ideal pro-environmental behavior. In one group of dyads, change agents applied MI (intervention group), whereas the other group of change agents served as a control group. Control change agents were instructed to give change recipients feedback on their environmental behavior and motivate them to improve their behavior.

The data for this study were gathered as part of a larger research project. We have reported detailed information about the reliability of our observational coding instrument in a previous publication (Klonek et al., 2015b).

In the present study, we investigated how verbal threats of social change agents differed between the two social-interaction based approaches (H1). To do so, we enumerated the frequencies of change agents' observed MI non-adherent behaviors and compared these between the MI and control group.

We also investigated how verbal threats (i.e., MI non-adherent behavior) in both groups affect social cognitions of change recipients (H2 and H3). To do so, we measured change recipients cognitive reactions after the interview.

Moreover, we were interested in the interpersonal dynamics between change agents' verbal threats and change recipients' motivation and self-defense (H4 and H5). Therefore, we also counted change recipients' instances of change and sustain talk during the conversation and related the frequency of these instances to change agents' social behavior.

Finally, we also measured environmental attitude and environmental intentions using a validated survey instrument (SEU-3; Schahn, 1999 and Schahn et al., 2000) before and after the intervention, in order to investigate whether the MI intervention positively affected environmental outcome measure in comparison to the control group.

# Materials and Methods

#### Sample

We excluded nine conversations from our analysis because change recipients in those dyads did not provide self-report data on empathy and confrontation. Our final sample contained 68 dyadic conversations about environmental behavior change. Twenty-six participants took the role of a change agent. The majority of the change agents in the MI group were enrolled as psychology majors (*n* = 14); one change agent in the MI group was enrolled in human resources development. The average age of the MI agents was 29 (SD = 8.33). The gender ratio was balanced with seven male change agents. Change agents in the control group (*n* = 11) were 31 years on average (SD = 13.63). The gender ratio was balanced (six female, five male). Six of the change agents in the control group had a technical or natural science background, and five change agents had a background in psychology or the social sciences.

Change agents in the MI group (*n* = 15) received training in MI before they took part in this study. The MI training comprised 21 h over a period of 3 months as part of their psychology coursework. The training was designed according to the eight stages of learning MI (Miller and Moyers, 2006) and contained exercises from the Motivational Interviewing Network of Trainers (2008) manual.

Change agents in both groups had a conversation with a change recipient about pro-environmental behavior change. All change agents were instructed to motivate their change recipients to increase pro-environmental behavior and to work out individual measures that participants should implement. In order to include a manipulation check of our intervention, change agents were asked to indicated their level of familiarity and proficiency in MI on a five-point Likert-type scale ranging from 1 (strongly disagree) to 5 (strongly agree). Change agents in the MI group scored significantly higher on familiarity with (M = 3.79 vs. M = 1.6, *p <* 0.01) and proficiency in MI (M = 2.93 vs. M = 1.8, *p <* 0.01), in comparison to change agents from the control group.

For a second and more objective manipulation check of our intervention variable, we also assessed change agents' MI spirit using a rating scale from the German MITI-d (MI Treatment Integrity; Brueck et al., 2009). The MI spirit scale assesses the overall skillfulness in using MI and indicates the extent to which the change agents in our sample showed evocation, collaboration and autonomy within the conversation. External observers were asked to estimate change agents' overall adherence to MI principles (Moyers et al., 2010).

Two observers rated the extent to which change agents showed MI spirit on a seven-point scale (1 = weak adherence, 7 = strong adherence). The extremes of the rating scales were verbally anchored with the definition of strong/weak MI spirit adherence. ICCs and Cronbach's αwere used to estimate inter-rater reliability for this measure (ICC = 0.76 and α = 0.87). Change agents in the MI group received significantly higher MI spirit adherence scores (M = 5.77) in comparison to change agents from the control group [M = 4.28, *t*(23.98) = −5.67, *p <* 0.01]. A value of 5 and higher on the MI spirit adherence scale is considered as basic proficiency in MI, whereas a value of 6 and higher is considered as solid proficiency in MI (Brueck et al., 2009).

Change recipients (*n* = 68) were 24 years old on average (SD = 7.83) and the majority was female (78%, *n* = 53). Most of them had a high school degree (76.5%, *n* = 52), 15% (*n* = 10) had a university degree, 7.4% had finished a vocational training (*n* = 5), and 1.5% (*n* = 1) held at least a secondary school-level education. Age, gender, prior vocational training, and educational level did not differ significantly between the MI and the control group. Prior to the conversation, participants in both groups reported their environmental behavior on a 28-item environmental behavior scale by Schahn (1999) and Schahn et al. (2000); e.g., "I only use energy-saving devices in order to save electric energy") on a five-point response format (1 = strongly disagree to 5 = strongly agree). We found no differences across the two groups regarding this measure (M = 3.36 in the MI and control groups, *p* = 0.96).

#### Procedure

Prior to data gathering, the study protocols were approved by the institutional review board for data security. All participants gave informed consent for videotaping their conversations. Based on their availability, change recipients were allocated to a conversation with either a change agent in the MI group (*n* = 49) or in the control group (*n* = 19). Change agents and recipients were unaware of the hypotheses of this study. Three sessions (two in the intervention and one in the control group) could not be recorded due to technical problems. Before each conversation started, the change recipients completed the self-report measure about their environmental behavior. The change agents received this information in order to talk about environmental behaviors that could be improved and a short written agenda that listed the following topics: setting the agenda, asking about current environmental behavior, giving feedback about environmental behavior to change recipients, asking for measures to increasing pro-environmental behavior, and planning measures/giving advice (see Appendices A–D for detailed information).

# Change Recipients' Environmental Behavior, Attitude, and Intentions

We measured environmental behavior, attitudes, and environmental intentions using three validated 28-item measures from Schahn (1999) and Schahn et al. (2000). A sample item for environmental behavior was "I only use energy-saving devices in order to save electric energy." A sample item for environmental attitude was "In my opinion, grocery shops still sell too many environmental harmful products." A sample item for environmental intentions was "In the future, I will specifically ask for environmentally friendly products and ask the grocery store to change the assortment of goods accordingly." Environmental attitudes and intentions were measured before and immediately after the conversation. Internal consistencies for environmental behavior (α = 0.70) was acceptable; internal consistencies for environmental attitude (α = 0.87–0.88) and environmental intentions (α = 0.88–0.86) were good (Kline, 2000).

# Environmental Action Plan

All change recipients had the opportunity to fill out a change plan worksheet after the interview (based on Kauffeld et al., 2009; Magill et al., 2010). This worksheet listed the sentence "I will carry out the following measures" followed by consecutively numbered opened spaces for writing down intended actions. The number of measures was summed up to derive a measure of "number of environmental actions."

# Change Agents' Verbal Threats

We measured verbal threats of change agents with an observational coding instrument from MI (i.e., the MITId, Brueck et al., 2009). The German MITI-d includes the behavior code "MI non-adherent behavior." This behavior code encompasses autonomy-restrictive behaviors, such as "confrontations" (e.g., directly and unambiguously disagreeing, arguing, correcting, shaming, blaming, criticizing, labeling, moralizing, or ridiculing), "advising without permission" (making suggestions, offering unsolicited advice), and "directing" (e.g., giving orders, commands, or imperatives). Within conversations about environmental behavior change, MI non-adherent behaviors comprised feedback about discrepancies between a desired environmental behavior and the current circumstance (i.e., threats) using a tone of uneven power sharing, disapproval, or negativity.

# Change Recipients' Social Reactions of Intrinsic Motivation and Self-Defense

We operationalized motivation and resistance to change (i.e., as a sign of self-defense) by coding change recipients' verbal reactions during the videotaped conversations. In line with previous research (Miller et al., 2008; Klonek et al., 2015a), statements with a positive inclination toward change were coded as change talk (e.g., "When I am not home the entire day, I do not need the lights on."), whereas statements that had a negative inclination toward change were coded as sustain talk (e.g., "In my opinion, changing my behavior will not make a difference," **Table 1**). Verbal statements containing no apparent inclination toward or against change were coded as follow/neutral (e.g., "I don't have a washing machine"). We further subclassified change and sustain talk utterances into reasons, activation/other, taking steps, and commitment to change or to sustain, respectively (cf., Miller et al., 2008). We used change talk codes as measures of verbally expressed motivation and sustain talk codes as measures of verbally expressed resistance and self-defense.

In order to adjust for time differences between conversations, we standardized the frequencies for each behavioral code to a 10 minute interval (i.e., "rates"; cf. Bakeman and Quera, 2011, p. 96 and 101).



<sup>1</sup>*This code is labeled as "Other" in the MI skill code (Miller et al., 2008).*

# Inter-Rater Reliability

Two independent observers were extensively trained in classifying the verbal behaviors of change agents and change recipients. They received a set of graded learning tasks, including scripted interactions from the developer of the MITI-d, MI video material (Demmel and Peltenburg, 2006), and recorded demo-interactions for learning how to code MI-relevant behavior. To train for coding change recipients' verbal behaviors, we provided additional transcripts and audio material (Project MILES, 2011). Further training material for training observers in using the MITI-d is also available online (cf., supplemental material in Klonek et al., 2015b). The videos were coded using INTERACT software (Mangold, 2010). For detailed psychometric information about the software-supported coding scheme, see Klonek et al. (2015b). A random sample of 13 interactions (19%) was coded twice. We calculated the Intraclass Correlations (ICC) for these sessions to obtain a measure of observer reliability for change and sustain talk measures. The ICC is a statistical index commonly used to estimate reliability because it adjusts for chance agreement and systematic differences between observers (Fleiss and Shrout, 1978; McGraw and Wong, 1996, p. 35). We classified all obtained ICC values according to the criteria proposed by Cicchetti (1994). Except for our measure of "sustain talk–taking steps" (ICC = 0.36), all behavioral codes in our sample showed fair to excellent inter-rater reliability (ICCs = 0.51–0.91).

# Cognitive Reactions

In terms of cognitive reactions to the change conversations, we measured change recipients' perceptions of empathy and confrontation immediately after the videotaped conversations. We slightly adapted the Rating Scales for the Assessment of Empathic Communication (REM, Nicolai et al., 2007), that is, we exchanged the term "doctor" from the original version with the term "interviewer." The REM is a two-factorial instrument with six items measuring empathy and three items measuring confrontation (Nicolai et al., 2007). The confrontation scale measures the perception of the change recipient of being talked down or being admonished (e.g., "Did the interviewer admonish you?"; M = 1.49; SD = 0.66; α = 0.73). By contrast, the empathy scale measures the change recipients' perception about the change agents' empathic behavior (e.g., "the interviewer put himself/herself in my position"; M = 4.64; SD = 0.42; α = 0.79). Change recipients rated each of these items on a five-point Likert scale that was behaviorally anchored for each item (cf., Nicolai et al., 2007; e.g., 1 = not at all; 5 = a lot).

### Overview of Statistical Analysis

Change agents in both groups (MI and control) had a conversation about environmental behavior change with 1–3 recipients. This one-to-many design results in interdependence between change agents and change recipients, that is, one change agent is nested within several clients, which can result in biased statistical parameters (cf., Kenny et al., 2006). In order to allow an evaluation of unbiased statistical tests, we randomly selected a subsample of unique agent-recipient dyads. We report analyses for the interdependent dyads sample (*n* = 68) and the reduced sample of unique agent–recipient dyads (*n* = 26).

# Results

### Social and Cognitive Effects of MI

Our first hypothesis posited that MI in environmental feedback conversations reduces the amount of verbal threats by change agents. We used the intervention type (MI vs. control group) as independent variable and used *t*-tests for independent samples to compare the group means for MI non-adherent behavior of change agents. When considering the interdependent sample, change agents in the MI group showed significantly less observable MI non-adherent behaviors in comparison to change agents in the control group [*t*(17.05) = 3.96, *p <* 0.01, *<sup>d</sup>* <sup>=</sup> 0.83; see **Table 2**]. Whereas the rate of MI non-adherent behavior by change agents in the MI group was close to zero (M = 0.06), change agents in the control condition threatened change recipients' about making changes at least two times within any 10-minute interval (M = 2.05). We replicated these findings in the unique dyads sample. This finding lends support to H1 by showing that MI indeed affects the social level of the interaction, in terms of the observable behavior of change agents.

Our second hypothesis stated that MI in environmental feedback affects change recipients' cognitions about the interaction in terms of (a) increasing perceived empathy and (b) decreasing perceived confrontation. We calculated *t*-tests to compare group means between the MI and control group concerning perceived confrontation and perceived empathy. When considering the interdependent dyad sample, perceived confrontation and empathy were significantly different between

TABLE 2 | Comparisons of social and cognitive outcomes between the MI group and the control group.


*MI, Motivational Interviewing; M, mean; SD, standard deviations are indicated in brackets;* <sup>∗</sup>*p < 0.05;* ∗∗*p < 0.01;* †*p < 0.05 (one-tailed).*

the MI group and the control group (**Table 2**). Change recipients in the MI group perceived confrontation from change agents to a significantly lesser extent (M = 1.37) compared to change recipients in the control condition [M = 1.79; *t*(65) = 2.39, *p <* 0.05, *d* = 0.65]. Change recipients in the MI group also perceived higher empathy in comparison to change recipients in the control group [M = 4.72 vs. M = 4.46, *t*(22.95) = −1.98, *p* = 0.07, *d* = −0.52]. We replicated these results in the unique dyad sample, with the exception of perceived empathy which did not differ between groups.

Although there was no consistent difference on the perceived empathy, these findings mostly support H2 concerning positive effects of MI on change recipients' social cognitions about the interaction.

# Interplay between Social Behavior and Cognitions

Our third hypothesis was that change agents' social behavior and recipients' perceptions of agents' behavior are intertwined, such that the amount of verbal threats is positively related to perceived confrontation and negatively related to perceived empathy. To test this hypothesis, we calculated intercorrelations between perceived confrontation, perceived empathy, and MI non-adherent behavior by change agents. **Table 3** shows that the observed MI non-adherent behavior by change agents was strongly correlated with change recipients' perceptions of being confronted (*r* = 0.54, *p <* 0.01). By contrast, there was a substantial negative correlation between MI non-adherent behavior and perceived empathy (*r* = −0.55, *p <* 0.01). In other words, an increase in MI non-adherent behavior was associated with an increased perception by change recipients that his/her change agent behaved in a confrontational manner. These results lend support to H3 concerning the social– cognitive interplay between the interaction behavior of change agents and the resulting cognitions in the respective change recipients. We also calculated these correlations separately for each group and found similar results. The rate of observed MI non-adherent behavior was positively correlated with perceived confrontation in MI dyads (*r* = 0.30, *p <* 0.05, *n* = 46) and control dyads (*r* = 0.76, *p <* 0.01, *n* = 18). Moreover, the rate of observed MI non-adherent behavior correlated negatively with perceived empathy both in MI dyads (*r* = −0.29, *p <* 0.05, *n* = 46) and control dyads (*r* = −0.66, *p <* 0.01, *n* = 18).

We replicated these findings in the unique dyad sub-sample, the direction and statistical significance of correlations were comparable (see **Table 2**). In the unique dyads sample, we also calculated correlations separately for each group (i.e., MI versus control group agents). Again, observed MI non-adherent behavior was positively correlated with perceived confrontation in MI dyads (*r* = 0.48, *p* = 0.068, *n* = 15) and control dyads (*r* = 0.78, *p <* 0.01, *n* = 11). Observed MI non-adherent behavior was negatively correlated with perceived empathy in MI dyads (*r* = −0.56, *p <* 0.05, *n* = 15) and control dyads (*r* = −0.63, *p <* 0.01, *n* = 11). These results support our third hypothesis that change agents' social behavior and recipients' perceptions of agents' behavior are connected.

# Change Recipients' Verbal Response to Threats in Social Interactions about Environmental Behavior Change

Our fourth and fifth hypotheses stated that verbal threats of change agents will be negatively related with change recipients' expressed motivation (i.e., change talk; H4) and positively related with change recipients' sustain talk (H5). To test these hypotheses, we calculated correlations between MI non-adherent behaviors by change agents and the respective verbal responses by change recipients (i.e., change talk or sustain talk). **Table 4** shows the results of these analyses separately for the interdependent and for the unique dyad sample. In both samples, MI non-adherent behaviors showed a substantial negative correlation with reasons to change (r = −0.40, *p <* 0.01). That is, the more change agents provided verbal threats to change environmental behavior, the less change recipients provided self-motivational statements such as reasoning why change might be beneficial for them. There were no significant associations between MI non-adherent behaviors and change recipients' sustain talk measures. Moreover, although we did not hypothesize about linkages between perceived empathy and recipient behavior, we still explored linkages between these variables. It would be reasonable to expect that change agents' empathy would address recipients' need for relatedness. Hence, we conducted additional *post hoc* analyses using perceived empathy as an independent variable and recipients' verbal responses (i.e., change and sustain talk) as dependent variables. **Table 4** shows that perceived agent empathy was positively related with reasons to change in both the unique dyad sample (*r* = 0.42, *p <* 0.01) and in the interdependent sample (*r* = 0.38, *p* = 0.053). In other words, the more change agents were perceived as empathic listeners that could relate to recipients' problems, the more recipients showed intrinsically motivated behavior,


*MI, Motivational Interviewing; REM, Rating scales for the assessment of empathic communication.* ∗∗*p < 0.01;* <sup>a</sup>*Cronbach's alpha;* <sup>b</sup>*Intraclass correlation.*



*Pearson's correlations, two-tailed. MI, Motivational Interviewing; ICC, Intraclass correlation.* ∗∗*p < 0.01;* <sup>∗</sup>*p < 0.05;* †*p < 0.10.*

in terms of stating reasons to change their environmental behavior.

# Effects of the MI Training Intervention on Environmental Attitudes and Intentions

We conducted separate two-factorial ANOVAs with the withinfactor "time" (pre vs. post-conversation) and the betweenfactor "intervention type" (MI vs. control group) for the two dependent variables environmental attitudes and environmental intentions. Participants' pro-environmental attitudes [*M*<sup>0</sup> = 3.69; *<sup>M</sup>*<sup>1</sup> <sup>=</sup> 3.80; *<sup>F</sup>*(1,66) <sup>=</sup> 5.16, *<sup>p</sup> <sup>&</sup>lt;* 0.05, <sup>η</sup><sup>2</sup> <sup>p</sup> = 0.07] and pro-environmental intentions increased significantly over time [*M*<sup>0</sup> = 3.43; *M*<sup>1</sup> = 3.64; *F*(1,66) = 55.07, *p <* 0.01, η<sup>2</sup> <sup>p</sup> = 0.46]. That is, conversations about pro-environmental behavior change positively affected environmental attitudes and intentions in both groups. We found no main effect for the factor "intervention type," neither for the dependent variable environmental attitudes [*M*MI = 3.75; *M*Control = 3.74; *F*(1,66) = 0.01, *p* = 0.93, η2 <sup>p</sup> = 0.00] nor for the dependent variable environmental intentions [*M*MI = 3.53; *M*Control = 3.57; *F*(1,66) = 0.12, *p* = 0.73, η2 <sup>p</sup> = 0.00]. There was also no significant two-factorial interaction (time × intervention type), neither for environmental attitudes [*F*(1,66) <sup>=</sup> 0.84, *<sup>p</sup>* <sup>=</sup> 0.36, <sup>η</sup><sup>2</sup> <sup>p</sup> = 0.01] nor for environmental intentions [*F*(1,66) <sup>=</sup> 1.22, *<sup>p</sup>* <sup>=</sup> 0.27, <sup>η</sup><sup>2</sup> <sup>p</sup> = 0.02]. These findings from the interdependent dyad sample were replicated in the unique dyad sample.

Overall, these results indicate that the conversation about environmental behavior change positively affected participants' environmental intentions and attitude. However, the improvement in environmental attitudes and intentions did not differ substantially between the MI conditions and the control group.

#### Environmental Action Plan

Finally, the number of planned environmental actions did not significantly differ between MI and the control group [*M*MI = 3.51; *M*Control = 4.16; *t*(66) = 1.2, *p* = 0.24, *d* = 0.32]. These findings were replicated in the unique dyad sample [*M*MI = 3.67; *M*Control = 4.00; *t*(24) = 0.52, *p* = 0.60, *d* = 0.21].

# Discussion

This study examined the cognitive and social reactions to potentially threatening conversations about pro-environmental behavior change. On the theoretical basis of SAT and SDT, we argued that individual feedback about environmental behavior (that is at odds with expected pro-environmental behavior) can pose a threat to individuals. Furthermore, social change agents can make these threats more salient by directly confronting or directing recipients toward a desired environmental behavior. According to SAT, participants use self-defense strategies to diminish such threats. Moreover, SDT suggests that threats harm the need of participants' autonomy and may therefore diminish participants' intrinsic motivation. In addition to examining the social and cognitive responses to potentially threatening feedback, we examined how feedback delivered using MI methods can help reduce perceived threats in social interactions.

Four main findings accrue from this study. First, MI affected the social level of interactions in terms of reducing change agents' verbal behaviors that are potentially threatening to change recipients (i.e., confrontations, warnings, autonomy-restrictive behavior). Moreover, MI showed benefits for the cognitive level of interactions, in terms of increasing change recipients' perceptions of a non-confrontational conversation.

Second, our findings illustrate the interplay between change agents' verbal behaviors and recipients' cognitions. Potentially threatening behaviors by change agents were negatively linked to recipients' perceptions of empathy and positively linked to perceptions of being confronted.

Third, our results show how participants respond verbally to threats in social interactions. As hypothesized, verbal threats by change agents were negatively related to change recipients' expressed motivation (i.e., verbal statements in favor of proenvironmental behavior change). Contrary to our expectations, however, verbal threats were not meaningfully related to change recipients' sustain language (i.e., verbal self-defense strategies).

Fourth, the change recipients in our study experienced an improvement in pro-environmental attitudes and proenvironmental intentions as a result of the conversation, regardless of the condition (MI versus control group).

Our findings have several theoretical implications for research on the interplay of cognition and motivation in the face of threats during social interactions in general, and for research on environmental behavior change in particular. First, we argued that discussing discrepancies between a desired state (i.e., improving environmental behavior) and the current (often imperfect) state of environmental behavior can be potentially threatening (cf. Jonas et al., 2014). Drawing from SAT (Steele, 1988) and SDT (Deci and Ryan, 2000, 2002; Ryan and Deci, 2000, 2002), we argued how such threats within social interactions can negatively affect interactional dynamics. Our findings underscore theoretical linkages between SDT and MI as discussed in previous research (see e.g., Markland et al., 2005; Vansteenkiste and Sheldon, 2006; Leffingwell et al., 2007).

Second, our study offers empirical support for the beneficial cognitive effects of MI in dyadic interactions, as indicated by our finding that MI change agents were perceived as significantly less confrontational compared to change agents in the control group. This result is in line with the notion that "[o]ne 'active ingredient' in MI may simply be a decrease in unhelpful counselor responses" (Miller and Rollnick, 2013, p. 381). However, cognitions about being confronted with behavior change could make favorable behavior change on the part of the recipient less likely (e.g., Francis et al., 2005; Klonek et al., 2014). In order to account for these cognitions, we highlighted recipients' perceptions of change agents' verbal threats as a relevant aspect for understanding the influence of MI.

Third, our finding that verbal threats by change agents were meaningfully connected to recipients' perceptions of behavioral threats aligns with previous findings from the medical field (Nicolai et al., 2007). However, our study extends this previous work in that we investigated the link between social behavior and cognitive appraisals of said behavior during environmental feedback conversations. Whereas medical feedback typically concerns problems of personal relevance, this does not necessarily apply to feedback regarding environmental behavior. Our findings thus contribute to the external validity of MI applications in the domain of environmental research and social interactions about environmental behavior. This is an important theoretical implication since MI is increasingly applied by environmental inspectors as a social change intervention (Forsberg et al., 2014). A review of the MI literature concluded that MI studies should also account for recipients' perceptions of the interaction (Madson et al., 2009). In addressing this call, we integrated observer-based measures of behavioral threats and change recipient's respective cognitive reactions.

Fourth, our findings shed light on the ways in which people respond to verbal threats in social interactions about environmental behavior change. That is, we compared both SAT and SDT in terms of their prediction for the underlying interactional mechanisms. SDT states that when individuals' need for autonomy is threatened, they express less motivation for behavior change. Our results suggest that the core principle of SDT (Deci and Ryan, 2000, 2002; Ryan and Deci, 2000, 2002) finds practical implementation in the use of MI. The principle of autonomy is implemented by asking change agents to avoid confrontational behavior (e.g., Miller et al., 2004). The current study tentatively supports this SDT-based explanation of how MI might unfold its positive effect by showing that the amount of verbal confrontations was indeed negatively associated with change recipients' expressed motivation, operationalized as verbally stated reasons to change. However, we found no negative relationships between verbal confrontations by change agents and other measures of change talk (i.e., activation, taking steps, commitment). Additional analyses revealed that empathy by change agents was positively related to change recipients' motivational language (i.e., reasons to change). Although we abstained from formulating any hypotheses regarding this link, this result does align with the notion derived from SDT that change agents should address interactional partners' need for relatedness in order to evoke intrinsic motivation. Moreover, because we calculated correlations, we can only conclude in terms of a general connection between the variables, instead of drawing directional conclusions.

Contrary to our expectations, our results do not align with the core tenet of SAT, which argues that threats to individuals' self-integrity trigger defense responses such as using counterarguments in order to restore their self-integrity. Our results showed no significant associations between change agents' verbal threats and change recipients' expressions of self-defense, operationalized as sustain talk. However, since this is the first study to investigate this relationship in the domain of environmental behavior, we would not necessarily discard the "threat results in defense" mechanism as posited in SAT. A metaanalysis of clinical studies in MI with over 1,000 participants reports a small (*r* = 0.07) but significant link between verbal threats (i.e., MI non-adherent behavior) and sustain talk (Magill et al., 2014). Our environmental behavior change intervention only supported the negative link between verbal threats and motivation to change (that was also reported by Magill et al., 2014). Our substantially smaller sample size and limited statistical power, in comparison to Magill et al. (2014), may have precluded us from establishing a significant relationship between threats and sustain talk. Moreover, verbal threats might affect recipients' self-defense mechanisms in other ways, such as emotional and cognitive rather than verbal responses. This suggests a need for including measures of emotional and cognitive self-defense mechanisms in addition to observations of recipients' verbal behavior in future research.

Finally, our findings indicate benefits for pro-environmental behavior not only in the MI condition, but also in the control group, suggesting an "any talk is good"-effect. However, our analyses also showed that the MI spirit of change agents in the control group was comparably high, presumably because control change agents also had a social science and psychology background. Miller and Rollnick (2013) argued that MI may benefit from a contrast effect, that is, MI will outperform social interaction based interventions that are mainly based on confronting and threatening change recipients to induce behavior change. Our control group condition may not have provided as extreme a contrast as previous studies comparing MI against confrontational therapeutic measures (e.g., in the treatment of alcohol addiction; see, e.g., Miller et al., 1993). Nevertheless, our findings lend themselves to implications for conversations about (environmental) behavior change in organizational practice.

# Practical Implications

Organizations are increasingly implementing social change agents in order to increase employees' motivation for sustainable behavior (e.g., Kraft and Neubeck, 2004; Carrico and Riemer, 2011; Steiner et al., 2011; Kauran, 2013; Klonek and Kauffeld, 2015). For organizational practice, our findings imply that change agents aiming to improve environmental conservation behaviors need to show interaction behavior that adheres to MI principles and aligns with the core tenets of SDT (i.e., change agents should respect recipients' need for autonomy and facilitate self-determined decisions). For example, an environmental intervention study by Werner et al. (2012, p. 419) used social change agents to promote energy-saving behavior and reported that agents "did not explicitly ask students to commit to turning off lights, but did ask if students were "on board" and willing to help [.] (...) [W]e did not want to use a stronger request that might feel coercive to students."

The authors reported that the conversational intervention outperformed a condition in which a sign (i.e., sticker) reminded change recipients to conserve energy. Our study adds implications for communicative interventions (or sociointeraction based interventions) by highlighting the need for change agents to respect recipients' need for autonomy, in line with SDT.

To assess whether this need is being met in change agent/recipient interactions, organizations could implement quality assurance measures (e.g., the MI spirit rating scale from the MITI). Finally, in order to reduce perceived confrontation during such interactions, organizations should consider training change agents in MI (see also Klonek and Kauffeld, 2015). Discussing discrepancies between current energy use and organizational benchmarks to save a certain amount of energy would likely not only be perceived by employees as less threatening but could also yield the intended behavior change as a result.

# Limitations and Future Directions

We acknowledge several limitations of this study. First, we gathered a student sample rather than studying environmental change agents and recipients in the field. However, it could certainly be argued that graduate students should show conscious environmental behavioral conduct similar to employees of organizations. Future research should aim to replicate our results in a field setting, although this may posit several challenges. Participants are often reluctant to allow video recordings in the field (Spee and Jarzabkowski, 2011; Meinecke and Lehmann-Willenbrock, 2015) and change agents do not always succeed in recording their conversations (Forsberg et al., 2014). Nevertheless, future studies should investigate the association of verbal threats and recipients' response in field settings and with larger sample sizes in order to substantiate our results.

Second, our measure of recipients' motivation to change showed somewhat limited reliability, such that the results for this measure should be interpreted cautiously. Previous research on observed change and sustain talk in clinical settings has reported similarly low ICCs for recipients' verbal codes (e.g., Baer et al., 2008; Gaume et al., 2008; Magill et al., 2010). Future research should strive to improve the reliability of observing recipients' motivation for change and further refine the coding instrument, perhaps by adapting it to certain features of the specific interaction context (in our case, pro-environmental conversations).

Third, whereas we obtained both social and cognitive measures, we did not include self-reports to measure recipients' motivation. Our decision to focus on observations of their verbal behavior was guided by prior research in MI. Following this approach allowed us to draw comparisons between our findings and previous research on MI in clinical settings (for an overview, see Magill et al., 2014). Moreover, previous research from the clinical field indicates that observational measures of motivation for or against change (e.g., observed resistance) have greater predictive value for behavior change compared to self-reported motivation measures, which is likely due to the elimination of self-representation bias when using observational measures (Westra, 2011). Future research should investigate whether this applies to the area of environmental change intervention as well. Moreover, based on our study, we cannot draw conclusions about actual behavior change. However, it would be interesting to assess how the use of MI in conversations about environmental behavior change might influence recipients' actual environmental behavior. Future research should strive to include follow-up measures that allow conclusions about actual change in behavior and lifestyle resulting from conversations where MI is applied.

Fourth, this study has shed light on the psychosocial and cognitive effects of MI in dyadic conversations about environmental behavior change. However, we did not consider the neural mechanisms that may be associated with threats and change language during social interactions. The analysis of a neuronal mechanism via the use of functional magnetic resonance or electroencephalography during face-to-face interactions is methodologically still very challenging. Nonetheless, there is preliminary evidence that change and sustain talk of change recipients have a neuronal correlate (Feldstein Ewing et al., 2011, 2014; Houck et al., 2013). Moreover, a recent neurocognitive study by Feldstein Ewing et al. (2014) emphasized that self-motivating change talk has to be generated from the change recipient rather than simply by repeating change-related statements. This underscores the core idea behind MI that motivation to change is created via the socialinteractive component of behavioral change interventions (Feldstein Ewing et al., 2014).

Fifth, the sample size of participants in the MI group was twice as large as participants in the control group. This implies larger SE in the control group and reduced power to detect significant differences between the two groups. Moreover, because change agents talked on average with three clients, interdependence between dyads should be considered. However, given our small sample of change agents, the analysis of multi-level models would result in biased parameter estimates (Maas and Hox, 2005). Therefore, we reran all analyses using a unique dyad sample. Future studies should incorporate larger change agent samples in order account for statistical interdependence by testing for actor effects, partner effects, and relationship effects.

Finally, whereas our study examined MI skills in the context of dyadic interactions about environmental behavior, future research should also investigate how MI skills can be implemented in change interventions aimed at the group or community level. For instance, one study has successfully applied MI for changing water disinfection practices within communities in Zambia (Thevos et al., 2000). Future research should examine how MI can facilitate environmental education that targets a broader audience, for example for increasing environmental engagement within an entire municipality.

In sum, this paper offers the following contributions: First, we integrated theoretical perspectives of SAT and SDT in order to investigate how participants respond to threats in social interactions. Next, we identified MI as a fruitful communication approach to reduce threats in social interactions and to promoting more favorable social and cognitive reactions in conversations about environmental behavior change. We tested this approach in a field experiment and tested how actual verbal threats and the perception of those threats differed in dyads in which an MI approach is used, in contrast to a control group. Our findings showcase the influence of interactional dynamics on both social and cognitive responses to threats during conversations about environmental behavior. We show how behavioral observations can provide a nonobtrusive measurement of motivation during conversations about behavior change. No study to date has assessed change recipients' perceptions and observational language measures during a conversation about environmental behavior change with change agents who use MI. We addressed this research gap while also adding to the few studies that have already investigated the use of MI in the context of environmental behavior (Tribble,

# References


2008; Forsberg et al., 2014). We discussed theoretical implications of our findings in the context of SAT and SDT and for organizational practice, particularly concerning environmental behavior change initiatives. Moreover, we demonstrated the value of a multi-method approach that integrates subjective perceptions and observational measures for gaining insights into the conversational dynamics surrounding perceived threats in social interactions and for untangling the interplay between interactional partners' behaviors and cognitions.

# Conclusion

This study provides support for the benefits of MI in terms of reducing actual verbal expressions of threats as well as alleviating negative perceptions of threats during social interactions. By connecting both the social and the cognitive level of conversational partners, our study contributes to a deeper understanding of people's responses to threats in social interactions.

# Acknowledgments

This project was supported by grants from the German Federal Ministry of Economics and Technology (BMWi; Grant 03ET1 0048). The authors wish to thank Bjoern Matthaei and Eva Ostmann for their coding work in this project.

# Supplementary Material

The Supplementary Material for this article can be found online at: http://journal*.*frontiersin*.*org/article/10*.*3389/fpsyg*.* 2015*.*01015


elaboration." *J. Environ. Psychol.* 29, 193–202. doi: 10.1016/j.jenvp.2009. 02.003

Westra, H. A. (2011). Comparing the predictive capacity of observed in-session resistance to self-reported motivation in cognitive behavioral therapy. *Behav. Res. Ther.* 49, 106–113. doi: 10.1016/j.brat.2010.11.007

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2015 Klonek, Güntner, Lehmann-Willenbrock and Kauffeld. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Unstable power threatens the powerful and challenges the powerless: evidence from cardiovascular markers of motivation

*Daan Scheepers1,2\*, Charlotte Röell1 and Naomi Ellemers1,2*

*<sup>1</sup> Social and Organizational Psychology Unit, Institute of Psychology, Leiden University, Leiden, Netherlands, <sup>2</sup> Leiden Institute for Brain and Cognition, Leiden University, Leiden, Netherlands*

Possessing social power has psychological and biological benefits. For example, during task interactions, people high in power are more likely to display a benign cardiovascular (CV) response pattern indicative of "challenge" whereas people low in power are more likely to display a maladaptive CV pattern indicative of "threat" (Scheepers et al., 2012). Challenge is marked by high cardiac output (CO) and low total peripheral resistance (TPR), while threat is marked by low CO and high TPR (Blascovich and Mendes, 2010). In the current work we addressed a possible moderator of the power-threat/challenge relationship, namely the *stability* of power. We examined the influence of the stability of power (roles could or could not change) on CV responses during a dyadic task where one person was the "chief designer" (high power) and one person was the "assistant" (low power). During the task, different CV-measures were taken [CO, TPR, heart rate, pre-ejection period). Whereas participants in the unstable low power condition showed a stronger tendency toward challenge, participants in the unstable high power condition showed a stronger tendency toward threat. Moreover, participants in the stable low power condition showed CV signs of task *dis*engagement. Results are discussed in terms of the importance of contextual variables in shaping the relationship between power and benign/maladaptive physiological responses.

#### Keywords: social power, social interaction, cardiovascular responses, challenge, threat

# Introduction

Social power—the ability to allocate or withhold resources from others—is one of the primary factors determining behavior in interpersonal, intra-group and inter-group settings (Cartwright, 1959; Mulder, 1977; Ng, 1980; De Waal, 1982; Galinsky et al., 2003; Keltner et al., 2003; Fiske and Berdahl, 2007; Guinote, 2007). Power not only defines the structural relationship between people (who "leads" and who "follows"), it also determines the affective and physiological responses of power holders and their subordinates, the cognitive strategies they use for solving problems, and the (non-verbal) behavior they direct to each other (Anderson and Berdahl, 2002; Keltner et al., 2003; Tiedens and Fragale, 2003; Anderson and Galinsky, 2006; Smith and Trope, 2006; Carney et al., 2010).

On the basis of the limited control they have, it seems only logical to assume that the powerless experience more stress than the powerful. There is indeed evidence for this assumption, also at

#### *Edited by:*

*Eva Jonas, University of Salzburg, Austria*

#### *Reviewed by:*

*Mike Prentice, University of Missouri, USA Cindy Harmon-Jones, University of New South Wales, Australia*

#### *\*Correspondence:*

*Daan Scheepers, Social and Organizational Psychology Unit, Institute of Psychology, Leiden University, Wassenaarseweg 52, 2333 AK Leiden, Netherlands scheepersdt@fsw.leidenuniv.nl*

#### *Specialty section:*

*This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology*

*Received: 29 November 2014 Accepted: 14 May 2015 Published: 27 May 2015*

#### *Citation:*

*Scheepers D, Röell C and Ellemers N (2015) Unstable power threatens the powerful and challenges the powerless: evidence from cardiovascular markers of motivation. Front. Psychol. 6:720. doi: 10.3389/fpsyg.2015.00720* the psychophysiological level. For example, compared to those high in power, those low in power do often have higher levels of cortisol (Wirth et al., 2006; Carney et al., 2010; Mehta and Josephs, 2010) and tend to display a cardiovascular (CV) response profile indicative of threat (Scheepers et al., 2012; Akinola and Mendes, 2013; Kraus and Mendes, 2014). Over time, frequent occurrence of these physiological response patterns can have serious negative health consequences (Chen et al., 2010).

However, as we will elaborate in more detail below, it is not self-evident that people with high power are always better-off in terms of stress. In the current research we examine the influence of the *stability* of power on CV-indices indicative of threat and challenge during a dyadic task situation that is characterized by a power difference. Our central hypothesis is that when power is unstable the powerless are relatively more challenged, and the powerful are relatively more threatened.

In the past 15 years, research on power has flourished within the field of social cognition (Galinsky et al., 2003; Keltner et al., 2003; Fiske and Berdahl, 2007; Guinote, 2007). A leading hypothesis in this field has been that possessing high power leads to activation of the behavioral activation system (BAS), while possessing low power leads to activation of the behavioral inhibition system (BIS; Keltner et al., 2003; Smith and Bargh, 2008). Related models have linked power to goal activation and goal pursuit (Guinote, 2007) and to a more general inclination to take action (Galinsky et al., 2003). In research revealing evidence for these predictions, power has typically been manipulated by means of priming tasks or specific role assignments in a (dyadic) task situation. Based on these models and procedures, low power has for example been related to a closed body posture, negative affect, a focus on possible losses, concrete thinking, and the avoidance of risks, while high power has been related to a more open body posture, positive affect, a focus on possible gains, abstract thinking, and risk-seeking behavior (Anderson and Berdahl, 2002; Keltner et al., 2003; Anderson and Galinsky, 2006; Smith and Trope, 2006; Carney et al., 2010).

The basic hypotheses regarding power, approach, avoidance, and related motivational tendencies have also been examined at the psychophysiological level. For example, using EEG measurements, Boksem et al. (2012), found that research participants who were primed with high (vs. low) power displayed a relative increase in left-sided frontal brain activity, which has been related to activation of the BAS. Furthermore, research on the neuroendocrine correlates of power has shown that high power is related to low levels of cortisol and high levels of testosterone (Schultheiss et al., 2005; Wirth et al., 2006; Carney et al., 2010; Akinola and Mendes, 2013) which represents a marker of approach tendencies during demanding situations (Mehta and Josephs, 2010).

Previous work on CV responses to power differences is most relevant to the current research. For example, Van Kleef et al. (2008) demonstrated that high power led to stronger parasympathetic regulation of heart rate (HR; as measured by *respiratory sinus arrhythmia*; RSA) during a potentially stressful event. An increase in RSA is indicative of more effective down-regulation of negative affect in demanding situations. This finding is in keeping with research that has applied the biopsychosocial model of challenge and threat (BPS-CT; Blascovich and Mendes, 2010) to the psychology of power. The BPS-CT, which is outlined in more detail below, identifies specific CV indices of "challenge" and "threat" motivational states. Building on this model, Scheepers et al. (2012) showed that research participants primed with high power displayed a challenge response while participants primed with low power displayed a threat response. Akinola and Mendes(2013) obtained similar results during a dyadic cooperative task where one participant was the "leader" and one person was the "support person."

The evidence summarized above may lead to the impression that people in relatively high power positions are in a quite comfortable position, while people low in power are more or less chronically threatened. This seems to be an oversimplification, however. Results from previous research suggest that there are important moderators for the relation between power and approach/avoidance tendencies, such as the perceived legitimacy and/or stability of the power difference (Magee et al., 2005; Maner et al., 2007; Lammers et al., 2008). For example, Lammers et al. (2008) showed that high power is only positively related to approach when the power had been obtained in a legitimate way. By contrast, when power differentials were based on an illegitimate procedure, low power was positively related to approach. This latter response has been interpreted as a sign of revolt by the powerless person. Regarding the stability of power, Maner et al. (2007) found that the classic effect that power leads to more risky decision-making was only observed when power relations were stable. Indeed, when power differentials were unstable the powerful people became more conservative in their decision-making.

The current research builds on this previous work but extends existing insights by addressing the more basic motivational and psychophysiological responses stemming from power (in)stability. Our central hypothesis is that the beneficial effects of high power (i.e., challenge) should only emerge when power relations are *stable*. When power differences are *unstable*—this should elicit threat in the powerful, as their privileged position is subject to change in the future, and challenge in the powerless, for whom there is scope to improve their position (see also Sapolsky, 2005).

Animal research provides indirect evidence for this hypothesis. In his seminal work on stress in primates, Sapolsky (2005) examined the relation between social rank and neuroendocrine response patterns. This research clearly indicated that rank in itself was a suboptimal predictor of maladaptive neuroendocrine stress profiles. Instead, what clearly predicted these stress profiles was the interaction between rank and the stability of ranks. When ranks were stable the low-ranked primates showed most physiological signs of stress, as might be expected. However, when ranks were unstable the highly ranked primates were the ones that showed most evidence of stress. In the current research we extend this prior work, by addressing the role of stability of interpersonal power relations in humans working together on a cooperative task.

Further evidence in support of the validity of our reasoning can be found in prior research on CV responses to the stability

of status differences between different social groups (Scheepers and Ellemers, 2005; Scheepers, 2009; see Scheepers, 2013 for an overview). In both lab-created minimal group contexts, as well as in more naturalistic groups (where status differences were based on gender) members of low status groups were found to display more CV signs of threat when status differences between groups were perceived to be stable. By contrast, members of high status groups revealed more CV signs of threat when intergroup status differences were perceived to be unstable. This prior work is about differences in social status (instead of high vs. low control due to power differences) and addresses relations between groups instead of individuals. Nevertheless, it seems relevant for the current investigation because it examines how the stability of social relations impacts on threat. Moreover, this prior work has used a methodology that makes it relevant to the present investigation, namely, the assessment of CV indices of threat following the BPS-CT. This model is explained in greater detail in the next paragraph.

The BPS-CT identifies specific CV-markers of the motivational states of challenge and threat during so-called motivated performance situations (e.g., athletic performance, doing a math test, giving a speech). According to the BPS-CT, threat and challenge result from the evaluation of a motivated performance situation in terms of its demands (e.g., effort, uncertainty, danger), as well as the person's resources (e.g., skills, knowledge, support, dispositions) to deal with these demands. When demands outweigh resources, a threat motivational state arises, whereas when resources approach or exceed demands, this induces a motivational state of challenge (Blascovich and Tomaka, 1996; Blascovich, 2008a,b; Blascovich and Mendes, 2010; Seery, 2013).

At the CV-level, challenge is marked by high cardiac output (CO, the amount of blood pumped out by the heart per minute), coupled with low total peripheral resistance (TPR, a measure of vascular resistance to blood flow), which enables the efficient mobilization and transportation of energy during motivated performance. Threat, by contrast, is marked by relatively high TPR and low CO, which leads to a less efficient mobilization and transportation of energy during motivated performance.

Two analytical strategies have been documented to interpret CV-data in the context of the BPS-CT. The first strategy is to examine *relative* differences in levels of CO and TPR *between* experimental conditions using (M)ANOVA. Relatively low levels of CO and high levels of TPR signal higher threat (and lower challenge) and vice versa. The second strategy is to examine *absolute* patterns of CV-reactivity *within* conditions, by examining increases and decreases in CV-responses compared to baseline levels. A pattern of threat CV-reactivity is indicated by significant increases in TPR in combination with unchanged CO, while a pattern of challenge CV-reactivity is indicated by significantly decreased TPR and increased CO (Blascovich and Mendes, 2010).

As indicated above, motivated performance situations constitute the context of the BPS-CT. A certain level of task engagement is required to be able to define a situation as involving motivated performance. As a check on whether task engagement is present, two additional CV-indices are commonly

examined in research using the BPS-CT: HR and *pre-ejection period* (PEP). Whereas HR refers to the *pace* with which the heart pumps, PEP, representing a measure of left-ventricular contractility, is a measure of the *force* with which the heart pumps. PEP is the most direct measure of sympathetic nervous system influence on heart activity (Brownley et al., 2000) and has additionally been described as the most direct measure of task engagement (as defined as effort; Kelsey, 2012), and the superior CV measure to index BAS activation (Brenner et al., 2005). Within the BPS-CT, engagement is indexed by significant increases in HR and decreases in PEP, compared to baseline levels.

The BPS-CT has been validated in dozens of studies, and has provided a new motivational perspective on a variety of topics, ranging from social facilitation to inter-ethnic interactions (see Blascovich and Mendes, 2010; Seery, 2013 for overviews). In the current research we examine the influence of power stability on CV threat – challenge responses, in line with the analysis provided by the BPS-CT model.

In the current study participants engaged in a cooperative, dyadic, computer-mediated task: designing and furnishing a house using a computer simulation program ("Sweet home 3D-R "). During the task, one person would be assigned the "chief designer" role (high power), and one person the "assistant" (low power). The stability of the power role was manipulated by specifying that the roles would remain the same for the duration of the task, or that the roles could possibly change after a first phase of the task. We predicted that when power was said to be stable the person with the low power role would be more threatened, and the person with the high power role would be more challenged; when power was said to be unstable the person with the high power role would be more threatened, and the person with the low power role would be more challenged.

In addition to CV-measures, we also included several self-report measures to capture some relevant outcome measures in the psychology of power. Specifically, we assessed positive and negative affect, regulatory focus (promotion and prevention focus; Higgins, 1997), action tendencies, and optimism (Anderson and Berdahl, 2002; Galinsky et al., 2003; Anderson and Galinsky, 2006).

# Materials and Methods

#### Participants and Design

Participants were 80 students (69% women; age: *M* = 21 years, range = 18–29) at Leiden University. They received €4 or course credits for their participation. Participants were randomly assigned to one of the four conditions of the 2(Power: Low vs. High) × 2(Stability: Stable vs. Unstable) design.

During the debriefing two participants displayed suspicion about the manipulations and their data were excluded from further analysis. In addition, the data of the first six participants were not included in the analyses due to a programming error in the manipulation check procedure; these participants were presented with incorrect feedback on their responses to the manipulation check items. Finally, due to signal loss or motion artifact we had missing or incomplete blood pressure data for 15 participants, and missing, incomplete or unscorable impedancecardiographic (ICG) and/or electrocardiographic (ECG) data for three participants. The participants that were included in the analyses were still evenly divided across conditions as evident from non-significant χ2-tests on the number of cases included in the analyses on the different dependent variables, χ2s *<* 1.12, *p*s *>* 0.773.

#### Cardiovascular Recording

Throughout the experimental session we continuously measured, ECG and blood pressure signals using a Biopac MP150 system (Biopac Systems Inc., Goleta, CA, USA). Physiological data was stored using *Acqknowledge* software (Biopac Systems, Goleta, CA, USA) and the ICG was scored using AMS-IMP software (Vrije Universiteit, Amsterdam, Netherlands).

For measuring ICG, the Biopac NICO100c module was used, together with four spot electrodes. Two electrodes were placed at the back of the neck (one at the base of the neck, the other ∼5 cm higher), and two spot electrodes were placed at the lower back (again ∼5 cm separated from each other). The distance between the two inner electrodes was ∼30 cm. The two outer electrodes injected a small (400 μA) alternating current while the two inner electrodes measured the voltage developed through the thorax volume. As output the NICO100c provided measures of basal impedance (Z0) and the rate of change in impedance (dZ/dt) which, in combination with the ECG, can be used to derive measures of PEP and CO. For determining CO we first calculated Stroke Volume (SV: the amount of blood pumped out on a single heartbeat), making use of the Kubicek formula (see Sherwood et al., 1990); Z0 was entered in the formula as a constant, for which we took the mean Z0 of both baseline and speech task. In turn, CO was calculated by multiplying HR and SV.

Electrocardiography was measured using an ECG100 module and two electrodes: one placed at the suprasternal notch above the top of the sternum, and one at the apex of the heart, on the left lateral margin of the chest approximately at the level of the processus xiphodius. We did not use a ground electrode as the participant was already grounded via the NICO100c. The ECG was used to determine HR and, in combination with the ICG, PEP.

Blood pressure was measured continually using a Vasotrac-R APM205a blood pressure monitor. This apparatus is equipped with a wrist sensor, which was placed over the radial artery of the participant's non-preferred hand to measure the pulse wave from the radial pulse. Every 15 s a measurement was taken. The monitor provided a measure of mean arterial pressure (MAP) which, in combination with CO, was used to calculate TPR, using the following formula: TPR = (MAP/CO) × 80.

In addition to examining CO and TPR separately, we also calculated a combined threat-challenge index (TCI) by calculating *Z*-scores of CO and TPR, then multiplying TPR with -1 and summing the result with the CO *Z*-score (Blascovich et al., 2004; Kassam et al., 2009; Seery et al., 2010). Higher scores on the resulting index—which maximizes the reliability of the CV measures (Seery et al., 2010)—indicate a greater challenge motivational state, whereas lower scores indicate a greater threat motivational state.

#### Procedure and Independent Variables

The research was conducted in conformity with the guidelines of the Ethics Committee of the Department of Psychology of Leiden University. The whole experiment was run on computers such that all information, tasks and manipulations were delivered via the computer. After arriving at the lab, the participant was seated in a cubicle, where sensors for physiological recording were applied. As part of the cover story relating to the power manipulation (see below) participants first completed a "leadership questionnaire" which consisted of a mixture of the eight items of the sense of power scale (Anderson and Galinsky, 2006; example: "If I want to, I get to make the decisions") and nine items from the multi-factor leadership questionnaire (Bass and Avolio, 1990; example: "I can inspire others"). After completing the items, 5 min of baseline CV-responses were collected during which the participant sat quietly and relaxed.

After the baseline period the participants were told that the study was about performance on a dyadic computer-mediated task. Although we led participants to believe that they would collaborate with another participant who was in an adjacent cubicle, in fact this person did not exist, and the experiment ended before participants actually performed on the task. The task consisted of first designing and then partly furnishing a house on the basis of the "Sweet home 3D-R " application. Participants were told that they would work together on the task, but that one person would take the role of "chief designer" (high power) and that the other person would take the role of "assistant" (low power). Before explaining the roles in more detail, participants were first provided with the opportunity to explore the "Sweet home 3D-R " application for some moments. The application was started, to allow participants to explore the different options and practice with the interface.

Participants were then assigned a role during the task, apparently on the basis of their scores on the leadership questionnaire (see Galinsky et al., 2003, Experiment 1 for a similar manipulation). In fact, the role assignment was made randomly. In the high power condition participants were told that they, as "chief designer," could design the house according to their wishes, and would instruct the assistant who would in turn carryout the practical steps within the "Sweet home 3D-R " program. In addition, the chief designer would determine the dimensions that should be used to evaluate the product, and would evaluate the assistant's performance. Participants in the low power condition ("assistants") were told that they would receive instructions from the chief designer and that they had to carry-out these orders to design the house according to the designer's wishes. Furthermore, participants in the low power condition were told that the chief designer would decide how the task would be evaluated, and would also evaluate the assistant's performance.

After manipulating power differences in this way, the stability of the power roles was manipulated. It was explained that the task would consist of two rounds: first designing the house, and then furnishing the house. In the stable condition participants were told that the roles (chief designer, assistant) would remain the same for the duration of both rounds of the task. In the unstable condition, participants were told that the power roles could possibly change after the first round, depending on "how the process of designing and building the house goes" (see Maner et al., 2007; Sligte et al., 2011 for similar manipulations of power stability).

After the manipulations of power and stability were induced in this way, participants delivered a short speech in front of the webcam. This represented the motivated performance situation we focused on with regard to CV-indices of challenge and threat (see Mendes et al., 2002; Weisbuch-Remington et al., 2005; Scheepers et al., 2012 for similar tasks). In the high power condition, the chief designer had to provide orders to the assistant in the speech, on how the house should be built. These instructions were said to be send to the assistant. In the low power condition the assistant could provide his or her view on the task, and indicate how they would like to see the house built in case they would have been in charge. This video was said to be recorded for "control purposes" – it would not be submitted to the chief designer. After the speech, participants completed the selfreport measures (see below) and then learned that the session had ended, after which they received a debriefing via the computer. The debriefing was concluded with the information that participants should open the door of the cubicle and call the experimenter. After the experimenter had removed the electrodes, the participant was verbally probed for suspicion and was given the opportunity to ask further questions. Finally, participants were compensated for their participation, and then dismissed.

#### Measures

Just before delivering the speech participants completed several items that were included to check the manipulations. First, participants were asked to indicate their role during the task by clicking on one of two buttons which were labeled: "chief designer" and "assistant," respectively. In addition, participants completed three items measuring how much power they had in designing the house (e.g., "How much control do you have in designing the house?" α = 0.94). Responses to these latter questions were given on seven-point scales with 1 ("very little"), and 7 ("very much") as end points.

The perceived stability of the power roles was checked by asking participants to indicate whether they could possible gain the chief designer role (in the low power condition) or whether they could possibly lose the chief designer role (in the high power condition). Participants responded by clicking on one of two buttons, which were labeled: "yes, I can gain [lose] the position of chief designer" and "No, I cannot gain [lose] the position of chief designer," respectively.

The primary dependent measure was the CV-reactivity during the speech. However, just after the speech we also administered several self-report measures. Responses to the items were recorded on seven-point scales with 1 ("very little"), and 7 ("very much") as end points. Positive and negative affect was measured by asking participants to indicate to what extent they experienced the following feelings and emotions: Happy, active, determined, positively challenged, alert, inspired, and strong (*positive affect*, α = 0.71), and threatened, scared, irritated, upset, and dejected (*negative affect*, α = 0.85). *Promotion focus* was measured using two questions (e.g., "I see the goal of building the house as well as possible primarily as an ideal," *r* = 0.29, *p* = 0.009). *Prevention focus* was also measured using two questions ("I see the goal of building the house as well as possible primarily as an obligation," *r* = 0.42, *p <* 0.001). *Optimism* was measured using three questions ("I feel certain about the success of this project"; α = 0.88). *Action-readiness* was measured using five items (e.g., "I'm prepared to take action"; α = 0.80). Finally, for control purposes we also assessed the expected *cooperation* during the task using two items (e.g., "I think that the cooperation will go smoothly," *r* = 0.60, *p <* 0.001).

# Results

All data were analyzed using 2(Power: Low vs. High) × 2(Stability: Stable vs. Unstable) ANOVAs and ANCOVAs, except where indicated otherwise. Different number of degrees of freedom across tests are due to that for different variables we had different numbers of missing or excluded cases1 .

#### Data Screening and Checks

On the dichotomous checks, only one participant indicated his/her role in the team incorrectly (power check) and only one participant indicated the stability of the roles incorrectly (stability check). The participants who gave incorrect responses were prompted with the correct response before proceeding with the experiment; therefore, the data of all participants were retained the in main analyses reported below.

Analyses of the power manipulation check scale only revealed a significant main effect for power, *F*(1,68) = 241.48, *p <* 0.001; η2 <sup>p</sup> = 0.780 (other *F*s *<* 2.13, *p*s *>* 0.149). Participants in the high power conditions experienced more power (*M* = 6.23, SD = 0.90) than participants in the low power conditions (*M* = 2.09, SD = 1.33). We conclude that the manipulations have been successful.

#### Cardiovascular Responses

In line with standard practice, mean levels of HR, PEP, CO, TPR, and TCI were calculated for the last minute of the baseline

<sup>1</sup>For eight persons all data was excluded from analyses due to technical errors (6) or suspicion (2). This means that we have data of 72 participants for the manipulation checks and self-report measures. The error terms of the ANOVAs that were conducted on these measures have 68 degrees of freedom. Regarding the baseline there was missing HR, PEP, and CO data for one participant. Regarding the speech task there was missing HR, PEP, and CO data for two persons. There was no overlap in missing data during the baseline and the speech task. Thus, a total of three participants had missing HR, PEP, and CO for either the baseline or speech task. As a consequence, the error terms of the ANCOVAs on task HR, PEP, and CO, in which baseline levels of the respective measure were added as covariates, had 72 – 3 – 5 = 64 degrees of freedom. As indicated in the main text, for the calculation of TPR, both CO and blood pressure data are necessary. Regarding the baseline there was missing TPR for seven participants, and regarding the speech task there was missing TPR for 16 participants; for six of these cases baseline TPR was also missing. As a consequence, the error term of the ANCOVA on task TPR, in which baseline TPR was added as a covariate, had 72 – 17 – 5 = 50 degrees of freedom.

period and the first minute of the speech. The resulting scores were then examined for outliers, which were defined as values 3.3 SD greater or smaller than the mean. There was one outlier on baseline CO and there were two outliers on baseline TPR; these cases were assigned a value of 1% higher than the adjacent nonextreme value (see Weisbuch-Remington et al., 2005; Van Beest and Scheepers, 2013 for a similar procedure).

#### Task Engagement

The means and standard deviations for HR and PEP during baseline and speech task period are presented in **Table 1**. Overall there were significant increases in HR from baseline to speech task (*M*baseline = 76.64; *M*speech = 86.49), *t*(68) = −8.89, *p <* 0.001, and decreases in PEP from baseline to speech task (*M*baseline = 122.32; *M*speech = 111.01), *t*(68) = 5.86, *p <* 0.001, indicating task engagement and thus the requirements are met for a further interpretation of CV-reactivity in terms of challenge and threat.

A closer inspection of task engagement in the different conditions indicated that in both high power conditions participants showed clear signs of engagement (increased HR, decreased PEP). However, while clear signs of engagement were also present in the unstable low power condition, only moderate task engagement was observed in the stable low power condition. That is, although HR increased in the stable low power condition, PEP, which represents the more direct index for engagement/BAS (Brenner et al., 2005; Kelsey, 2012) did not differ significantly from zero in the stable low power condition.

To examine between-condition differences in HR and PEP we conducted separate ANCOVAs on HR and PEP during the speech task, with baseline HR and PEP as covariate in the respective analysis. In the case of HR this did not result in significant effects of power, stability, and their interaction, *F*s *<* 2.36, *p*s *>* 0.129. In the case of PEP this resulted in a significant main effect of stability, *<sup>F</sup>*(1,64) <sup>=</sup> 6.18, *<sup>p</sup>* <sup>=</sup> 0.026; <sup>η</sup><sup>2</sup> <sup>p</sup> = 0.075, which was qualified by a marginally significant interaction between power and stability, *<sup>F</sup>*(1,64) <sup>=</sup> 3.31, *<sup>p</sup>* <sup>=</sup> 0.074; <sup>η</sup><sup>2</sup> <sup>p</sup> = 0.049. **Figure 1** displays the predicted means of PEP during the speech (controlling for baseline PEP), as a function of power and stability. Recall that lower PEP indicates more engagement. In keeping with the results of the within-condition analyses reported above, a test of the simple main effects showed that there was no difference in PEP between the stable and unstable high power condition, *<sup>F</sup>*(1,64) <sup>=</sup> 0.11, *<sup>p</sup>* <sup>=</sup> 0.739; <sup>η</sup><sup>2</sup> <sup>p</sup> = 0.002. However, participants in the stable low power condition had significantly higher PEP than participants in the unstable low power condition, *<sup>F</sup>*(1,64) <sup>=</sup> 8.49, *<sup>p</sup>* <sup>=</sup> 0.005; <sup>η</sup><sup>2</sup> <sup>p</sup> = 0.117. Moreover, when power was stable, PEP was somewhat higher in the low power condition than in the high power condition, *<sup>F</sup>*(1,64) <sup>=</sup> 2.92, *<sup>p</sup>* <sup>=</sup> 0.092; <sup>η</sup><sup>2</sup> <sup>p</sup> = 0.044. When power was unstable, there were no differences between the low and high power condition, *<sup>F</sup>*(1,64) <sup>=</sup> 0.74, *<sup>p</sup>* <sup>=</sup> 0.394; <sup>η</sup><sup>2</sup> <sup>p</sup> = 0.011. Together this indicates that participants in the low power condition *dis*engaged from the task when the position was stable. It is also noteworthy that–of all experimental conditions–participants in the unstable low power condition showed the strongest signs of engagement.


*The reported t-tests test CV-reactivity against 0 (i.e., baseline);* ∗*p < 0.05;* ∗∗*p < 0.01;* ∗∗∗*p < 0.001.*

#### Challenge and Threat

Mean levels of CO and TPR as a function of power and stability are presented in **Table 2**. To examine *relative* differences in CV markers of challenge and threat, ANCOVAs on CO, TPR, and TCI during the speech were performed; baseline levels of the respective measure were added as covariates in the models. Regarding CO this analysis yielded a marginally significant main effect of stability, *<sup>F</sup>*(1,64) <sup>=</sup> 3.11, *<sup>p</sup>* <sup>=</sup> 0.083; <sup>η</sup><sup>2</sup> <sup>p</sup> = 0.046, which was qualified by a marginally significant interaction between power and stability, *<sup>F</sup>*(1,64) <sup>=</sup> 3.31, *<sup>p</sup>* <sup>=</sup> 0.073; <sup>η</sup><sup>2</sup> <sup>p</sup> = 0.049. A test of the simple main effects showed that there was no difference in CO between the stable and unstable high power condition, *F*(1,64) = 0.01, *p* = 0.983; η<sup>2</sup> <sup>p</sup> *<* 0.001. However, participants in the unstable low power condition had significantly higher CO than participants in the stable low power condition, *<sup>F</sup>*(1,64) <sup>=</sup> 6.52, *<sup>p</sup>* <sup>=</sup> 0.013; <sup>η</sup><sup>2</sup> <sup>p</sup> = 0.092. There were no significant differences in CO between the low and high power condition when power was either stable or unstable, *F*s *<* 2.69, *p*s *>* 0.106.

The only significant effect in the analysis on TPR was an interaction among power and stability, *F*(1,50) = 5.49, *p* = 0.023; η2 <sup>p</sup> = 0.099. When power was high, TPR was higher in the unstable than in the stable condition, *F*(1,50) = 4.07, *p* = 0.049; η2 <sup>p</sup> = 0.075. There were no differences between the unstable and stable low power condition, *<sup>F</sup>*(1,50) <sup>=</sup> 1.47, *<sup>p</sup>* <sup>=</sup> 0.231; <sup>η</sup><sup>2</sup> <sup>p</sup> = 0.029. Moreover, when power was stable, TPR was somewhat higher in the low power condition than in the high power condition, *F*(1,50) = 3.66, *p* = 0.061; η<sup>2</sup> <sup>p</sup> = 0.068. When power was unstable there were no differences between the low and the high power condition, *F*(1,50) = 1.97, *p* = 0.167; η<sup>2</sup> <sup>p</sup> = 0.038.

The only significant effect on the TCI was an interaction between power and stability, *F*(1,50) = 4.12, *p* = 0.048; η2 <sup>p</sup> <sup>=</sup> 0.076. This interaction is displayed in **Figure 2**. In the low power condition there was a stronger tendency toward challenge when power was unstable than when it was stable, *F*(1,50) = 3.61,

TABLE 2 | Cardiac output (CO) and total peripheral resistance (TPR) during baseline and speech task as a function of power and stability.


*The reported t-tests test CV-reactivity against 0 (i.e., baseline);* ∗*p < 0.05;* ∗∗*p < 0.01;* ∗∗∗*p < 0.001.*

*p* = 0.063; η<sup>2</sup> <sup>p</sup> = 0.067. There were no other (marginally) significant simple main effects, *F*s *<* 2.41, *p*s *>* 0.127.

An examination of the *absolute* patterns of CV reactivity (see **Table 2**) provides some additional evidence for our main predictions. As can be seen in the table, there is only one condition where TPR significantly increases (in keeping with the threat pattern): the unstable high power condition. Despite that CO did also increase in this condition, the increased TPR corroborates with the finding that TPR was higher in the unstable high power condition than in the stable high power condition, and thus provides additional evidence for threat in the former condition. It is also noteworthy that despite significant increases in CO in the unstable low, and stable high power condition, there were no significant decreases in TPR. Thus, although there is evidence for relatively more challenge in these two conditions (see **Figure 2**), in absolute terms we cannot speak of a "full blown" challenge response. We will return to this issue in the discussion.

#### Self-Report Measures

The mean and SD for the different self-report measures (affect, promotion focus, prevention focus, action readiness, optimism, and cooperation) are presented in **Table 3**. We found main effects of power on positive affect, *<sup>F</sup>*(1,68) <sup>=</sup> 6.26, *<sup>p</sup>* <sup>=</sup> 0.015; <sup>η</sup><sup>2</sup> <sup>p</sup> = 0.084, and promotion focus, *<sup>F</sup>*(1,68) <sup>=</sup> 6.39, *<sup>p</sup>* <sup>=</sup> 0.014; <sup>η</sup><sup>2</sup> <sup>p</sup> = 0.086. Replicating earlier findings (Anderson and Berdahl, 2002; Keltner et al., 2003), participants in the high power condition scored higher on positive affect (*M* = 4.47, SD = 0.71) and promotion focus (*M* = 4.95, SD = 0.86) than participants in the low power condition (*M* = 4.02, SD = 0.80; and *M* = 4.33, SD = 1.21 respectively).

For optimism there was a significant interaction between power and stability, *<sup>F</sup>*(1,68) <sup>=</sup> 4.63, *<sup>p</sup>* <sup>=</sup> 0.035; <sup>η</sup><sup>2</sup> <sup>p</sup> = 0.064. This interaction is displayed in **Figure 3**. As can be seen in the figure, when power relations were stable, participants in the low power condition were less optimistic than participants in the high power condition, *F*(1,68) = 4.83, *p* = 0.031; η<sup>2</sup> <sup>p</sup> = 0.066; when power relations were unstable, however, these differences between those with high vs. low power did not emerge, *F*(1,68) = 0.75, *p* = 0.391; η2 <sup>p</sup> = 0.011.

TABLE 3 | Self-report measures as a function of power and stability.


*Scales run from 1 ("not at all") to 7 ("very strong").*

There were no other significant effects on the self-report measures, *F*s *<* 2.28, *p*s *>* 0.135.

# Discussion

In the current research we examined the influence of the stability of power relations on CV-markers of motivation during a dyadic task. Three results are particularly noteworthy. First, participants in the low power condition were relatively more challenged when power differences were unstable than when they were stable, which was evident from higher CO and TCI in the former than in the latter case. Second, participants in the high power condition were relatively more threatened (as evident from high TPR) when power differences were unstable than when they were stable. Both these observations provide some evidence for our hypothesis. Finally, participants in the low power condition showed CV signs of *dis*engagement from the task when their position was stable. In the following we discuss the implications of these findings for work on power, health, and team performance.

The current research shows that having power is not always a positive state that is related to approach tendencies. That is, when power holders could possibly lose their privileged position they showed a maladaptive CV pattern, indicative of threat. In the context of motivated performance, the state of threat has been conceptualized as a conflict between approach and avoidance tendencies (Blascovich, 2008a). Thus, the current work adds a physiological dimension to work showing important moderators of the power–approach relationship (Maner et al., 2007; Lammers et al., 2008). This previous work has indicated that when power differences are perceived to be illegitimate or unstable, power holders tend to display a variety of avoidance related tendencies, like becoming more risk-aversive. Future work could examine whether the currently demonstrated physiological states function as mediators between power (in)stability and these behavioral consequences.

In a more general sense, the current work shows that having power does not necessarily lead to unconstrained freedom and pleasure, but that having power can actually be quite demanding. These demands of power can stem from uncertainty about one's position, as illustrated in the current research, but also from other sources, like the meaning of power (i.e., how power is cognitively construed). Recent research shows that power holders can construe their power as an *opportunity* but also as a *responsibility* (Sassenberg et al., 2012). Power-holders who construe their power as an opportunity experience freedom and feel enabled to do what they want, while power-holders who construe their power as a responsibility experience the inner demand to do what is needed, and feel privileged and committed to act against impediments to the adherence to values and standards. There is evidence that power construed as responsibility is more demanding (and therefore often less attractive; Sassenberg et al., 2012) than power construed as responsibility. In effect, when exercising their power, power holders who were led to construe their power as a responsibility displayed a CV response profile indicative of threat, while power holders who construed their power as an opportunity displayed a CV response profile indicative of challenge (Scholl et al. under review). Thus, this work on the meaning of power fits nicely with the general conclusion of the current work, namely that it is not always "great to be the boss," but that having power can sometimes be a burden.

The importance to move beyond main effects and take into account what power *means* is also in keeping with one of the main conclusions from an extensive review of work on the relationship between rank and health in primates (Sapolsky, 2005). As this work shows, the relationship between rank and stress is complex, and rank is in itself an imperfect predictor of health-related neuro-endocrine stress responses. Sapolsky (2005) discusses the stability of ranks among the chief moderators of the rank–health relationship. He concludes that when a hierarchy is stable the low-ranked primates display the strongest neuroendocrine signs of stress but that when the hierarchy is unstable the highly ranked primates display strongest signs of stress. Although the current work addressed a different population (humans) and a different physiological process (CV responses), drawing a parallel with the analysis of Sapolsky (2005) seems to some extent justified, also given the negative health consequences of the threat CV profile (see Blascovich, 2008b for a discussion). The identification of specific subgroups that are most vulnerable for stress during social interactions, e.g., in work settings, can serve as a basis for more specifically targeted interventions.

The current results do also have implications for interpersonal behavior in teams, and how these teams in turn perform. Apart from a CV threat profile, we also found that unstable power erased the self-reported tendency for optimism that is commonly found in those with high power and one of the hallmarks of charismatic leadership and the ability to engage in the pursuit of challenging goals. These findings resonate with prior research revealing that individuals who are insecure about their position in a work group, display dysfunctional task behavior, for instance by rejecting valid contributions made by others (Rink and Ellemers, 2015). In relation to this, the CV threat profile has also been related to suboptimal decision-making, as threatened persons become rigid (De Wit et al., 2012; see also Kassam et al., 2009; Jamieson et al., 2014). In sum, role insecurity might not only have long term health implications for those in power, but may also have immediate effects on task performance, and undermine broader social relations, with decreased productivity as a likely consequence.

So far we have mainly focused on the implications of the results for (unstable) high power. However, it is also important to look at the "other side of the coin," namely the motivational processes in those with low power. The current results indicated that the prospects of change (i.e., instability) stimulate strong task engagement and benign CV arousal (challenge) in the powerless. However, when power was *stable* the powerless showed a tendency to disengage from the task as indicated by unchanged PEP. Thus, the current work shows that it is important to offer at least some prospects to those who have low power in team situations, in order to keep them positively involved in the task at hand (Ellemers et al., 2013).

At this point one may wonder about the extent to which the current results fit with the results of our earlier research where we found threat in the case of low power and challenge in the case of high power (Scheepers et al., 2012). That is, one might argue that the stable conditions in the current design are the ones that come closest to the low and high power conditions in our previous work where we did not manipulate power stability. While in our previous studies we found threat among those low in power, in the current study we found disengagement in the stable low power condition. Furthermore, while in our previous studies we found challenge among those high in power, in the current study the CV response profile for participants in the stable high power condition is more ambiguous (increased CO but stable TPR). We argue, however, that in both cases methodological differences can account for the asymmetry in results.

Regarding disengagement vs. threat in the (stable) low status condition it should be noted that in the current paradigm it was the low power person's task to simply follow the orders that were given by the power-holder, which does not seem to be very engaging, especially in the absence of ways to improve one's position. In contrast, in e.g., our previous negotiation study (Scheepers et al., 2012, Experiment 2), the low power negotiator might still have been engaged in the negotiation in order to reach the best deal possible, despite having relatively low resources and, as a consequence, showing a threat CV profile. Another factor contributing to disengagement in the current study might be that we provided explicit feedback about power stability while in our earlier research (some of) the participants in a low power position might have actually seen some prospects to improve their position, and thus remained engaged, despite still being threatened at the same time.

Regarding the absence of a strong challenge response in the (stable) high status condition it should be noted that we did not find evidence for challenge in the form of strongly decreased TPR in any of the conditions in the current design. Thus, it seems that the current paradigm has moved participants in the direction of threat, which might be explained by the demanding nature of the currently used task. For example, we purposely kept the instructions for the task somewhat vague ("design your ideal house"), to provide the power holder with some freedom to exercise his or her power. However, this might also have resulted in some task ambiguity. In addition, although participants had a few minutes to explore the "Sweet home 3D-R " program, this might have been insufficient to get a good impression of all the options. Finally, some functions, like rotating the model, were a bit delayed, due to the limited memory capacity of the computer we ran the program on. Together, these task features and aspects of the procedure may have increased task uncertainty, introducing this as a source of threat across the board, which made it more difficult for a CV challenge response to emerge.

Finally, it should also be noted that despite that the patterns of results were in line with our hypotheses, some of the effects were just marginally significant. For example, even though the interaction on the TCI was significant and the pattern of means was in keeping with our hypotheses, the specific simple main effect tests failed to reach the conventional level of significance. This might be explained by the relatively low statistical power, in particular for the tests on the CV responses, which was due to signal loss during CV recording. Despite that the current effects are not extremely strong, we remain confident in their validity not in the last place because they nicely fit with the broader literature. That is, the effects are in keeping with work on neuroendocrine responses to unstable hierarchies (Sapolsky, 2005) and also our earlier research on CV responses to stable and unstable inter-group status hierarchies (Scheepers and Ellemers, 2005; Scheepers, 2009). However, we acknowledge that the current findings are preliminary and need to be replicated in future research.

The current work has illustrated that having power does not always lead to approach and benign physiological responses, but can be rather threatening when one's powerful position can possibly change. Moreover, while low power can lead to disengagement when power is stable it can also lead to increased engagement and challenge when power is unstable. These results add (a physiological dimension) to the examination of moderators of the power-approach relationship, and can have implications to health and performance in work- and other team settings. Follow-up research should examine how CV responses as a function of (un)stable power differences mediates performance and other (behavioral) outcomes.

# References


of dissent during intragroup conflict. *J. Exp. Soc. Psychol.* 55, 221–227. doi: 10.1016/j.jesp.2014.07.011


course. *Psychophysiology* 47, 535–539. doi: 10.1111/j.1469-8986.2009. 00945.x


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2015 Scheepers, Röell and Ellemers. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

**REVIEW ARTICLE** published: 02 February 2015 doi: 10.3389/fpsyg.2015.00014

# The Trier Social Stress Test as a paradigm to study how people respond to threat in social interactions

### *Johanna U. Frisch\*, Jan A. Häusser and Andreas Mojzisch*

Institute of Psychology, University of Hildesheim, Hildesheim, Germany

#### *Edited by:*

Eva Jonas, University of Salzburg, Austria

#### *Reviewed by:*

Bernadette von Dawans, University of Freiburg, Germany Daan Scheepers, Leiden University, Netherlands

#### *\*Correspondence:*

Johanna U. Frisch, Institute of Psychology, University of Hildesheim, Universitätsplatz 1, 31141 Hildesheim, Germany

e-mail: frischjo@uni-hildesheim.de

In our lives, we face countless situations in which we are observed and evaluated by our social interaction partners. Social-evaluative threat is frequently associated with strong neurophysiological stress reactions, in particular, an increase in cortisol levels. Yet, social variables do not only cause stress, but they can also buffer the neurophysiological stress response. Furthermore, social variables can themselves be affected by the threat or the threat-induced neurophysiological stress response. In order to study this complex interplay of social-evaluative threat, social processes and neurophysiological stress responses, a paradigm is needed that (a) reliably induces high levels of social-evaluative threat and (b) is extremely adaptable to the needs of the researcher. The Trier Social Stress Test (TSST) is a well-established paradigm in biopsychology that induces social-evaluative threat in the laboratory by subjecting participants to a mock job-interview. In this review, we aim at demonstrating the potential of the TSST for studying the complex interplay of socialevaluative threat, social processes and neurophysiological stress responses.

**Keywords:Trier Social StressTest, social-evaluative threat, cortisol, social support, social cognition, social behavior**

#### **INTRODUCTION**

Stress is a complex interplay of neurophysiological, psychological, behavioral—and also social variables. In their seminal paper, Dickerson and Kemeny (2004) argue that threats to the goal of maintaining the "social self" trigger stress responses, including substantial elevations in cortisol levels. Prototypical situations in which we experience this type of threat are those that bear the danger of a negative evaluation of important and valued aspects of oneself by others (e.g., oral examinations, presentations or job interviews). Such social-evaluative threats have been found to be very potent stressors triggering strong neurophysiological stress responses (see Dickerson and Kemeny, 2004 for a meta-analysis). Responses to these threats include negative self-related cognitions, increases in cortisol, and changes in other neurophysiological variables (Dickerson et al., 2004, 2009). Interestingly, these neurophysiological stress responses are often modulated by social variables (e.g., social support) or their cognitive representations (e.g., the knowledge of belonging to a social group; Häusser et al., 2012). In reverse, social-evaluative threat and the corresponding neurophysiological responses also affect social cognition and social behavior (e.g., Merz et al., 2010; von Dawans et al., 2012).

In order to study both types of relationships in an experimental fashion, a reliable and effective paradigm to induce high levels of social-evaluative threat in the laboratory is needed. The Trier Social Stress Test (TSST; Kirschbaum et al., 1993), which is the gold standard and most commonly employed paradigm in biopsychological stress research (Kudielka et al., 2007), has only recently begun to be used in research examining the interplay between social-evaluative threat, neurophysiological stress responses, social cognition and social behavior. In this review article, we will discuss and integrate the empirical evidence from TSST-studies that have examined research questions related to this interplay.

In what follows, we will first describe the TSST and its variations. We will then summarize and integrate studies that have investigated which social variables (e.g., social support, social status) buffer the neurophysiological stress reaction in response to the TSST. Thereafter, we will turn to studies that have examined the effects of threat-related neurophysiological responses (e.g., cortisol) on social cognition (e.g., social memory) and social behavior (e.g., prosocial behavior). Finally, we will discuss methodological and conceptual issues related to the use of the TSST to study the interplay between social and neurophysiological variables in reaction to threat. Also, we will propose some avenues for future research.

#### **DESCRIPTION OF THE TSST AND TSST VARIATIONS**

In short, the TSST (Kirschbaum et al., 1993; Kudielka et al., 2007) can be described as a mock job interview. The participants are instructed to imagine having applied for their "dream job" and that they are now invited to a job interview (see **Figure 1**). The TSST consists of three successive phases: (1) A preparation period (3 min), (2) a free speech task in which the participants have to argue why they are the best candidate for the job they wish to apply for (5 min), and (3) a mental arithmetic task in which participants have to sequentially subtract an odd two-digit number from an odd four-digit number (e.g., 17 from 2023; 5 min). The two tasks are performed in front of a selection committee (two or three female and male members), dressed in white lab coats, acting in a reserved manner and providing no facial or verbal feedback. Additionally, participants are video-taped and told that their performance will be evaluated and a voice analysis will be conducted (see Kudielka

et al., 2007 for a detailed description of the standard TSST protocol).

The TSST has been found to reliably activate the hypothalamic– pituitary–adrenal (HPA) stress axis and to trigger a two- to threefold release of the stress hormone cortisol (compared to non-stress control conditions) in about 70–80% of participants (Dickerson and Kemeny, 2004; Kudielka et al., 2007). Moreover, various other indicators confirm the stress-inducing potential of the TSST: The activity of the sympathetic–adrenal–medullary (SAM) axis—the other main stress axis besides the HPA axis—can be assessed by changes in cardiovascular parameters (Kirschbaum et al., 1993) or in salivary alpha amylase (Nater and Rohleder, 2009). Additionally the TSST has also been shown to affect immunological parameters (e.g., interleukins, von Känel et al., 2006) and leads to high levels of self-reported stress and anxiety (e.g., Hellhammer and Schubert, 2012). However, since cortisol is the most prominent and most widely assessed indicator of the physiological response in TSST research (Kudielka et al., 2007; Hellhammer et al., 2009), we will focus on cortisol as an indicator of the neurophysiological stress response.

In addition to the standard protocol described above, several variations of the TSST have been developed and validated. The two most important variations concern the development of control conditions for the TSST (Het et al., 2009; Wiemers et al., 2013) as well as the development of a TSST group-version (von Dawans et al., 2011). The Placebo-TSST is a parallelized control condition for the TSST, in which participants have to talk loudly about a movie, novel or holiday trip, and have to do a simple addition task while standing in an upright position, but the social-evaluative component is missing (i.e., no committee, no video camera; Het et al., 2009). For almost all participants, the Placebo-TSST does not lead to a stress response, although it is identical to the TSST in terms of the general procedure, duration, and cognitive and physical load. It is therefore particularly useful when investigating the effects of stress (i.e., as an independent variable) on cognitive or affective outcomes while controlling for cognitive and physical load.

In addition to the Placebo-TSST, another control condition, the friendly TSST (Wiemers et al., 2013), has been developed. Similar to the Placebo-TSST, the participants experience the same cognitive and physical load as participants in the TSST, but additionally they have to perform the tasks in front of a friendly non-threatening committee. To reduce any kind of socialevaluative threat, participants are explicitly told that they are in the control condition, no video cameras are present, the committee members wear no lab coats, and behave in a friendly manner and give positive non-verbal feedback. The friendly TSST has been found to lead to no significant increase in cortisol or increase of negative affect (Wiemers et al., 2013). In its originally proposed version, the friendly TSST does not include a simple addition task like the Placebo-TSST; however, this should be added if also using the TSST in its original form.

The TSST-G (von Dawans et al., 2011) is a group version of the TSST and allows the simultaneous induction of socialevaluative threat in a group of up to six participants. In the TSST-G, the participants stand in a row facing the committee and the video cameras. Participants are separated from each other by dividing walls in order to inhibit social contact between them. The task instructions and sequence are the same as in the original TSST except that participants are asked to perform the tasks one after the other. In the corresponding control condition (Placebo-TSST-G), participants are asked to simultaneously read out a text in a low voice and to simultaneously perform a simple addition task. The TSST-G leads to comparable increases in cortisol levels and self-reported anxiety and stress as the original (individual) TSST. Importantly, in the corresponding Placebo-TSST-G no significant increases in salivary cortisol and self-reported anxiety or stress were observed (von Dawans et al., 2011).

The TSST has also been shown to effectively induce stress and trigger a cortisol response—although somewhat smaller using virtual reality systems (e.g., Kelly et al., 2007; Jönsson et al., 2010; Merz et al., 2010; Schmid and Schmid Mast, 2013). These variations differ in the presentation of the virtual reality (head-mounted display vs. projections vs. presentation on a TV screen), the used characters (avatars vs. "real" people), the possibility of verbal interaction with the committee (not possible vs. pre-recorded answers) and the number of tasks performed in front of this committee (speech task only vs. speech and mental arithmetic task). These variations bear the advantages of a highly standardized committee-behavior and some of these versions can be used under financial and spatial restrictions (e.g., in an fMRI-scanner).

Moreover, the TSST has also been adapted to different age groups. Buske-Kirschbaum et al. (1997) developed a TSST for children from 7 to 14 years. Instead of the job interview, the children were told the beginning of a story. The children are then asked to finish telling it in front of a committee which—in contrast to the original TSST—provides the children with positive verbal and non-verbal feedback. The mental arithmetic task is also adapted to the numeracy skills of children. When investigating older (retired) adults, the instruction for the job interview can be slightly changed to applying for a part-time job (e.g., child caring, housekeeping; Kudielka et al., 2007). In these two age groups, the TSST also leads to an increase in cortisol (Buske-Kirschbaum et al., 1997; Kudielka et al., 1998). However, for children, the response magnitude seems to be reduced by 30–50% compared to the cortisol response of adults (Buske-Kirschbaum et al., 1997).

#### **EMPLOYING THE TSST TO INVESTIGATE THE INTERPLAY BETWEEN SOCIAL AND NEUROPHYSIOLOGICAL VARIABLES**

Studies that have employed the TSST to examine the interplay between social and neurophysiological variables in reaction to threat can be broadly categorized into two distinct lines of investigation. The first line of investigation deals with the question of which social variables (or the cognitive representation of these social variables) buffer the neurophysiological stress reaction in response to social-evaluative threat (e.g., "Is social support effective in buffering the neuro-endocrine response triggered by the TSST?"). In these studies, the cortisol response to the TSST is the main dependent variable. The second line of investigation focuses on the reverse direction. Here, the TSST is used to induce changes in threat-related neurophysiological responses—mainly increased levels of cortisol—in order to examine their impact on social processes, such as interpersonal behavior or social cognition (e.g., "Do threat-induced elevations of cortisol levels affect social memory or prosocial behavior?"). Hence, in these studies, interpersonal behavior or social cognition are the dependent variables. In what follows, we will review both (a) studies examining the effects of social variables on neurophysiological responses, and (b) studies examining effects of threat-related neurophysiological variables on social processes.

#### **LITERATURE SEARCH AND STUDY SELECTION**

Studies were identified by searching the PsychINFO, MEDLINE, and PSYNDEX databases using the keyword "Trier Social Stress Test" (all text search). This search generated 1003 hits. The abstracts of these hits were checked as to whether the TSST or an adapted version of it was used to investigate a research question related to the interplay between social and neurophysiological variables in healthy adults. Using a snowball search system, reference lists of all identified studies were checkedfor additional studies that had not been found through the computerized search. The final sample consisted of 17 studies that examined effects of social variables on the neurophysiological stress reaction (Kirschbaum et al., 1995; Hellhammer et al., 1997; Heinrichs et al., 2003; Gruenewald et al., 2006; Ditzen et al., 2007, 2008; Mendes et al., 2007; Taylor

et al., 2007, 2010; Cosley et al., 2010; Page-Gould et al., 2010; Chen et al., 2011; Buchanan et al., 2012; Häusser et al., 2012; Schmid and Schmid Mast, 2013; Engert et al., 2014; Frisch et al., 2014) and nine studies that examined effects of threat-related neurophysiological reactions on social cognition and social behavior (Takahashi et al., 2004; Roelofs et al.,2005; Smeets et al.,2009; Merz et al.,2010; Starcke et al., 2011; von Dawans et al., 2012; Leder et al., 2013; Vinkers et al., 2013; Tomova et al., 2014).

#### **EFFECTS OF SOCIAL VARIABLES ON NEUROPHYSIOLOGICAL RESPONSES**

In this part, we will review studies that address the question of how social processes (e.g., social support) or the cognitive representation of social processes (e.g., social status) affect the neurophysiological response (e.g., the release of cortisol) to social-evaluative threat.

#### *Social support*

In our lives, we face countless situations in which we are observed and evaluated by other people and our social self is threatened (e.g., oral examinations or job interviews). Fortunately, in some of these situations we are not alone but receive support from others. There has been a wealth of research on the effectiveness of social support in stressful situations (see Uchino et al., 1996; Thorsteinsson and James, 1999 for a meta-analysis and a review). Intriguingly, although social support is often effective, there are also situations in which social support has no (Taylor et al., 2010) or even detrimental effects (e.g., Maisel and Gable, 2009). Experimental research in the laboratory using the TSST as a paradigm to induce social-evaluative threat has helped to identify some important qualifications of the effect of social support.

At first glance, it may seem counterintuitive to invite participants to the laboratory in order to study the effects of social support—something that we experience regularly in our lives and that might, hence, be studied best in those real life contexts (e.g., via questionnaires or diary studies). However, an experimental approach to study the effects of social support offers at least two major advantages: First, using an experimental approach, and especially the TSST, guarantees that all participants are confronted with the very same kind of stressor. Second, within the TSST protocol the properties of the support situation can be exactly determined by the experimental manipulation of (a) the characteristics of support recipient and provider (e.g., sex, personality), (b) the relationship between support recipient and provider (e.g., sharing of a social identity), (c) the type of support provided (e.g., verbally or non-verbally; emotional or instrumental) or (d) the availability of other hormones (i.e., oxytocin) in order to target the underlying mechanism of the beneficial effect of social support on cortisol. In what follows, we will discuss and integrate the findings from studies that have addressed these moderators of social support.

*Characteristics of support recipient and provider.* One of the first TSST studies investigating the stress-buffering effects of social support was conducted by Kirschbaum et al. (1995). In their study, male and female participants either received social support by their romantic partners, by an opposite-sex stranger (i.e., a trained confederate) or no support immediately before the TSST.

The support was provided verbally and entailed aspects of emotional and instrumental social support (e.g., focusing on positive appraisals and information on effective self-presentation). Surprisingly, social support attenuated the stress reaction only for men. Thus, men showed a lower cortisol response when supported by their partners and a tendency for attenuation when supported by strangers. Women, however, did not benefit from social support; they even showed a tendency for increased levels of cortisol when obtaining support from their partners. This pattern was not found for the psychological stress response: In all conditions participants reported a moderate level of stress. Furthermore, self-reported stress was not associated with the cortisol response. At first glance, it seems that the effectiveness of social support is moderated by the sex of the support receiver and by familiarity with the support provider. But importantly, since the sex of the support provider was not included as an experimental factor, the sex of the support provider and recipient were confounded (i.e., male participants were always supported by females and vice versa). Therefore it remains unclear if the results are due to gender differences in support reception or support provision (Kirschbaum et al., 1995). The stress-buffering effect of social support was confirmed by a study by Ditzen et al. (2008). Specifically, Ditzen et al. (2008)suggested that the attachment style of the support recipient might also play a crucial role in the effectiveness of social support. In their study, male participants either received verbal social support by their romantic partners prior to the TSST or received no support. Additionally, the attachment style of the participants was assessed via self-reports. Attachment styles can be described on the two dimensions attachment anxiety (i.e., fear of losing the partner) and attachment avoidance (i.e., striving for independence from the partner; Fraley et al., 2000). Ditzen et al. (2008) found that social support buffered the cortisol response independent of attachment style—particularly during the phase of threat anticipation. In contrast to Kirschbaum et al. (1995), a stress-buffering effect of social support was found for psychological stress, too. More specifically, this effect was moderated by attachment style, that is, social support buffered psychological stress only for securely attached men (i.e., low on the anxiety and low on the avoidance dimension). In contrast, for insecurely attached participants, social support had no effect (compared to the no support condition).

Apart from the sex of the recipient (or provider) of social support, a further moderating variable for the effectiveness of social support may be the cultural background of the support recipient and the specific type of social support (i.e., explicit vs. implicit support). Taylor et al. (2007) proposed that for Asian Americans explicit support (i.e., seeking and using emotional and instrumental support) should have detrimental effects, because they are afraid of potential negative consequences that this focus on their own needs might have for their relationship with others. Instead, they should benefit more from implicit types of social support (i.e., reminiscence of belonging to a valued social group). In order to test this hypothesis, immediately before the start of the TSST, Asian Americans and European Americans were asked to either write a letter to a friend in which they asked for his or her support in the upcoming task (explicit support), to think and write about a close group (implicit support) or

to work on an unrelated task (no support). As predicted, for Asian Americans explicit support led to a higher cortisol response and to more self-reported stress as compared to the implicit support and no support condition. Conversely, European Americans showed a higher cortisol response when faced with implicit support as compared to the explicit and no support condition, but there was no difference in self-reported stress between the groups.

It is important to note that in the studies described above social support was always provided by a third person (i.e., a confederate, friend, or partner) that was not directly involved in the stress situation. However, in many real life situations one cannot rely on one's partner or best friend providing support (you would definitely not bring them along to a job interview). As is often the case, the only potential source of social support available is the stressor him- or herself: Imagine, for example, an oral examination in school or at university. Wouldn't it be nice if you had a teacher to emotionally support you? Surprisingly, the results of a study by Taylor et al. (2010) suggest that it would not really matter; social support offered by the source of threat (i.e., the two members of the TSST committee) was not effective. In their study, Taylor et al. (2010) manipulated the behavior of the TSST-committee. Instead of being neutral and providing no feedback (as in the standard version of the TSST), the committee behaved either non-verbally supportive (e.g., by leaning forward and smiling) or non-verbally unsupportive (e.g., by showing signs of boredom like frowning or sighing). Strikingly, compared to a control condition without any committee, participants in both committee-present-conditions showed an equally strong cortisol response. In other words, a supportive TSST committee did not attenuate the cortisol stress reaction. Cosley et al. (2010) suggested that whether or not support by the stressor is supportive is moderated by inter-individual differences in the trait "compassion for others." People with high levels of compassion should perceive the support provider as being more compassionate, interpreting the offered support in a more positive way. Similar to Cosley et al. (2010), Taylor et al. (2010) manipulated the behavior of the TSST-committee. Half of the participants faced an emotionally supportive committee whereas the other half faced the neutral standard TSST-committee. Consistent with their expectations, in the social support condition participants with high levels of compassion for others showed lower cortisol responses than those with low levels of compassion. However, in the neutral committee behavior condition, compassion was not associated with the cortisol response.

Taken together, the five studies reviewed above have identified important moderators of the effect of social support on the neurophysiological response to social-evaluative threat, such as sex (Kirschbaum et al., 1995), attachment style (Ditzen et al., 2008), cultural background (Taylor et al., 2007) and personality (Cosley et al., 2010) of the support recipient. Furthermore, characteristics of the support provider also influence the effectiveness of social support: Support by the stressor itself is not per se effective (Taylor et al., 2010) and, at least for men, support from their romantic partner tended to be more effective than support by a stranger (Kirschbaum et al., 1995). This last finding suggests that the relationship between provider and recipient of social support

might be an additional important moderator that might also play a role in explaining the finding of Taylor et al. (2010).

*Relationship between provider and recipient.* In line with this idea, Frisch et al. (2014) proposed that the results of Taylor et al. (2010) can be explained when taking the relationship between support recipient and provider into account. Building on the social identity approach (Haslam, 2004), Frisch et al. (2014) extended Taylor et al.'s (2010) original study design by including a manipulation of social versus personal identity as an additional experimental factor. In order to do so, the TSST committee consisted of two confederates who pretended to be real participants and who were designated as the TSST committee by a faked drawing of lots procedure. Prior to the TSST, the salience of either a shared social identity (i.e., a feeling of "we"-ness) between the participant and the TSST committee or a personal identity was manipulated: To make a social (vs. personal) identity salient, participants had to wear same (vs. different) colored T-shirts, were asked to think of similarities (vs. differences) and worked alone on an idea generation task in which group (vs. individual) performance was analyzed. In the following TSST, similar to Taylor et al. (2010), half of the participants faced an emotionally supportive committee whereas the other half was confronted with an unsupportive committee. As hypothesized, a stress-buffering effect of social support (i.e., a decreased cortisol reaction) was only found in the social identity condition. In the personal identity condition, the results resembled the findings of Taylor et al. (2010), that is, participants showed the same cortisol reaction regardless of being supported or not. Interestingly, for self-reported stress, no stressbuffering effect of social support in the social identity condition was found. Taken together, the study by Frisch et al. (2014) suggests that social support buffers the cortisol stress reaction only if a shared social identity between the provider and recipient of support has been established.

Whereas Frisch et al. (2014)showed that a shared social identity is an important moderator of the effectiveness of social support, Häusser et al. (2012) found that a shared social identity per se can be an effective stress-buffer. In their study, participants underwent either the TSST-G (von Dawans et al., 2011) or the Placebo-TSST-G in groups of four. Beforehand, for half of the participants a shared social identity with their fellow group members was made salient and for the other half a personal identity was activated. As expected, a shared social identity worked as a stressbuffer. Thus, participants in the shared social identity condition showed a significantly reduced cortisol reaction in response to the TSST-G. Importantly, since participants were not allowed to interact with each other during the whole study, no overt transmission of support was possible. In other words, the mere cognitive representation of belonging to the same social group (i.e., "we are going through this together"), buffered the cortisol reaction in response to social-evaluative threat. Again, this stress-buffering effect was not found for self-reported stress. This study provides a nice example of how even a cognitive representation of social processes can be effective in coping with social-evaluative threat.

In sum, especially the study by Frisch et al. (2014) highlights that when facing social-evaluative threat a shared social identity is an important precondition to benefit from support. A shared social identity may provide group members with a common interpretive framework (things are perceived and evaluated in a similar fashion by group members) and may increase feelings of trust. These processes may facilitate the interpretation of the offered social support as wholehearted and in the spirit it was intended thereby making the provision of support more effective (Haslam et al., 2012; van Dick and Haslam, 2012).

However, two limitations of this research have to be put forward: First, in both studies the predicted stress buffering effects were only found for the neuroendocrine stress reaction but not for self-reported stress (see also Kirschbaum et al., 1995; Ditzen et al., 2007 for similar findings). Second, the specific mechanisms mediating the effects of social identification on the neuroendocrine stress reaction are still far from clear.

*Type of support.* It is not only the characteristics of support recipient and provider and their relationship that should be taken into account, but also the type of support. For example, as already described above, in the study by Taylor et al. (2007), the effects of the cultural background of the recipients of social support were moderated by the type of support (explicit vs. implicit support). In another study by Ditzen et al. (2007), female participants received either (a) verbal support or (b) a shoulder massage from their romantic partner or (c) received no support prior to the TSST. In line with Kirschbaum et al. (1995), women who received verbal partner support did not profit from it and showed a similar increase in cortisol as compared to the no support control condition. However, since verbal support was provided exclusively from male partners it remains unclear whether this finding results from an ineffective support provision by males, or ineffective support reception by females. In contrast, women who received a massage had an attenuated cortisol reaction. For self-reported stress and anxiety, no differences between the three conditions were found. Hence, although a stress-buffering effect on the physiological stress response was found, again, this was not found on the subjective-psychological level.

*Availability of oxytocin.* Recently, the activity of the hormone oxytocin has been discussed as one underlying biological mechanism of the stress-buffering effect of social support (see Campbell, 2010; Hostinar et al., 2014 for reviews). Predominantly in studies with animals, but also in some human studies, it has been shown that oxytocin is released in positive social contexts and that it has dampening effects on the activity of the HPA axis (Hostinar et al., 2014). However the direct mediation of the stress-buffering effect of social support in situations of social-evaluative threat has not been demonstrated so far, which may be partly due to problems with measuring peripheral oxytocin (McCullough et al., 2013).

Heinrichs et al. (2003) investigated the stress-buffering effects of both verbal support and oxytocin for male participants. Specifically, in addition to the manipulation of social support (support by best friend vs. no support), they also administered intranasal oxytocin to half of the participants whereas the other half received a placebo about an hour prior to the TSST. The results showed that social support as well as oxytocin suppressed cortisol responses, with participants receiving both treatments having the lowest cortisol response. These findings suggest that oxytocin is involved

in the down-regulation of the HPA-axis of humans and that it enhances the beneficial effect of social support. However, the study design employed by Heinrichs et al. (2003) does not allow testing of whether the effect of social support was mediated by oxytocin secretion (i.e., social support increases oxytocin levels which, in turn, buffer the cortisol reaction). More evidence comes from a study of Chen et al. (2011). They investigated whether variations in the receptor gene of oxytocin are associated with the stress-buffering effect of social support. One special singlenucleotide polymorphism in this gene (rs53576) has been found to be related to reduced social abilities (e.g., Bakermans-Kranenburg and van Ijzendoorn, 2008; Tost et al., 2010) and less searching for social contact (Kim et al., 2010). Chen et al. (2011) hypothesized that participants carrying this special allele variation might also profit less from social support. In their study, male participants either received social support from a female friend or no support prior to the TSST-G. As predicted, rs53576 G carriers seemed to benefit more from social support than individuals with the AA genotype, who showed almost identical subjective and cortisol stress reactions in both the support and no-support conditions. These results indicate that genetic variations of the oxytocin system modulate the effectiveness of social support as a buffer against social-evaluative threat. Again, however, these results provide no direct evidence for the idea that the stress-buffering effects of social support are mediated by oxytocin. Rather, the findings point to an interactive effect of social support and oxytocin. Since oxytocin has been found to increase trust (Kosfeld et al., 2005), it is tempting to speculate that the interactive effects of oxytocin and social support are due to the fact that oxytocin increases the probability that the recipient of support trusts more in the wholeheartedness of the provided support, thereby making the provision of support more effective. Future research is needed to test this hypothesis.

#### *Observation of threat*

In almost all of the studies employing the TSST to examine how social variables influence neurophysiological responses, the focus is on the individual being threatened and his or her reaction to the TSST. By contrast, two recent studies shifted this focus toward the neurophysiological reactions of the persons being the stressor (i.e., the committee members; Buchanan et al., 2012) or to persons who observed the TSST participant (Engert et al., 2014). Particularly, both studies investigated the relationship between the neurophysiological responses of the stressor/observer and the neurophysiological responses of the participants being exposed to the TSST. In other words, these studies aimed at exploring whether the response of the stressor/observer resonates with that of the participant. Physiological resonance means that the stress response of the stressor/observer is a function of the stress response of the TSST participant (Engert et al., 2014). In the study of Buchanan et al. (2012), the cortisol responses of the TSST-participants indeed were predictive for the cortisol response of the TSST-committee members (i.e., trained research assistants). Moreover, cortisol responses were generally higher for those committee members with higher levels of trait empathy, indicating that empathy plays a crucial role in physiological resonance. However, the direction of the effect is not entirely clear. It is easy to imagine that the TSST can also be a stressful experience for the committee members (Engert et al., 2014). For example, being responsible for the distress of another person, or the demand to suppress spontaneous supportive behavior (such as smiling or nodding) could lead to stress in the TSST-committee. Therefore, it seems also possible that a stressed committee causes more stress in the participant accounting for the relationship between the cortisol responses. In order to investigate if a real empathic stress response can be elicited by the TSST, Engert et al. (2014) used a different approach. A passive observer (either a stranger or a romantic partner) witnessed the participant undergoing the TSST either through a one-way mirror or via video. Hence, in contrast to Buchanan et al. (2012), the participant could not see the observer and could therefore not be influenced by his or her stress response. Twenty-six percent of all observers showed a significant increase in cortisol levels, with the strongest cortisol responses found in observers watching their own partners through a one-way mirror. Furthermore, Engert et al. (2014) also demonstrated resonance since the cortisol stress response of the observers was to some degree predicted by the cortisol stress response of the participants. The finding that the neuroendocrine responses of the observer/stressor and the participant resonate is intriguing. Although emotional contagion—the "catching" up of the emotion of the interaction partner (Hatfield et al., 1993)—has been shown at a behavioral level (e.g., facial mimicry, Mojzisch et al., 2006), or at a cardiovascular level (Konvalinka et al., 2011), these two studies are the first to show this resonance of stress on a neuroendocrine level. However, one limitation of both studies is that they do not address the underlying mechanism of this effect. Both studies highlight the importance of empathy, but it remains completely unclear what cues (e.g., facial expression, voice, and other behaviors) of the threatened participants trigger the HPA axis activity in the observer. Furthermore, since no selfreports of the stressor/observers were obtained we do not know which feelings accompany this HPA activation. As Dickerson et al. (2004) suggested and have shown empirically (Gruenewald et al., 2004; Dickerson et al., 2008), the threat of one's social self is associated with self-conscious emotions and cognitions (i.e., shame or embarrassment). Hence, it would be interesting to examine whether the same feelings are triggered in the observer.

Furthermore, at this point, we can only speculate about the implications of this resonance, but it might enhance the understanding for the situation or for the needs of the threatened person. Moreover, it might enhance the provision of social support or might also have beneficial effects for the long-term relationship between the threatened person and the observer. Future research should address these implications as well as the psycho-physiological pathways of transmission of this resonance.

#### *Social status and power*

Social self-preservation theory (Dickerson et al., 2004, 2009) argues that social-evaluative threats trigger a coordinated psychophysiological and behavioral response in order to prevent negative effects, like loss of status or social exclusion. Gruenewald et al. (2006) hypothesized that this relationship is moderated by the social status of the individual. Individuals with low status should react stronger to additional threats of their already low status, as compared to high status individuals. To test this hypothesis, Gruenewald et al. (2006) assessed the subjective selfreported social status from college students living in a residential dormitory before letting them undergo either the TSST or the Placebo-TSST. In stark contrast to their expectations, they found that only students high in social status exhibited the typical cortisol reaction whereas students low in status showed a blunted cortisol reaction. These findings are, however, in line with a study by Hellhammer et al. (1997) who observed male army recruits during boot camp training over several weeks and confronted them with the TSST. The recruits were divided into small groups, and the social status of each of the recruits was assessed somewhat more objectively than in the study of Gruenewald et al. (2006) by asking every recruit to indicate how he perceived his fellow recruits. In reaction to the TSST, the recruits with a higher status showed the strongest cortisol reaction whereas low status recruits only showed a weak response.

Albeit speculatively, these findings could be explained by lower ego-involvement of participants low in status. Low status participants might be in general less concerned with evaluation situations—which could also be seen as one reason for their low status (Hellhammer et al., 1997; Gruenewald et al., 2006). In contrast, for individuals with a high status there might have been more at stake and therefore their fear of a potential status loss led to an increase in cortisol (Hellhammer et al., 1997).

Note that these two studies (Hellhammer et al., 1997; Gruenewald et al., 2006) investigated the influence of the status of individuals in already existing groups; hence a quasi-experimental design was used. Since quasi-experimental designs are prone to the influence of confounding variables, they should be supplemented with more controlled experimental studies. Such a study was conducted by Schmid and Schmid Mast (2013) who actually found the exact opposite pattern of results as the two previous studies (i.e., their results were in support of the original proposition of Gruenewald et al., 2006). In two experiments, using adapted TSST versions, the experimental priming of high power in social situations compared to low power resulted in a weaker increase in heart rate (unfortunately cortisol was not assessed). Moreover, participants in the high power condition reported less fear of evaluation, showed less non-verbal signs of nervousness and performed better in the speech task, which was rated by different raters based on the recorded videos.

In sum, these three studies convey an inconclusive picture on the effects of social status on the reaction to threat. Although all three studies confirmed that experience of previous social interactions (e.g., the emergence of status in a social group) affects the stress response and even the performance, the direction of this effect remains unclear for the present. The picture might be even more complex when acknowledging that power and status can be thought of as slightly different concepts, for example, a relatively high status does not necessarily go along with high levels of power in social situations (e.g., the recruits high in status in the study of Hellhammer et al., 1997 were also dependent on their supervisors). Further research is therefore needed to specify the conditions under which high status/power has a protective effect.

Interestingly, much of the current research on the influence of rank on reactions to stress is based on animals, but here results are often inconsistent (see Sapolsky, 2005 for a review). Sapolsky (2005) suggests that there are several moderators (i.e., stability or personality) that determine whether primates of low or high status experience more stress: For example, when hierarchies are rather stable then individuals with low ranks experience more stress than those with high ranks, whereas in the case of unstable hierarchies the relation is inversed. In the context of inter-group competition, a corresponding pattern has been found for humans (Scheepers and Ellemers, 2005; Scheepers, 2009). Scheepers (2009)found that when hierarchies were stable, members of low status groups, if confronted with an inter-group competition, exhibited a more pronounced cardiovascular threat pattern (e.g., Blascovich and Mendes, 2000) than members of high status groups. By contrast, when hierarchies were unstable, members of low status groups showed a challenge pattern, whereas members of high status groups displayed a threat pattern. Since the underlying motivation of maintaining a positive (social) self is very similar in situations of inter-group threat and in those of social-evaluative threat, stability of hierarchies might also be an important moderator in these latter situations.

#### *Racial bias/intergroup threat*

A mounting body of evidence shows that interacting with people of different races can produce threat and stress reactions (e.g., Mendes et al., 2002). However, the neuroendocrine reactions to intergroup threat are likely to be shaped by the individuals' racial bias. To test this idea, Mendes et al. (2007) examined the influence of implicit racial attitudes of White participants on their neurophysiological reaction toward a TSST in which the committee members were either part of an in-group (i.e., White) or of an out-group (i.e., Black). The implicit racial attitudes were assessed with the Implicit Association Test (Nosek et al., 2005). The results revealed that the cortisol response did not depend on the group membership of the committee members or the racial-biases of the participants. However, more egalitarian attitudes of the participants were associated with a more salutary stress response (as defined by the ratio of the hormone dehydroepiandrosterone to cortisol): For participants facing Black committee members, a low racial bias was associated with a more salutary stress response, a lower report of threat appraisals, and less signs of anxiety than a high racial bias. Page-Gould et al. (2010) replicated and extended this study by including both White and Black participants. Thus, Black and White participants underwent the TSST facing either a Black or White committee. The results of this study confirmed that the cortisol reaction in response to the TSST did not depend on whether the committee members were part of the in-group or the out-group. However, the amount of prior intergroup contact—which was assessed beforehand—was positively related to the physiological recovery in both intergroup conditions (i.e., Black participants and White committee members or White participants and Black committee members). Thus, participants reporting more prior intergroup contact had a steeper decline in cortisol levels following an intergroup stressor than participants with only few prior intergroup contacts. Interestingly, the race of the participants did not moderate these results.

Taken together, these two studies highlight that not only specific behaviors of the interaction partner in situations of social-evaluative threat can affect neurophysiological responses, but also stereotypes and intergroup contact. Both studies, however, did not find effects on the immediate cortisol reaction, but rather on indicators of recovery. This is an important finding suggesting that social variables may not only exert influences on the immediate cortisol reaction but also on the rate of recovery from stressful events. Since especially the failure to recover from such events may have several negative health consequences (e.g., McEwen, 1998), it seems worthwhile to focus not only on peak neuroendocrine responses but also to analyze rates of recovery (see Linden et al., 1997 for a review).

#### **EFFECTS OF NEUROPHYSIOLOGICAL RESPONSES TO THREAT ON SOCIAL PROCESSES**

All of the studies reviewed so far used the TSST as a tool to test whether the manipulation of specific social variables (e.g., social support) buffers the neurophysiological stress response. In these studies, the cortisol response to the TSST is the main dependent variable. By contrast, a different line of research examines the social-cognitive effects of stress. In these studies, stress is the independent variable (i.e., TSST vs. Placebo-TSST) and the main dependent variables are participants' social cognitions or behaviors in response to this manipulation. For example, this line of research has tested whether acute stress affects prosocial behavior. In the next section, we will focus on this second line of research.

#### *Social cognition/social memory*

Social cognition, defined as "the mental operations that underlie social interactions and includes the ability to attribute mental states (e.g., emotions, thoughts, intentions) to oneself and others" (Smeets et al., 2009, 507), is one essential prerequisite for successful interactions. Smeets et al. (2009) investigated the influence of social-evaluative threat on the ability to infer the non-emotional and emotional states of other individuals. After the TSST/Placebo-TSST, participants were asked to indicate the emotional and non-emotional states of different characters in a short movie. The results showed an effect of threat on the ability to infer the states of other individuals which, however, was moderated by sex as well as by the magnitude of the cortisol response: When exposed to social-evaluative threat, male high-cortisol responders were better at identifying emotional, and non-emotional states than male low-cortisol responders, however, they were not better than the non-stressed control group. By contrast, female participants showed the opposite pattern, that is, when exposed to social-evaluative threat, female low-cortisol responders performed better than female high-cortisol responders and the non-stressed control group. However, these results are somewhat inconsistent with the results of a later study by Tomova et al. (2014). In this study, the ability to distinguish between the self and the other another import prerequisite for empathy and mentalizing—was assessed in three tasks (e.g., a perspective taking task in which participants had to arrange objects according to instructions of another person with a different visual perspective). For women, the ability to distinguish between the self and the other was increased under conditions of social-evaluative threat (TSST-G vs. Placebo-TSST-G), whereas it was decreased for men. Moreover, in this study, cortisol was not correlated with the ability of self-other-distinction.

From a more theoretical point of view, it is plausible that abilities such as emotion recognition should be enhanced during threat—particularly in women: The tend-and-befriend model of Taylor et al. (2000) posits that women do not respond to stress with fight-or-flight (cf. Cannon, 1932) as men do, but show a more affiliative stress response. This involves nurturing behavior in order to protect the offspring (tending) as well as activities to create and maintain the social network (befriending). Improved emotion recognition and increased empathy are beneficial in forming these social bonds (Tomova et al., 2014). Future research is needed to confirm and to disentangle the diverging finding of these two studies.

In another study, Leder et al. (2013) investigated the effects of social-evaluative threat on the ability of strategizing in a decision making context. The ability of strategizing—that is, thinking about what other actors might think and do—is important for many decision situations, especially in economic decision-making where asset prices are less affected by the fundamental value of the asset but more by what people think everyone else thinks the asset value is. This ability was assessed by using the Beauty Contest game: Following previous research (e.g., Nagel, 1995), four participants were asked to choose a number between 0 and 100. They were told that the participant whose number is closest to the average of all chosen numbers multiplied by 2/3 will be the winner of this game. Hence, in order to win this game, participants have to anticipate the answers of the other participants. For example, participants who show no signs of strategic reasoning would pick a random number. However, participants with a higher level of reasoning would pick numbers around 33, since they assume the other participants would have chosen random numbers (which would result in a mean around 50 that then has to be to be multiplied by 2/3). Furthermore, if a participant expects that all other participants will figure this out, then he or she would choose a number close to 22.22. With increasing iterations, the number converges toward zero (i.e., the Nash equilibrium). Leder et al. (2013) found that threatened male participants (TSST-G) chose higher numbers in the beauty contest game than non-threatened individuals (Placebo-TSST-G), indicating lower levels of strategic reasoning. The relationship between social-evaluative threat and strategic decision making was mediated by the threat-induced increase in cortisol. Additional analyses revealed that it took stressed individuals longer to learn and understand the strategic nature of the game compared to participants in the control group. This is in line with previous findings on the effects of stress on impaired feedback processing (Starcke and Brand, 2012), but might also be due to impaired mentalizing, that is, the ability to anticipate what interaction partners might think or do (Tomova et al., 2014).

Apart from having the ability to tune into others or to anticipate their behavior, it is important for social interactions to encode, memorize and update important information about the interaction partner. Stress and cortisol have been found to negatively affect the retrieval of declarative memory contents (see Wolf, 2009 for a review). Two studies show that this also holds true for the memory retrieval of social information. For example, Takahashi et al. (2004) found an impaired social memory for face-name associations under conditions of social-evaluative threat in men, compared to a non-stressed participant. Moreover, cortisol was negatively associated with performance in the memory task. However, since both the encoding and retrieval took place after the TSST, it remains unclear which process was affected by threat. To disentangle the potential effects on encoding and retrieval, Merz et al. (2010) conducted a study in which male and female participants had to learn biographical information about two persons (e.g., gender, hometown, birth date) before the TSST and were asked to recall them after being exposed to the TSST. Compared to their own performance in a control session without social-evaluative threat, participants made more mistakes in the retrieval of the biographical material after being threatened. Furthermore, and in line with Takahashi et al. (2004), cortisol levels were negatively associated with recall performance. Interestingly, social-evaluative threat and cortisol have been found to also affect declarative memory—for non-social contents—positively (e.g., Kuhlmann and Wolf, 2006). Whether memory is positively or negatively affected seems to be dependent on the memory processes involved; whereas cortisol has a negative influence on memory retrieval, it has a positive effect on memory consolidation (Wolf, 2009). Future research should aim at examining whether these positive effects of threat and cortisol can also be found for the consolidation of social memory contents.

In sum, these findings suggest that social-evaluative threat has mostly negative effects on emotion recognition in others, selfother distinction, strategizing, and memory retrieval for social information. These findings are remarkable insofar as emotion recognition or effective updating of social information are valuable resources when coping with social-evaluative threat. Hence, social-evaluative threat negatively affects the very abilities that are needed to cope with threat. In other words, an ancient biological stress response interferes with the social-psychological demands of typical modern threats.

#### *Approach/avoidance behavior*

Roelofs et al. (2005) sought to investigate how social-evaluative threat influences approach and avoidance behavior to social stimuli. To this end, they used a computerized approach-avoidance task, in which participants saw either happy or angry faces and either had to push a button requiring arm flexion (i.e., approach behavior) or had to push a button requiring arm extension (i.e., avoidance behavior) in response to these stimuli. Employing a within-subjects-design, male and female participants were tested both before and after the TSST. Before the TSST, participants showed the well-established congruency effect—a faster reaction in trials in which movement and stimulus were congruent (i.e., angry face and arm extension, happy face and arm flexion) than in incongruent trials (Solarz, 1960). However, after the TSST, participants who had a high cortisol response no longer showed the congruency effect, that is, the reaction times in the congruent trials became slower and similar to those in the incongruent trials. This effect was not evident in low cortisol responders. Albeit speculatively, increased reaction times can be considered as a freezing reaction, similar to what has been found in animal studies, where neither a preference for avoiding nor approach behavior exists anymore (Roelofs et al., 2005). Hence, similar to the effects on social cognition and social memory retrieval, stress has a dysfunctional effect on behavior actually needed to effectively cope with the stressful situation. Although the classic perception of the human stress reaction is that it is functional in helping the organism to overcome the stressful event (e.g., by supplying it with additional energy), this may not necessarily be the case when it comes to social-interactive coping resources.

#### *Prosocial behavior*

Given that social support and social identification in groups are likely to buffer the neuroendocrine stress reaction in response to social-evaluative threat (e.g., Häusser et al., 2012), engaging in prosocial behavior (e.g., providing help and support to each other) might be a functional response to social-evaluative threat. Building on this notion, von Dawans et al. (2012) examined the effect of social-evaluative threat on subsequent prosocial and antisocial behavior. They found that after being exposed to the TSST-G,male participants were more prosocial in economic games. Compared to the non-stressed control group (Placebo-TSST-G), they trusted their partners more, were themselves perceived more trustworthy and shared more money with others. Importantly, the results also suggest that prosocial behavior following social-evaluative threat is not due to an unspecific increase in the readiness to bear risks. Thus, social-evaluative threat specifically affected the willingness to accept risks arising through social interactions, whereas nonsocial risk taking was not affected. Also, the stress manipulation had no influence on negative social interactions (i.e., punishment behavior). Taken together, this study provides first experimental evidence suggesting that men—and not only women—engage in prosocial tending-and-befriending behavior in response to stress (Taylor et al., 2000).

Similarly,Vinkers et al. (2013) investigated the effects of socialevaluative threat on reactions to unfair offers in an ultimatum game. In an ultimatum game, one player is given a sum of money that he or she can allocate between herself and another player. The recipient has the option of accepting or rejecting this offer. If the offer is accepted, the sum is divided as proposed. If it is rejected, neither player receives anything. In addition, Vinkers et al. (2013) used a one-shot variant of the Dictator Game in which participants received 10€ with the possibility to donate any amount to "Unicef" and keep the remaining amount to themselves. Interestingly, Vinkers et al. (2013) also varied the timing of the social decision making tasks, where half of the male participants worked on the tasks immediately after the TSST-G/Placebo-TSST-G (in order to examine rapid non-genomic effects of cortisol) and the other half worked on the tasks 75 min after the cessation of the TSST (in order to examine slow genomic effects of cortisol; see Joels and Baram, 2009 for a description of the different phases of the stress response). They found that the effects of threat on behavior in the ultimatum game were time-dependent. In the direct aftermath of the threat—and in line with the findings of von Dawans et al. (2012)—no effects of threat were found. By contrast, in the delayed condition, threatened participants rejected fewer unfair offers (i.e., less altruistic punishment). In the Dictator Game, a time-independent negative effect of stress on prosocial behavior was found: Stressed participants donated less money to a charitable organization.

A different type of prosocial behavior was investigated in a vignette study by Starcke et al. (2011). This study examined whether social-evaluative threat affects prosocial behavior in everyday moral decision making. Participants had to decide on everyday moral dilemmas, each offering a more egoistic and a more altruistic decision alternative (i.e., "You find a 20\$ note on the pavement. Then you see a homeless man looking for food in the dustbin. Would you give him the money?"; Starcke et al., 2011, 217). The results revealed no significant differences between the stressed TSST and the not-stressed control group. There was, however, a significant negative correlation between cortisol and the morality of the decisions, indicating that participants (male and females) with a stronger cortisol reaction made more egoistic decisions in the dilemmas.

Taken together, the three TSST-studies described above suggest that social-evaluative threat can influence prosocial behavior in several different ways. The direction of this effect seems to depend on specific conditions. Thus, social-evaluative threat has been found to lead to more prosocial decisions when the decision is targeted toward single individuals (von Dawans et al., 2012), but not when it is targeted toward a charitable organization (Vinkers et al., 2013) or involves just predicting one's own behavior in hypothetical situations (Starcke et al., 2011). Notwithstanding the importance of these studies, it has to be noted that they all examined prosocial behavior in a rather constrained fashion, that is, by using vignettes (Starcke et al., 2011) or decision paradigms from behavioral economics (von Dawans et al., 2012; Vinkers et al., 2013). By contrast, numerous studies from social psychology have examined spontaneous prosocial behavior in a more unconstrained setting. For example, in a prototypical study, the experimenter accidentally spills some pencils on the floor, and the dependent variable is whether participants help him or her to pick them up (e.g., Greitemeyer and Osswald, 2010). Hence, it would be an interesting avenue for future research to study the impact of social-evaluative threat on spontaneous prosocial behavior using paradigms from social psychology. Also, this research might examine whether the effects of social-evaluative threat on prosocial behavior are mediated by the accessibility of prosocial thoughts.

#### **DISCUSSION**

The present review aimed at evaluating the potential of the TSST to study the interplay between social and neurophysiological factors during threatening social interactions. To this end, we reviewed research using the TSST to examine either the effects of social factors on threat-related neurophysiological stress responses, or the effects of social-evaluative threat and neurophysiological stress responses on social processes.

#### **SOCIAL-EVALUATIVE THREAT—INSIGHTS FROM RESEARCH EMPLOYING THE TSST**

The most robust finding in all of the reviewed studies is that socialevaluative threat leads to a strong neuroendocrine stress reaction as indicated by elevated levels of cortisol and high levels of selfreported stress. This is remarkable since in all of these studies the participation in the TSST, of course, had no real life consequences for the participants. Thus, none of the participants would have been excluded from a social group or would have lost his or her job if he or she failed in the TSST. Although no direct real life consequences emerge, it might be argued that at least to some extent the experienced stress in the TSST may stem from the anticipation of real life consequences in the future. Thus, participants may fear a similarly poor performance in a comparable real life situation (e.g., a job interview).

Luckily, however, studies employing the TSST have also examined which social factors can buffer neuroendocrine stress reactions. Most prominently, several studies have tested whether receiving social support works as a stress buffer. Somewhat counterintuitively, social support per se has been found to be often insufficient in buffering the neuroendocrine stress reaction, and the studies reviewed in this article have identified some important moderators like, for example, the relationship between support provider and recipient (e.g., Frisch et al., 2014), the type of support offered (e.g., Ditzen et al., 2007) or the sex of the support recipient (e.g., Kirschbaum et al., 1995). Apart from direct social support also other—more subtle—social processes, like social status (e.g., Gruenewald et al., 2006) or racial biases (e.g., Mendes et al., 2007) have been found to affect the neuroendocrine response or recovery. Interestingly, there is also preliminary evidence for contagion effects regarding the neuroendocrine stress response, that is, the participants' neuroendocrine responses to stress resonate with the TSST committee members or with passive observers (Buchanan et al., 2012; Engert et al., 2014).

Social variables and processes do not only influence the neurophysiological reaction to threat, but social-evaluative threat and the corresponding neurophysiological responses can also backfire on social factors. Ironically, social-evaluative threat seems to have negative effects on the very social abilities that are particularly helpful to cope with that threat, like the recognition of emotions of others (Smeets et al., 2009; Tomova et al., 2014) or the anticipation of their behavior in strategic interactions (Leder et al., 2013). On the other hand, exposure to social-evaluative threat can also lead to functional protective responses, such as an increase in prosocial behavior (von Dawans et al., 2012).

Although the studies reviewed above substantially advance our understanding of social-evaluative threat, they also raise some conceptual issues that should be addressed by future research. In particular, these conceptual issues comprise (a) the dissociation of neurophysiological and psychological stress responses as well as (b) the specific mechanisms underlying the relationship between social variables and threat-related neurophysiological processes.

#### *Dissociation of neurophysiological and psychological stress responses*

The TSST reliably triggers increases in both neurophysiological and psychological stress indicators (e.g., Kudielka et al., 2007). However, especially in studies examining the stress attenuating influence of social support on stress reactions, a dissociation of both response levels has been found. In most of these studies (Kirschbaum et al., 1995; Ditzen et al., 2007; Taylor et al., 2010; Häusser et al., 2012; Frisch et al., 2014) social support effectively buffered the neurophysiological stress response but did not affect the self-reported stress levels—which always remained high. In line with this, TSST studies often do not find significant correlations

between both indicators of stress. Eight studies reviewed in this paper directly tested the association between the neurophysiological and the psychological stress response. Only two of them (Gruenewald et al., 2006; Ditzen et al., 2008) report a significant correlation whereas in the other six studies cortisol was not related to self-reported stress (Kirschbaum et al., 1995; Roelofs et al., 2005; Ditzen et al., 2007; von Dawans et al., 2012; Frisch et al., 2014; Tomova et al., 2014). This absence of a direct relationship is confirmed in a review article by Campbell and Ehlert (2012) who found that only 27% of studies employing the TSST or similar stress induction paradigms have found a significant correlation between cortisol levels and self-reported stress.

In light of this dissociation, two questions arise: (1) Why do the neurophysiological and the psychological stress responses so rarely correlate? (2) And more specifically, why does social support buffer the neurophysiological stress response but is less likely to have an effect on the psychological stress response?

Regarding the first question, various reasons for the absence of a direct relationship between the neurophysiological and the psychological response to social-evaluative-threat have been discussed. For example, Hellhammer et al. (2009) argue that in addition to the involvement of brain structures related to the experience of emotions, many other neuroendocrine factors exert their influence on the stress-induced cortisol response. Hence, a low association of physiological and psychological stress responses is not very surprising. Further explanations aim at methodological issues as the timing of the assessment of self-reported stress (Hellhammer and Schubert, 2012) and some authors even doubt the link between cortisol and self-reported stress (e.g., Gruenewald et al., 2004).

Regarding the second question, the failure of social support to buffer the self-reported stress response might be—at least partially—explained by demand characteristics and self-reported biases. For example, since the aim of the TSST—stressing the participants—is quite obvious, participants may think that it is expected of them to report high levels of stress (e.g., Orne, 1962). But the opposite may also be true, due to self-protective mechanisms or avoidance motivations (e.g., Gramzow et al., 2003), participants may deny the level of stress experienced and indicate low amounts of stress. In both cases—either reporting more or less stress than actually experienced—an underestimation of the correlation between self-reported and physiological stress results and the effects of the experimental manipulation on the psychological level are undermined.

Moreover, a recent study by Het et al. (2012) suggests that high levels of cortisol in response to an acute stressor do not need to be associated with a negative emotional outcome, but may even have a mood-enhancing effect leading to less negative affect after the cessation of the TSST. Based on this finding, one could argue that participants who received no support prior or during the TSST indeed experienced higher levels of subjective stress than supported participants, but that these higher levels of subjective stress were buffered by the higher cortisol response. As a result, similar levels of subjective stress are reported by unsupported and supported participants. Although these explanations are speculative, they may serve as a starting point for further

research. Furthermore, as the results of the study by Ditzen et al. (2008) suggest, moderators on the psychological level, like the attachment style, should also be considered. Particularly secure attached individuals might profit more from social support on the neuroendocrine as well as on the subjective level.

Beyond the potential reasons for the dissociation, the question remains as to whether social support can be claimed to be effective when it only buffers the neuroendocrine reaction but not the subjectively experienced stress response. Stated differently, what is the relative impact of (a) neurophysiological and (b) psychological stress on health and general well-being? Future research is needed to address this question. Finally, there is one interesting implication of the finding that social support is frequently ineffective in buffering the subjectively experienced stress response: People may fail to capitalize on social support because in their experience it does not make them feel better—thereby giving away the benefits on the neurophysiological level (Frisch et al., 2014).

#### *Underlying mechanisms*

Another important limitation of the reviewed studies is that most of them do not address the specific mechanisms underlying the relationship between social variables and threat-related neurophysiological processes.

Regarding the effects of social variables on neurophysiological processes, it might be worthwhile to study, for example, by which pathways social support and social identification affect the neurophysiological response. One promising key to understand why social support can be effective would lie in examining its role in situational appraisal processes (e.g., Ben-Zur and Michael, 2007; Gallagher et al., 2014).

Regarding the influence of threat and the corresponding neurophysiological response on social processes, the underlying mechanisms are also largely unknown. Although most studies do report a correlational relationship between cortisol and the social processes under study, only in one of the studies has a formal analysis of mediation been conducted (Leder et al., 2013) following the Baron and Kenny (1986) approach. But, of course, even this analysis is not suited to establish causality, since it is essentially correlational. In order to unequivocally establish causality, a valuable alternative to this measurement-of-mediation design is to experimentally investigate the proposed casual chain (Spencer et al., 2005). Note that, from a neurophysiological point of view, the effects of social-evaluative threat on social cognition and social behavior may be due to (a) an increase in cortisol levels (via the HPA axis), (b) an increase in noradrenergic activity (via the SAM axis), or (c) an interaction of concurrent glucocorticoid and noradrenergic activity. To disentangle these processes, researchers can pharmacologically manipulate both glucocorticoid activity and noradrenergic activity, for example, by the administration of hydrocortisone and yohimbine, thereby employing a 2 × 2 experimental design (e.g., Schwabe et al., 2012). Another method for disentangling the neurophysiological effects of glucocorticoid and noradrenergic activity is to selectively suppress either the glucocorticoid or the noradrenergic stress response. For example, the glucocorticoid stress response can be suppressed by employing the dexamethasone suppression test (e.g., Andrews et al., 2012), whereas noradrenergic activity can be suppressed by

propranolol administration (Andrews and Pruessner, 2013). Thus, future studies might examine the neurophysiological underpinnings of social-evaluative threat by exposing participants to the TSST while simultaneously suppressing either the glucocorticoid or the noradrenergic stress response.

#### **EVALUATION OF THE TSST—INSIGHTS FROM RESEARCH ON THREAT IN SOCIAL INTERACTIONS**

In this review, we sought to evaluate the potential of the TSST which is the gold standard in biopsychological stress research (Kudielka et al., 2007)—as a method to examine the interplay between neurophysiological and social factors in threatening social interactions. From the studies reviewed above, we also can learn a lot about the TSST as an experimental paradigm, with respect to its range of applications, strengths and weaknesses. In what follows, we will summarize these insights and will evaluate the TSST. Let us start with the weaknesses of the TSST.

#### *Weaknesses of the TSST*

First, the scope of the TSST is limited to inducing a specific kind of threat, namely social-evaluative threat. If aiming at examining the consequences of, for example, threats to physical integrity, other paradigms might be more appropriate, such as the cold-pressor test in which participants immerse their hand for a few minutes into ice water (first described by Hines and Brown, 1936) or the socially evaluated cold-pressor test (Schwabe et al., 2008) which combines the cold-pressor test and TSST elements.

Second, a further problem of the TSST is that it actually contains two stress-inducing-elements: Apart from being socially evaluative, the TSST situation is highly uncontrollable for the participants. Thus, participants cannot influence their potential negative evaluation and only partially know what tasks they have to face and how things will proceed (Dickerson and Kemeny, 2004). Indeed, in many real life situations social-evaluative threat is tightly coupled with uncontrollability, since the behavior of the interaction partners in such situations can almost never be completely foreseen. However, uncontrollability and social-evaluative threat might still have independent effects, and these effects might differ with respect to different social processes. A way to remove or at least reduce the uncontrollability component could be to tell participants beforehand the exact procedure of the TSST (they could even been shown a video of the procedure).

Third, the TSST is not well suitedfor repeated measures designs: The HPA axis has been found to be highly sensitive to the effects of repeated stimulation with TSST and reacts with habituation to it (Pruessner et al., 1997; Schommer et al., 2003). This should be kept in mind when planning on using the TSST more than once in a sample.

Fourth, the findings of Buchanan et al. (2012) showing that the TSST can elicit an empathic stress reaction of the committee members—even at the neuroendocrine level—could be a problem for the internal validity of the TSST. Quite obviously, from a methodological point of view, the committee members' behavior should be exactly identical for all participants. Yet, the findings of Buchanan et al. (2012)suggest that this might not be the case since the committee members tend to contagiously catch the stress of the

participants. Whereas this contagion of stress across individuals is adaptive for coordinating the behavior of groups, it is problematic with regard to the internal validity of the TSST. However, these weaknesses of the TSST are balanced by several methodological strengths.

#### *Strengths of the TSST*

The TSST can be described as a reliable and effective, highly standardized psychosocial stress induction tool that simplifies the comparison and integration of findings of different studies but is still very flexible, so it can be used to study a variety of research questions.

First, the TSST leads reliably to a strong cortisol response as has been demonstrated in previous studies (Dickerson and Kemeny, 2004) as well as in all of the reviewed studies. Moreover, it is very effective, since about 70–80% of participants show a two- to threefold increase in cortisol levels (Kudielka et al., 2007). However, as we have seen in some of the studies (e.g., Smeets et al., 2009), it can be important to distinguish between high- and lowcortisol responders or to exclude cortisol non-responders (for discussion of exclusion criteria, see Miller et al., 2013). Hence, upon deciding to employ the TSST as a paradigm to induce socialevaluative threat, one can be sure that the basic precondition—the existence of a solid and high physiological and psychological stress reaction—is met for the majority of participants. Moreover, although it causes a strong physiological and psychological stress reaction, the TSST is still in accordance with established ethical research standards, like the declaration of Helsinki.

Second, the TSST protocol is highly standardized. The procedure is well-documented and since no special apparatuses (apart from a video camera and a microphone) or questionnaires are needed it can be easily applied in nearly every laboratory.

Third, this high degree of standardization allows for comparisons between different studies in one field of research. This facilitates the integration of these findings in reviews and meta-analysis as well as the replication and extension of previous studies.

Fourth, notwithstanding the high degree of standardization, the TSST is still flexible and can be adapted to the specific needs of the researcher, thereby paving the way to investigate a huge variety of research questions (e.g., examining the effectiveness of social support, or the existence of empathic stress reactions). Hence, to date several variations of the TSST exist for specific research questions (e.g., different control groups, group version) or populations of participants (e.g., children, older adults). One frequent employed variation is the TSST-G (von Dawans et al., 2011). For example, the TSST-G can be used to study the effects of social-evaluative threat in groups of participants (e.g., Häusser et al., 2012). Moreover, due the simple composition of four main elements (i.e., anticipation/preparation period, free speech task, mental arithmetic task and social-evaluative component) (Kirschbaum et al., 1993), the TSST can be adapted to the specific research question and context of each study by simply altering these elements or by adding new elements. For example, as we have seen in the research on the effectiveness of social support, it was possible to add a supportive element (i.e., friend, stranger or confederate) to the TSST protocol. Moreover, the relationship between the TSST committee and the participant or

even between the participants in a TSST-G can be manipulated (Häusser et al., 2012; Frisch et al., 2014). Furthermore, the temporal sequence of the three phases of the TSST (i.e., preparation, speech task, math task) allows the measuring of behavior or selfreported emotions/cognitions at different time points and may help to reveal time-dependent effects: (a) Before the participants are told about the TSST (baseline), (b) after the preparation phase (anticipatory stress reaction), (c) during the two task (stress reaction) (d) directly and after the TSST (post-stress reaction) and (e) at several measurement points after the TSST (recovery reaction) (e.g., Kirschbaum et al., 1993; Hellhammer and Schubert, 2012).

Taken together, the TSST is perfectly suited for investigating the complex interplay of social-evaluative threat, social processes, and neurophysiological stress responses. We hope that our review article will stimulate new research directions and propel this field forward.

#### **ACKNOWLEDGMENTS**

We would like to thank Markus Germar for valuable discussions as well as for the willingness to pose on the picture of the TSST procedure together with Arzu Cetin and Daniela Röttinger.

#### **REFERENCES**


the test. *Psychoneuroendocrinology* 37, 119–124. doi: 10.1016/j.psyneuen.2011. 05.012


*Cure: Identity, Health and Well-Being*, eds J. Jetten, C. Haslam, and S. A. Haslam (London: Psychology Press), 175–194.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 30 October 2014; accepted: 06 January 2015; published online: 02 February 2015.*

*Citation: Frisch JU, Häusser JA and Mojzisch A (2015) The Trier Social Stress Test as a paradigm to study how people respond to threat in social interactions. Front. Psychol. 6:14. doi: 10.3389/fpsyg.2015.00014*

*This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology.*

*Copyright © 2015 Frisch, Häusser and Mojzisch. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited andthatthe original publication inthis journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Regulatory focus and generalized trust: the impact of prevention-focused self-regulation on trusting others

#### *Johannes Keller 1\*, Ruth Mayo2, Rainer Greifeneder <sup>3</sup> and Stefan Pfattheicher <sup>1</sup>*

*<sup>1</sup> Abteilung Sozialpsychologie, Institut für Psychologie und Pädagogik, University of Ulm, Ulm, Germany, <sup>2</sup> The Hebrew University of Jerusalem, Jerusalem, Israel, <sup>3</sup> University of Basel, Basel, Switzerland*

The current research suggests that taking self-regulatory mechanisms into account provides insights regarding individuals' responses to threats in social interactions. In general, based on the notion that a prevention-focused orientation of self-regulation is associated with a need for security and a vigilant tendency to avoid losses and other types of negative events we advocate that a prevention-focused orientation, both as a disposition as well as a situationally induced state, lowers generalized trust, thus hindering cooperation within social interactions that entail threats. Specifically, we found that the more individuals' habitual self-regulatory orientation is dominated by a prevention focus, the less likely they are to score high on a self-report measure of generalized trust (Study 1), and to express trust in a trust game paradigm as manifested in lower sums of transferred money (Studies 2 and 3). Similar findings were found when prevention focus was situationally manipulated (Study 4). Finally, one possible factor underlying the impact of prevention-focused self-regulation on generalized trust was demonstrated as individuals with a special sensitivity to negative information were significantly affected by a subtle prevention focus manipulation (versus control condition) in that they reacted with reduced trust in the trust game (Study 5). In sum, the current findings document the crucial relevance of self-regulatory orientations as conceptualized in regulatory focus theory regarding generalized trust and responses to threats within a social interaction. The theoretical and applied implications of the findings are discussed.

#### Keywords: generalized trust/distrust, regulatory focus, self-regulation, sensitivity to negative information

# Introduction

#### Vigilant Prevention-Focused Self-Regulation and Generalized Trust

Self-regulatory orientations influence many fundamental social cognitive as well as social interactive mechanisms and affect individuals' thought processes, emotions, and behavioral tendencies in social interactions (for an overview of the field of self-regulation research, see the volume edited by Vohs and Baumeister, 2011). One theoretical perspective figures particularly prominently in this research on self-regulatory orientations: regulatory focus theory (RFT) introduced by Higgins (1997, 1998, 2012a,b). The present contribution aims to explore the role of RFT with respect to a fundamental and important social interactive phenomenon: generalized trust. Specifically, we

# *Edited by:*

*Eva Jonas, University of Salzburg, Austria*

#### *Reviewed by:*

*Roland Deutsch, Technische Universität Dresden, Germany Ann-Christin Posten, Harvard University, USA*

#### *\*Correspondence:*

*Johannes Keller, Abteilung Sozialpsychologie, Institut für Psychologie und Pädagogik, University of Ulm, 89081 Ulm, Germany johannes.keller@uni-ulm.de*

#### *Specialty section:*

*This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology*

> *Received: 29 October 2014 Accepted: 20 February 2015 Published: 17 March 2015*

#### *Citation:*

*Keller J, Mayo R, Greifeneder R and Pfattheicher S (2015) Regulatory focus and generalized trust: the impact of prevention-focused self-regulation on trusting others. Front. Psychol. 6:254. doi: 10.3389/fpsyg.2015.00254* address the relations between prevention- and promotionfocused self-regulation and generalized trust, mainly as affecting responses to social interactions that involve threats. The basal motivational orientations as conceptualized in RFT play a crucial role in dealing with threats and uncertainty (prevention focus) as well as with opportunities for growth and attainment of maximal goals (promotion focus). Situations where a decision is pending whether one is willing to trust others reflect a state of uncertainty in that one is exposed to the threat that trusting others may prove to be disadvantageous. In this regard, we propose that especially prevention-focused self-regulation is crucial regarding the willingness to trust others in social interactions that entail threats. Overall, the present research adds to and extends a growing body of research, which puts the spotlight on the self-regulatory character of psychological phenomena (cf. Carver, 2006).

To build a common basis for our arguments, we start out with a brief discussion of the core assumptions proposed in RFT. Based on this, we offer a conceptual analysis to explore the relation between prevention-focused self-regulation and generalized trust.

# Core Assumptions of Regulatory Focus Theory

Extending the basic hedonic principle that people approach pleasure and avoid pain, RFT holds that it is necessary to differentiate distinct types of pleasures and distinct types of pain and to assess the specific strategic orientations and types of goal pursuit that reflect self-regulation guided by two distinct motivational systems – promotion focus and prevention focus (Higgins, 1998, 2012a,b; Molden et al., 2008). Self-regulation with a promotion focus is characterized as the motivation to attain growth and nurturance, to bring one's actual self into alignment with one's ideal self, as well as the desire to reach gains (and to avoid nongains). In contrast, self-regulation with a prevention focus entails the motivation to attain security, to bring one's actual self into alignment with one's ought self (i.e., fulfilling one's duties and obligations), as well as the desire to avoid losses (and to attain non-losses).

Both types of regulatory orientations are presumed to be related to specific consequences. RFT postulates several consequences of self-regulation with a promotion focus: (a) a special sensitivity to the presence or absence of *positive* outcomes, (b) application of *eager* strategic means (i.e., to insure hits and to insure against errors of omission), (c) ambitious and keen striving to reach one's aspirations as reflected in tenacious goal pursuit that is focused on *maximal* goals (i.e., goals differentiating a positive region of outcomes from a non-positive/neutral region; cf. Brendl and Higgins, 1996), and (d)*cheerfulness-dejection* emotions in response to positive and negative events. In contrast, according to RFT, self-regulation with a prevention focus is associated with the following consequences: (a) a special sensitivity to the presence or absence of *negative* outcomes, (b) application of *vigilant* strategic means (i.e., to insure correct rejections and to insure against errors of commission), (c) a defensive orientation in the pursuit of *minimal* goals (i.e., goals differentiating a negative region of outcomes from a non-negative/neutral region; cf. Brendl and Higgins, 1996), and (d) *quiescence-agitation* emotions in response to positive and negative events. These theoretical assumptions have been supported by substantial empirical evidence (see Higgins, 1998, 2012a,b; Higgins and Spiegel, 2004, for reviews).

Particularly important in the present context, RFT posits that individuals may differ in their predominant *chronic* or *habitual* self-regulatory orientation, and several measures to assess these individual differences have been developed (e.g., regulatory focus questionnaires, cf. Higgins et al., 2001; Lockwood et al., 2002; Keller, unpublished regulatory strength measures, cf. Shah et al., 1998). The two self-regulatory systems can also be manipulated, that is, situationally induced or primed (Roney et al., 1995; Shah et al., 1998; Friedman and Förster, 2001; Freitas et al., 2002).

It is important to note that the two modes of self-regulation, promotion and prevention, have been conceptualized as independent constructs and therefore may vary independently. That is, individuals can be high on both, low on both, or predominantly prevention- or promotion-focused. Accordingly, measures of the two modes of self-regulation (chronic promotion and prevention-focused orientation) have been found to be largely uncorrelated or slightly positively correlated (cf. Higgins et al., 2001; Lockwood et al., 2002; Keller, unpublished). Promotion-focused self-regulation is thus *not* the opposite pole of prevention-focused self-regulation. In consequence, the two modes of self-regulation may display very different patterns of relations with other constructs, such that one of the two modes may be related to a certain phenomenon or construct while the other mode is not.

The present contribution aims to explore the role of RFT with respect to the fundamental and important social interactive phenomenon of trust, specifically as reflected in response to social situations that entail threats. Based on the notion that a *prevention-focused* orientation of self-regulation is associated with a need for security and the desire to attain it, a tendency to avoid dangers and threats, as well as a proclivity to be defensive and vigilant, we hypothesized that such an orientation is likely to lower generalized trust, in turn hindering the potential for cooperation in a social interaction. Given that trust involves a willingness to accept vulnerability and to take risks – a tendency which appears largely incompatible with a concern for safety and security – it seems reasonable to assume a negative relation between a prevention-focused mode of self-regulation and trust. We discuss the general logic underlying this proposition in the next paragraphs.

# Trust and Its Relation to Self-Regulatory Mechanisms

According to the Oxford English Dictionary, trust is "[c]onfidence in or reliance on some quality or attribute of a person or thing, or the truth of a statement," and the act of trusting is defined as "to accept or give credit to without investigation or evidence." What is probably the most widely cited definition of trust in the psychological literature was introduced by Rotter (1971, p. 444) who defined trust as "an expectancy held by an individual or a group that the word, promise, verbal, or written statement of another individual or a group can be relied upon." In all of these definitions, trust is seen as lending credibility to a person or group without (conclusive) knowledge about the actual credibility or reliability of the person or group. This aspect also figures prominently in Rempel et al.'s (1985) conceptualization of trust, which refers to predictability, dependability (the willingness to put oneself at risk through reliance on another's promises), and faith (confidence in caring responses) as crucial components. In a more recent definition, Rousseau et al. (1998, p. 395) proposed a conceptualization of trust according to which "[t]rust is a psychological state comprising the intention to accept vulnerability based upon positive expectations of the intentions or behavior of another.". This perspective specifically highlights an aspect of trust that is also (if only implicitly) entailed in the definitions mentioned above, namely that trust involves a certain level of *willingness to accept risks and uncertainty* (in the sense that the individual accepts the possibility that he or she might suffer a loss, injury, or harm as a consequence of engaging in an interaction) based on positive expectations about the other person's intentions. Thus, the core of generalized trust is the global belief that people are likely to be reliable, sincere, cooperative, benevolent, and truthful (Simpson, 2007).

As mentioned, our work focuses on *generalized* trust. This perspective differs from the dyadic interpersonal perspective on trust that rose in the partnership- and relationship-centered programs of research initiated in the 1980s (cf. Larzelere and Huston, 1980; Johnson-George and Swap, 1982; Rempel et al., 1985; Holmes and Rempel, 1989). Specifically, trusting a well-known partner in a close relationship might be different to generalized trust including unknown others and people in general (cf. General Discussion).

Trust can be examined from several perspectives and levels. On the *cognitive* level, trust entails the expectation that most others have benign intentions (e.g., Acar-Burkay et al., 2014). In social interactions individuals often do not know how other individuals will behave. That is to say, many social interactions entail a threat that one is exploited by uncooperative and antisocial behavior of others (e.g., Balliet and Van Lange, 2013a,b). In threatening social situations trust decreases, especially among individuals who possess few resources and who feel powerless to deal with the threat (Ross, 2011). However, if individuals do not trust others, they miss the possibility of beneficial social interactions and positive reciprocity (Berg et al., 1995; McCabe et al., 2003). Some amount of trust is therefore necessary for ongoing and beneficial social interactions.

On the *behavioral* level, trust reflects a willingness to accept a state of dependency on another individual who has the power to return harm or benefits (Ainsworth et al., 2014). Behaving trustworthily toward others increases cooperative behavior in social groups (Balliet and Van Lange, 2013a) and lowers transaction costs because interaction partners need to be less monitored (Cook et al., 2005). In close relationships, showing trust is positively related to felt security, constructive strategies in coping, and emphasizing positive aspects of the relationship (Mikulincer, 1998; Rempel et al., 2001). Thus, showing trust also counters uncertain and threatening social situations.

Research regarding trust – and its counterpart, distrust – has also included the prevalence of deception in our life (DePaulo et al., 1996; DePaulo and Kashy, 1998; Feldman et al., 2002), and social perceivers' attempts (usually unsuccessful) to decode deception (DePaulo et al., 1997; Anderson et al., 1999) and the differentiation of cooperators from deceivers (Yamagishi et al., 2003). Recent research has investigated the impact of hormones, especially oxytocin, on trust behavior (Kosfeld et al., 2005; Van IJzendoorn and Bakermans-Kranenburg, 2012). However, somewhat surprisingly, the role of basic *motivational* mechanisms with respect to generalized trust has been largely neglected to date. Given the eminent relevance of trust in social life, it seems important to understand the underlying motivational mechanisms involved in generalized trust, trust behavior, as well as the boundary conditions that may lead to (or inhibit) the expression of generalized trust. We address this issue by focusing specifically on the role of basic self-regulatory mechanisms as proposed in RFT.

As outlined in RFT (Higgins, 1997, 1998, 2012a,b), prevention-focused self-regulation reflects a desire to avoid losses and need for security, the desire to reach safety. In this regard, particularly prevention-focused individuals consider security to reflect a desirable human value (Leikas et al., 2009). Moreover, prevention-focused self-regulation is associated with a special sensitivity to (potential) negative events (e.g., Keller and Pfattheicher, 2013; Pfattheicher and Sassenrath, 2014). In this sense, Pfattheicher and Keller (2013) showed that strongly prevention-focused individuals punish antisocial others in a social dilemma situation more than weakly prevention-focused individuals. Neural correlates also supports this assumption indicating a greater activity in the amygdala, anterior cingulate, and extrastriate cortex for prevention-focused individuals when negative (versus positive) information is presented (Cunningham et al., 2005). Furthermore, preventionfocused self-regulation has been related to vigilant and careful strategic tendencies reflecting threat, defensiveness, and cautiousness (note that evidence documenting a significant negative correlation between measures of a prevention-focused selfregulatory orientation and sensation-seeking as a measure of disinhibition supports this notion; Uskul et al., 2008; cf. Keller, unpublished).

Social situations that demand trusting others entail a threat that one is exploited by uncooperative and antisocial behavior of others (e.g., Balliet and Van Lange, 2013a,b). That is to say, when trusting others individuals are in a situation of insecurity, vulnerability, and uncertainty. As is evident from this list of concepts that are related to the prevention focus, a prevention-focused selfregulatory orientation reflects a mode of self-regulation that is concerned with security while avoiding negative events to happen. In fact, the defining concepts that characterize trust and prevention focus seem to be incompatible (e.g., the willingness to accept vulnerability, uncertainty, and risk as characteristic feature of trust compared to a concern for safety and security in prevention focus). Accordingly, we assume that there is a negative relation between a prevention-focused mode of self-regulation and generalized trust.

In contrast to the close conceptual relation between prevention-focused self-regulation and trust, the defining aspects involved in the conceptualization of promotion-focused self-regulation appear not particularly strongly associated with the willingness to accept vulnerability, uncertainty, and risk, as the defining characteristic of trust. Moreover, since both types of self-regulation (promotion and prevention) have been conceptualized as independent constructs, proposing a negative association between prevention focus and generalized trust does not imply the reverse, that is to say, a positive association between promotion focus and generalized trust.

In sum, the conceptual analysis outlined above leads us to hypothesize a negative relation between prevention-focused self-regulation and generalized trust. In contrast, based on the conceptual analysis outlined above, it seems most plausible to expect that promotion-focused self-regulation is largely unrelated to generalized trust. As a starting point, we tested these basic assumptions in three studies assessing the relation between self-report measures of regulatory focus and (a) a self-report measure of generalized trust (Studies 1a,b), and (b) a measure designed to assess the behavioral tendency to trust another unfamiliar person based on the experimental trust game paradigm as a case of social interaction that entails a threat (Studies 2–5). To further bolster confidence in these findings, we manipulated prevention focus in Study 4. Finally, going beyond the global relationship between prevention-focused self-regulation and generalized trust, we elaborate on the specific mechanisms that may be involved by manipulating a prevention focus problem cue and testing its effect on trust behavior as a function of one specific dispositional tendency (sensitivity to negative information) that is known as a characteristic element of prevention-focused self-regulation (Study 5).

All studies of the present work were conducted in line with the ethical standards of the American Psychological Association (APA). All participants were given written informed consent prior to the study. In order to assure anonymity, participants did not provide information that allows inferences to the participants (e.g., names). Participants were paid in private at the end of each study and debriefed.

# Study 1

The goal of Study 1 was to assess whether individual differences in generalized trust are related to differences in chronic regulatory focus.

#### Procedure

This study involved two independent samples of students at the University of Mannheim (Study 1a: *N* = 88; *M*age = 22.9; 45 females; Study 1b: *N* = 117; *M*age = 23.3; 56 females)1 . Participants in the two questionnaire studies completed measures designed to assess habitual levels of generalized trust and regulatory focus. In Study 1a, responses were assessed using 9-point response scales; in Study 1b, we used 7-point scales. In both studies, scale endpoints were labeled *not at all true* and *completely true*. We used a German version of Rotter's (1967) trust scale to assess generalized trust (Amelang et al., 1984). A sample item of this measure reads: "In dealing with strangers one is better off being cautious until they have provided evidence that they are trustworthy." The scale reached Cronbach's alpha levels of 0.76 (Study 1a) and 0.77 (Study 1b).

We assessed regulatory orientations using a German version (Keller and Bless, 2006) of the regulatory focus scale developed by Lockwood et al. (2002), which consists of prevention and promotion focus subscales. A prevention scale sample item reads: "In general, I am focused on preventing negative events in my life." And a promotion scale sample item reads: "I frequently imagine how I will achieve my hopes and aspirations." The promotion scale reached Cronbach's alpha levels of 0.77 (Study 1a) and 0.81 (Study 1b); the prevention scale reached Cronbach's alpha levels of 0.76 (Study 1a) and 0.80 (Study 1b)2 .

In Study 1b, additional measures of basic dimensions of personality (extraversion; neuroticism) were assessed using a German version (Eggert, 1983) of Eysenck's personality inventory in order to test whether the assumed association between individuals' regulatory orientation and their level of generalized trust remains robust when controlling for these broad personality dimensions. Both scales reached acceptable levels of internal validity, Cronbach's α = 0.78 (extraversion) and 0.87 (neuroticism).

#### Results and Discussion

To test the relations between regulatory orientation and generalized trust we computed zero-order correlations (see **Table 1**) as well as regression analyses. In both samples (Studies 1a,b) prevention focus was significantly and negatively correlated with generalized trust (see **Figures 1** and **2**). These results support the proposed negative association between the two constructs. The regression analyses controlling for promotion scores (Study 1a; top panel of **Table 2**) as well as scores on the neuroticism and

<sup>1</sup>We are well aware of the fact that following the false positive debate it is now considered desirable to conduct studies with a determination of sample sizes based on power analyses. Please note that the studies reported in the current manuscript were conducted well before the false positive debate was initiated in 2011. Accordingly, we are not in a position to refer to power analyses that were conducted prior to the data collection. What we did was that we took into

consideration previous studies in the field of regulatory focus research regarding correlations between measures of individual differences in regulatory foci and measures of other trait level constructs. These considerations suggested that sample sizes between 80 and 120 typically provided enough power to detect meaningful associations.

<sup>2</sup>We also assessed an alternative regulatory focus instrument developed in our lab (Keller, unpublished) that has been used in several studies (cf. Keller and Bless, 2008; Leder et al., 2013, 2015). The alternative regulatory focus scale was assessed after the Lockwood scale in this study (with six items taken from the need for cognition scale as filler items between the Lockwood scale and the alternative regulatory focus scale). A sample promotion-item of this instrument reads: "If I know that my performance is being evaluated by other people that spurs me on and increases my ambition to do well." A sample prevention-item reads: "My life is often shaped by fear of failure and negative events." Both scales were internally consistent and reached acceptable Cronbach's alpha levels (Promotion: 0.65; Prevention: 0.81). Note that these scales are conceptually similar to the Lockwood et al. (2002) scales (cf. Keller, unpublished). For reasons of brevity, and given that the Lockwood et al. (2002) instrument is the more established measure of chronic regulatory focus, we focus on these latter scales in the results section.

TABLE 1 | Zero-order correlations between variables included in Studies 1a,b.


*Results obtained in Study 1a are shown below the diagonal; results of Study 1b are depicted above the diagonal.* + *< 0.07;* ∗*p < 0.05,* ∗∗*p < 0.01,* ∗∗∗*p < 0.001.*

extraversion scales (Study 1b; bottom panel of **Table 2**) reveal that this relation is robust.

One unexpected aspect of the findings that emerged in Study 1 is the fact that the promotion scale did show a modest negative association with generalized trust scores. One may speculate that this modest negative relation may be attributed to the fact that promotion–focused self-regulation has been TABLE 2 | Regression analyses testing regulatory focus, neuroticism, and extraversion as predictors of generalized trust scores in Studies 1a,b.


∗*p < 0.05,* ∗∗*p < 0.01,* ∗∗∗*p < 0.001.*

proven to be positively related to human values reflecting self-enhancement (achievement and power; cf. Leikas et al., 2009; Keller, unpublished) and negatively related to values reflecting self-transcendence (universalism and benevolence; cf. Leikas et al., 2009; Keller, unpublished). Given that selfenhancement values haven been found to relate negatively to generalized trust (e.g., Tatarko, 2013), the modest negative relation between promotion focus and trust observed in our study may reflect the specific value connotation of promotion-focused self-regulation. However, this is a *post hoc* consideration and as outlined above, there are no strong reasons to predict a negative relation between the promotion focus and generalized trust on conceptual grounds (and in line with this reasoning we did *not* find such a negative relation in the studies reported below).

In combination, the present results support our reasoning that a prevention-focused self-regulatory orientation is inversely related to generalized trust. In the next studies we turn to test our hypothesis within a social interaction that entails threats.

# Study 2

We designed Study 2 to replicate the negative relation between prevention-focused self-regulation and trust using the trust game introduced by Berg et al. (1995). In the trust game, the decision to transfer money to an unfamiliar interaction partner is taken as a good proxy of trustful behavior (e.g., Bohnet and Zeckhauser, 2004; Buchan and Croson, 2004; Kosfeld et al., 2005; Greifeneder et al., 2011). In this experimental game, participants initially receive a certain amount of money. They are informed that they can transfer any amount of this money to a second person and that this amount will be tripled before the other person receives the transfer amount. At this point, the other person (who receives the money) is free to decide whether or not to reciprocate by sending a certain amount of money back. Thus, this paradigm is designed to assess the level of trust on the part of the initiator (i.e., the person who can decide on how much money he or she is willing to put in the second person's hands). Individuals with a strong tendency to trust are expected to send the full amount to the interaction partner, while the lowest level of trust is represented by the decision not to send any money to the interaction partner. We predict that the more prevention-oriented the person in the position to initiate the money transfer is, the smaller are the chances that he or she is willing to trust another unfamiliar person, in other words – to respond to a threat (of receiving little or no money back) with trust. Accordingly, we expect an inverse, that is, negative, relation between participants' scores on the measure of prevention-focused self-regulation and the amount of money offered to a (hypothetical) second player.

#### Procedure

Seventy-nine students at the University of Mannheim (*M*age = 21.4; 43 women, 32 men; four participants did not indicate their sex) filled out a questionnaire that included a short version of the regulatory focus scale (with three items per subscale; Cronbach's alpha Promotion: 0.65; Prevention: 0.61) as applied in Study 1 (Lockwood et al., 2002).

To operationalize generalized trust other than by responses to questionnaire items, we relied on a scenario version of the trust game paradigm (cf., e.g., Buchan and Croson, 2004, for uses of the trust game in scenario version). In this scenario version, we asked participants to imagine that they were participating in a study on social behavior and that each participant in this imaginary study initially receives a payment of 12€ for showing up. Moreover, our participants learned that in the imaginary study participants were randomly assigned to the role of either a money transfer initiator, or transfer receiver. They were told that participants in the role of a money transfer initiator would have the option of transferring an amount between 0 and 12€ (the amount of money each participant had received for showing up) to another anonymous person (the receiver) who would then decide on how much money he or she wanted to send back. It was further explained that the amount of money the initiator was willing to transfer (between 0 and 12€) would be *tripled*. That is, if the initiator decided to transfer 4 Euro, the receiver would get 12€ and could then decide on how much of this amount he or she was willing to send back to the initiator. Participants also learned that in the imaginary study the persons involved in the interaction would remain absolutely anonymous and had no chance to communicate with each other.

Following the detailed description of the scenario, participants were asked to imagine themselves in the role of the transfer initiator and to decide on the amount of money they would be willing to transfer to the receiver (on a scale with 13 response options representing units of 1 between 0 and 12€).

#### Results and Discussion

In line with our theoretical analysis and the findings obtained in Study 1, we expected a negative association between participants' chronic prevention focus and the amount of money they would be willing to transfer to the interaction partner. Responses in the trust game paradigm varied between 0 and 12€ and the distribution was clearly non-normal (multiple peaks emerged: 17 participants selected 6€, 12 participants selected 10€, and 10 participants selected 12€). Accordingly, we applied non-parametric analyses to assess the relation between regulatory focus scales and trust game responses. We computed non-parametric correlations (Kendall's τ) between the Lockwood et al. (2002) focus scales and trust game responses. As expected, the prevention focus was negatively related to generalized trust as reflected in the tendency to give money to a stranger (Kendall's τ = −0.18, *p <* 0.03; see **Figure 3**). In contrast, the promotion focus was slightly positively related to trust game responses (Kendall's τ = 0.05, *n.s.*) 3 .

When we entered promotion and prevention scale scores simultaneously in ordered logit regression analyses, we found parallel results. Specifically, prevention scale scores emerged as a significant predictor, Wald χ<sup>2</sup> = 8.12, *p <* 0.01, the respective odds ratio of 0.61 indicates that a one-unit change in the prevention scale score decreases the odds of the decision to transfer the full amount available (versus the combined other response categories) by 39%. The coefficient for the promotion scale scores was not significant, Wald <sup>χ</sup><sup>2</sup> <sup>=</sup> 1.07, *<sup>p</sup> <sup>&</sup>gt;* 0.30, odds ratio 1.25.

# Study 3

The findings reported thus far consistently support the hypothesis that prevention-focused self-regulation is negatively related to

<sup>3</sup>Parallel results emerged in this study (as well as in the studies reported below) when we conducted parametric analyses (i.e., "classic" regression analyses). Also, the observed pattern of results was corroborated when analyzing the relation between trust and prevention focus as measured with Keller's scale. This scale was assessed directly preceding the short version of Lockwood scale. Results based on Keller's (unpublished) scale showed a similar picture. The prevention scale revealed a negative relation (Kendall's τ = −.16, *p <* 0.06); the promotion scale was marginally significantly positively related with trust game responses (Kendall's τ = 0.15, *p >* 0.07).

generalized trust. Specifically, prevention-focused self-regulation leads to responding with less trust in a social interaction paradigm that involves a threat. Given that Studies 1 and 2 were based on self-report measures of generalized trust and a scenario version of the trust game, it remains to be tested whether the relation holds when we assess actual trust behavior. To address this critical question, we assessed whether participants' actual transfer of real money in the trust game paradigm was similarly affected by the individuals' level of prevention-focused orientation.

## Procedure

Seventy-seven students at the University of Mannheim (*M*age = 22.4; 37 women) participated in this study and received monetary compensation (2€ plus extra money depending on their response in the experimental setting, see description below). Four participants were excluded because they did not follow experimental procedures adequately, leaving a sample size of 73 for subsequent analyses.

When participants arrived at the lab, a confederate of the experimenter appeared and served as ostensible other participant so that participants were made to believe that there was another participant present with whom they would later be interacting. There was no vocal communication between confederate and participants. Participants were told that the other participant would be sitting in an adjacent room and completing the same material. Once seated in their cubicle, participants received a questionnaire that contained the measure of chronic regulatory focus used in Study 1 [Lockwood et al.'s (2002) scales; αPromotion = 0.81; αPrevention = 0.76].

# Trust Game Paradigm Involving Real Money

After completing the regulatory focus measure, participants received new materials from the experimenter and read that the researchers were interested in individuals' reactions in social interactions involving financial decision-making. As stated above, participants initially received 2€ (in form of ten 20 Cent coins) for showing up and read in the description of the trust game paradigm that they could exchange this money. Participants learned that they would be randomly assigned either to the role of a money transfer initiator or the role of a transfer receiver (in reality, all participants were assigned to the role of money transfer initiator). Participants in the role of a money transfer initiator had the option to transfer any amount between 0.204 and 2€ to another person (the receiver), who would then decide how much money he or she wanted to send back. In line with the general logic of the trust game (as outlined in Study 2 above), it was further explained that the amount of money the initiator was willingto transfer (between 0.20 and 2€) would be *tripled*. Moreover, participants were told that the individuals involved in the interaction would remain strictly separated and had no chance of communicating with each other in the course of the study or after.

Following the detailed description of the scenario, participants were asked to decide on the amount of money they would be willing to transfer to the receiver and to put the respective amount in an envelope provided by the experimenter, who then ostensibly brought the envelope to the transfer receiver in the adjacent room. The experimenter then returned, handed over one of several previously prepared envelopes containing the amount of money that either reflected a repayment of 100% (for participants randomly assigned to the norm violation condition) or a repayment of 160% (for participants randomly assigned to the control condition) and provided the participant with further materials that are not relevant in the present context (i.e., the trust game was part of a different line of research on selfregulation and aggression that has been reported by Keller et al., 2008).

# Results and Discussion

As in Study 2, the distribution of scores representing money transfer decisions was clearly non-normal. Specifically, a large proportion of participants (*n* = 38 or 52.1%) decided to transfer the complete available amount of 2€. This behavior can best be described as a decision to "give all" while the remaining participants decided to "give some" of the available money. Accordingly, we conducted an analysis predicting the decision to "give all" (transfer of 2€) or to "give some" by way of a logistic regression with the respective dummy variable as the criterion. We expected findings replicating the patterns obtained in Study 2, that is, a negative relation between prevention scale scores and the decision to "give all." Results of the logistic regression analysis revealed that prevention scale scores predicted the decision, *<sup>B</sup>* = −0.52, Wald <sup>χ</sup><sup>2</sup> <sup>=</sup> 3.68, *<sup>p</sup>* <sup>=</sup> 0.055, the respective odds ratio was 0.59. Promotion scale scores were not reliably associated with the trust game decision, *B* = −0.17, Wald <sup>χ</sup><sup>2</sup> <sup>=</sup> 0.32, *<sup>p</sup>* <sup>=</sup> 0.57, odds ratio <sup>=</sup> 0.83. The same pattern was found using zero-order correlations: analyses revealed that prevention scale scores were significantly negatively related to the decision to "give all" (*r* = −0.25, *p* = 0.03), whereas promotion focus scores were not significantly related (*r* = −0.12, *p* = 0.32).

These results speak to the fact that the inverse relation between prevention-focused self-regulation and generalized trust can be documented not only based on self-report measures of generalized trust and a scenario version of the trust game but also based on actual trust behavior involving real money in the trust game paradigm.

In this context, one could argue that the willingness to transfer money in the trust game paradigm is primarily due to the individuals' willingness to incur risk. According to this perspective, knowing participants' tolerance of risk may be sufficient to predict their transfer behavior, regardless of how much or how little they trust the other player. However, such a perspective is incompatible with empirical findings documenting that there are no meaningful relations between measures of risk taking and behavior in the trust game paradigm. For instance, Eckel and Wilson (2004) revealed that neither the Zuckerman (1994) sensation seeking scale (a widely used measure of risk taking) nor behavioral risk measures (involving lottery choices; cf.

<sup>4</sup>In this particular study, we offered no option to refrain from transferring money (0€) because the trust game paradigm was used in the context of another research project in order to manipulate perceived fairness in terms of the back transfer of money by the ostensible interaction partner (cf. Keller et al., 2008); accordingly, the 0€ response option was omitted in this version of the trust game paradigm to ensure that a money exchange was actually involved in every session of the study.

Holt and Laury, 2002) were related to the decision to trust in the trust game paradigm. Moreover, participants in the trust game seem to care not only about the payoff outcome but "behave as though there is a betrayal cost above and beyond any dollar losses" (Bohnet and Zeckhauser, 2004, p. 474). Kosfeld et al. (2005) showed that risk taking can be differentiated from trust and that risk calculations are of minor relevance in a trust game. Accordingly, it appears fair to conclude that trust game responses do not reflect mere risk taking, but a generalized tendency to trust in the face of a social threat of not receiving something back, as is evident in several investigations documenting substantial positive correlations between trust game responses and self-report measures of trust or reported past trusting behaviors (e.g., Glaeser et al., 2000; Evans and Revelle, 2008). Taken together, this evidence clearly speaks to the validity of the trust game paradigm and is incompatible with the proposition that responses in this paradigm reflect nothing but willingness to take risks. We also want to mention that the findings including the trust game are complemented by measures assessing individual differences in trust (Studies 1 and 2) where risk calculations are less likely to be relevant. Here, the same pattern emerged, that is, prevention focus is negatively related to individual differences in trust.

# Study 4

Studies 1–3 demonstrate that chronic prevention focus is inversely related to generalized trust. In a next step we extended these finding and tested whether a parallel relation of preventionfocused self-regulation and generalized trust can be documented when prevention focus is situationally activated. It is evident that this would be an important contribution to the current line of research in that we could draw firmer conclusions regarding the (causal) nature of the relation in question. Accordingly, we conducted an experimental study involving the manipulation of the prevention focus and we assessed participants' responses in the trust game paradigm (parallel to Study 2) following exposure to this manipulation.

### Procedure

Sixty students at the University of Mannheim (*M*age = 22.0; 30 women) participated in this study. They were randomly assigned to one of two experimental conditions (see below) and received a questionnaire containing the relevant materials.

# Prevention Focus Manipulation

In order to induce a prevention focus, half of the participants were asked to report on their duties and obligations (following the logic proposed by Higgins et al., 1994; see also Freitas et al., 2002). Specifically, we provided half of the participants with a description of the concept of the ought self followed by a statement referring to the fact that a defining feature of the ought self was the fear of (a) not possessing the respective elements of the ought self and (b) being punished or rejected because of an existing discrepancy between the actual and ought self. Following this introduction, participants noted three traits or characteristics that reflected elements of their ought self. Finally, we asked participants to indicate the person(s) that considered the indicated elements of their ought self as relevant. That is, participants mentioned the person(s) who made them feel responsible (or obligated) to possess the respective trait or characteristic and who may punish or reject them in case of an existing actual-ought discrepancy. The other participants (in the control condition) did not receive this part of the questionnaire.

As a measure of generalized trust, we used again a scenario version of the experimental trust game paradigm (parallel to Study 2; see also Buchan and Croson, 2004). That is, the main dependent variable in this study was the amount of money participants were willing to transfer to the receiver (on a scale with 13 response options representing values between 0 and 12€).

# Results and Discussion

In line with our theoretical analysis and the findings obtained in Studies 1–3, we expected a lesser amount of money to be transferred in the condition of prevention focus induction compared to the control condition. Responses in the trust game paradigm varied between 0 and 12€ and the distribution was again clearly non-normal (multiple peaks emerged: nine participants selected 6€, and 17 participants selected 12€). Accordingly, we applied a non-parametric test to assess the effect of regulatory focus priming on the amount of money transferred. A Mann–Whitney-*U*-Test revealed that participants in the prevention prime condition offered significantly less money to their ostensible interaction partner (*Median* = 5.0) than their control group counterparts (*Median* = 8.5, *z* = 1.97, *p <* 0.05). The results of Study 4 thus support the proposition that prevention focus decreases the expression of generalized trust. This finding– which is based on an experimental manipulation of the prevention focus–speaks to the causal nature of the relation between vigilant prevention-focused self-regulation and generalized trust as well as to the fact that situational variations (in extension to chronic differences) in prevention focus affect generalized trust.

Interestingly, alternatively to what we hypothesized, one could argue that the experimental group differed from the control group not only in the activation of a prevention focus, but also with regard to other factors such as negative mood or selfawareness. To address such alternative hypotheses, we assessed potential effects of the experimental procedure (as applied in the current study) on participants' mood state in a separate pre-test (*n* = 39). The item used to assess participants' mood state reads "How do you feel right now?" with response scale end poles labeled (1) *in a good temper* and (9) *in a bad temper*. We did not observe a meaningful effect, *F*(1,37) = 0.002, *p >* 0.96. Accordingly, an alternative interpretation referring to mood seems not particularly likely. Note that the possibility that the procedure applied in Study 4 may have increased experimental participants' self-awareness or general self-focus will be addressed in Study 5 where we made use of a different procedure to induce a prevention focus that is unlikely to trigger a self-focus or self-awareness.

# Study 5

Study 4 complemented the correlational findings observed in Studies 1–3 by situationally inducing a prevention-focused self-regulatory orientation. Although intriguing and consistent, this evidence is silent about the specific mechanisms involved in the relation. Study 5 therefore relies on an experimental paradigm designed to test the specific impact of two crucial elements proposed as characteristic aspects of prevention-focused self-regulation: sensitivity to negative information and the need for security.

In this study, we made use of a fairly subtle 'problem cue' referring to the need for security that is known to be related to prevention-focused self-regulation (Friedman and Förster, 2001). This cue should decrease individuals' trust in another person. Hence, ceteris paribus, the provision of a preventiontriggering cue should result in lower levels of trust. Moreover, we assessed habitual differences in the sensitivity to negative information as a second crucial factor. Given our previous results, one may suppose that both factors – sensitivity to negative information and the need for security – affect individuals' tendency to trust strangers (which would be reflected in main effects of both factors). However, as an alternative one may expect an interplay of both factors such that the fairly subtle problem cue (referring to the need for security) is only effective in individuals with a strong sensitivity to negative information. In our view, such an interplay seems likely to occur given the theoretical and empirical background as discussed in the next section.

# Differential Sensitivity to Problem Cues Signaling Potential Negative Events

As outlined in the description of the proposed consequences associated with each style of self-regulation, RFT holds that individuals are sensitive to different outcomes and consequences depending on the mode of self-regulation that is guiding their regulatory system (Higgins, 1997, 1998). Specifically, RFT predicts that the prevention focus is related to a special sensitivity concerning negative outcomes and consequences, whereas the promotion focus is related to a special sensitivity concerning positive outcomes and consequences. Starting from this proposition, we suggest (based on previous evidence; Keller et al., 2008; cf. Keller, unpublished) that a sensitivity to negative outcomes and consequences involves a specific vigilance regarding environmental cues that signal insecurity and potential losses (i.e., signals indicating that there is a potential for negative consequences in the current situation). And we propose that this special sensitivity to cues that signal insecurity and potential losses can be related to the inverse relation between prevention-focused self-regulation and generalized trust (as observed in the previous studies).

In fact, it appears quite adaptive to lower one's willingness to trust another person in situations where one perceives situational cues indicating potential negative consequences. However, individuals differ in how quick they are in detecting and interpreting cues as signals of insecurity and potential losses – and the style of self-regulation is most likely associated with such individual differences. We obtained empirical evidence in support of this notion (Keller, unpublished) in two independent

correlational studies, which yielded significant positive correlations between prevention (but not promotion) focus scale scores and two instruments designed to assess individual differences in the sensitivity to negative information as reflected in (a) a cognitive tendency to focus on negative information (Noguchi et al., 2006) and (b) a tendency to follow negative emotions (Gasper and Bramesfeld, 2006).

Building on these previous findings documenting a close association between prevention-focused self-regulation and sensitivity to negative information, we assume that individuals high on prevention-focused self-regulation have a strong chronic sensitivity to negative information and therefore are particularly vigilant with respect to subtle cues signaling insecurity, and a potential for negative consequences. Accordingly, we argue that confrontation with a subtle prevention focus problem cue (a subtle trigger of the need for security) is most likely to result in defensiveness and a decreased willingness to express trust among individuals characterized by a special sensitivity to negative information or cues in the environment.

To test these considerations, we conducted an experimental study involving the manipulation of a subtle prevention focus problem cue and we assessed participants' sensitivity to negative information as a critical boundary condition.

# Procedure

Forty students at the University of Mannheim (*M*age = 25.6; 20 women) participated in this study and received 1€ and a chocolate bar as compensation. They were randomly assigned to one of two experimental conditions. All materials were combined in one questionnaire, including the manipulation of the prevention focus problem cue, the scenario version of the trust game paradigm, and a personality measure of sensitivity to negative information5 .

# Manipulation of Prevention Focus Problem Cue

We applied a procedure developed by Friedman and Förster (2001) as a subtle prevention focus priming task. In this task, participants complete a simple paper-and-pencil maze that is designed in a way that there is a cartoon mouse depicted trapped inside the maze and participants are instructed to "find the way for the mouse." Outside the maze, a brick wall is depicted containing a mouse hole. In the prevention focus version of the maze task, an owl is depicted as hovering above the maze presumably ready to fly down and capture the mouse unless it could escape the maze and retreat through the mouse hole. This manipulation reflects a subtle activation of the need for security. Participants in

<sup>5</sup>We assessed sensitivity to negative information after the manipulation and the assessment of trust game responses in order to avoid possible negative sensitivity priming effects that may have emerged if we had asked participants to respond to the items of the scale prior to the manipulation. Given that the measure has been designed as an instrument to assess a stable disposition it seems unlikely that the preceding procedures could affect participants' scores on this scale. Supporting this assumption, we found that the problem cue prime had no impact on participants' sensitivity to negative information scores, *F <* 0.2. Note that a similar posttest measurement strategy has been used successfully in several influential social psychological studies (cf. Devine and Elliot, 1995; Lepore and Brown, 1997; Wittenbrink et al., 1997).

the control group condition received a questionnaire that did not contain this maze task.

# Trust Game Paradigm

The description and procedure of the trust game scenario was parallel to that applied in Studies 2 and 4.

# Sensitivity to Negative Information

We assessed participants' sensitivity to negative information with a measure comprising 10 items of the "attention to negative information scale" developed by Noguchi et al. (2006; a sample item reads "I notice when something is not going well even if it's a trivial thing") as well as the four items taken from Gasper and Bramesfeld's (2006) "following negative emotions" subscale of the "Following Affective States Test" (one sample item reads "I do pay attention to my negative feelings"). The combined 14-item scale was internally consistent with Cronbach's α = 0.86.

# Results and Discussion

In parallel to Studies 2–4, participants' trust game responses were non-normally distributed which is why we conducted nonparametric tests. In an analysis using ordered logit regression we observed no significant main effects whereas a significant interaction effect involving sensitivity to negative information and the problem cue dummy variable emerged, Wald <sup>χ</sup><sup>2</sup> <sup>=</sup> 0.4.42, *p <* 0.04. In additional focused analyses (cf. Aiken and West, 1991; Hayes, 2013) we found that the prevention focus problem cue manipulation had the expected significant effect. Specifically, participants scoring relatively high on the sensitivity to negative information scale (i.e., 1 *SD* above mean) who were confronted with the prevention focus problem cue decided to transfer significantly less money to the ostensible other player as compared to the control condition (*B* = −2.38, *t* = −2.82, *p <* 0.01). In contrast, the prevention focus problem cue manipulation had no significant effect among participants scoring relatively low (i.e., 1 *SD* below mean) on the sensitivity to negative information scale (*B* = 0.26, *t* = 0.30, *p* = 0.76). In sum, the findings obtained in this experimental study suggest that the special sensitivity to negative information that is characteristic of preventionfocused self-regulation is associated with a strong tendency to react with distrust when confronted with a social threat signaling danger and insecurity. This supports the RFT perspective according to which prevention-focused individuals' defensiveness is associated with a particular sensitivity to environmental cues that signal insecurity and a potential for negative consequences. Moreover, the obtained results support the notion that trust game responses vary as a function of situationally induced prevention focus mechanisms. This indicates – parallel to Study 4 – that there is indeed a causal relation between prevention focus mechanisms and generalized trust. This finding complements and extends the correlational results obtained in the Studies 1–3.

In Study 5, we did not observe a main effect of the prevention focus problem cue manipulation. In our view, this is most likely due to the fact that we used a rather *subtle* method to elicit prevention-focused-concerns to which only individuals with a strong sensitivity to negative information are responsive.

# General Discussion

In relating dis-/trust to RFT (Higgins, 1998, 2012a,b), we propose that a prevention-focused mode of self-regulation is inversely related to generalized trust. The obtained findings support this hypothesis. Across a series of studies, participants consistently scored lower on measures of generalized trust the more they were prevention-focused in their self-regulatory orientation. In these studies, we applied different measures to assess individual differences in regulatory focus, we also situationally manipulated prevention-focused mode of self-regulation and tested the effect regarding different measures of generalized trust. The proposed relation was observed irrespective of the type of measurement applied and for both prevention-focused orientation as a disposition as well as a situational factor. Accordingly, the obtained evidence can be considered strong support for the proposed inverse relation between prevention-focused self-regulation and generalized trust.

Going beyond the 'first generation of research' (Zanna and Fazio, 1982) that looked at the question of whether a relation between (prevention-focused) self-regulatory mechanisms and generalized trust can be observed, we also addressed the second-generation question focusing on the specific factors involved in the observed relation (see also Fiedler and Krueger, 2013). Specifically, we tested differential sensitivity to negative information and the need for security as relevant factors and found empirical support for the notion that these specific factors – in combination – play a crucial role and contribute to substantial differences in the tendency to trust unfamiliar others.

The reported research offers several innovative insights. First, in relating RFT to generalized trust we document the crucial impact of self-regulatory mechanisms with regard to a very important social interactive phenomenon that reflects a fundamental basis of social life. Trust is an important factor in interpersonal relationships, intergroup relations, as well as in social organizations. Accordingly, understanding the mechanisms involved in the emergence of trust (or distrust) among unfamiliar persons is extremely relevant, and our work contributes to this field of research by providing insights from a self-regulatory perspective. The role of self-regulatory mechanisms has not been assessed previously in the analysis of generalized trust, and the present contribution therefore represents a new and promising approach. Our analysis reveals the relevance of vigilant self-regulatory mechanisms regarding individuals' behavior in important interpersonal situations.

Second, we put the differential sensitivity assumption entailed in RFT to a critical test in the context of trust and observed strong support. Specifically, we found that participants with a strong sensitivity to negative information were particularly sensitive with respect to a subtle prevention focus problem cue. The findings obtained in Study 5 suggest that prevention-focused individuals are particularly likely to react defensively, specifically with distrust, when confronted with a subtle sign related to danger and insecurity. This particular aspect of prevention-focused self-regulation has not been documented before and thus represents an extension of our knowledge on the specific mechanisms characteristic of this distinct self-regulatory mode.

In sum, the current findings can be viewed as a contribution to the understanding of the role self-regulatory mechanisms play concerning generalized trust, and they enrich our knowledge about the factors that contribute to the expression of trust (or distrust) among unfamiliar persons. As such, the current studies open a new avenue of research for studying trust that incorporates the crucial impact of self-regulatory mechanisms.

#### Avenues for Future Research

We want to note that the analysis of self-regulatory mechanisms with respect to interpersonal trust in the domain of (close) ongoing relationships may result in different predictions and results than those discussed in the present contribution focusing on generalized trust. Specifically, one could speculate that in the context of close relationships individuals with a strong prevention focus could be particularly motivated to override their fears of becoming vulnerable and to trust the persons in their intimate social environment in order to attain security that a trusted other can provide. That is, the desire for security may be satisfied via the establishment of trusting close relationships and this strategy may be particularly attractive to prevention-focused individuals. Consequently, the conclusions offered in the present contribution may not necessarily translate to interpersonal trust in (close) ongoing relationships. Of course, the fact that we prefer to remain silent with respect to the domain of (close) relationships does not imply that we consider self-regulatory mechanisms as irrelevant or unimportant in the domain of ongoing partnerships or close relationships. To the contrary, it appears to be an intriguing topic on the agenda of research on self-regulation and trust that should be addressed in future research.

One could also discuss boundary conditions of the observed relation between prevention-focused self-regulation and generalized trust. Scholer et al. (2010) show that a prevention focus results in increased risk taking under the conditions of a loss. In the present studies, individuals are not in the state of a loss because they have received a specific amount of money to make their trust decisions. However, on basis of the work by Scholer et al. (2010) one could argue that prevention-focused individuals behave differently in a state of a loss by increased trust. Indeed, the possible distinct effect of dealing with a *potential* loss, as considered in the current studies, compared to dealing with a situation of a factual loss as considered by Scholer et al. (2010) is an interesting aspect for future research to address.

# References


#### Implications

Given the strong relevance of generalized trust in everyday life, we would like to end with a discussion of several practical implications of the present work with respect to potential means for increasing individuals' tendency to trust others. In the most general terms, our finding of an inverse relation between prevention-focused self-regulation and trust suggests that individuals' tendency to trust may be enhanced by reducing the strength or relevance of needs and motives that are related to prevention-focused self-regulation. That is, it seems plausible to assume that individuals are more likely to express trust and behave in a trusting way when the needs and motives underlying prevention-focused self-regulation have been satisfied and are therefore not likely to control and govern the self-regulatory orientation in the respective situation. A related important factor is the salience of losses and potential negative outcomes in the situation and hence the accessibility of related thoughts and the activation of a pessimistic mindset. Our findings suggest that trust is more likely to emerge the less individuals are guided by the desire to avoid losses or negative outcomes. Thus, situational cues and inputs that trigger an optimistic perspective (and decrease the accessibility of pessimistic and misanthropic thoughts and related motives) are likely to contribute to an increased tendency to express trust in people.

Another possible strategy to counter prevention-focused individuals' tendency to distrust could be to make use of preventionfocused individuals' respect for normative standards and to emphasize that trusting behavior is normatively appropriate. Based on the notion that prevention-focused individuals are particularly concerned with fulfillment of oughts and responsibilities, such a strategy seems, in theory, particularly meaningful.

In sum, the present work builds the basis for a systematic analysis of factors related to basic self-regulatory mechanisms that contribute to the expression of trust (or distrust) and it has the potential to contribute substantially to our understanding of possible strategies to foster the development and expression of generalized trust. Given that generalized trust is declining in the general population (cf. Paxton, 1999; Robinson and Jackson, 2001), and given that generalized trust is important for organizations and democratic systems to function appropriately (cf. Almond and Verba, 1963; Uslaner, 1999), this seems to be an issue of great relevance.

(Interpersonal Trust) [On some experiences using a German scale to assess interpersonal trust]. *Diagnostica* 30, 198–215.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2015 Keller, Mayo, Greifeneder and Pfattheicher. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# **Attracted to power: challenge/threat and promotion/prevention focus differentially predict the attractiveness of group power**

*Annika Scholl <sup>1</sup> \*, Claudia Sassenrath <sup>1</sup> and Kai Sassenberg 1,2*

*<sup>1</sup> Department of Social Psychology, Institute of Psychology and Education, University of Ulm, Ulm, Germany, <sup>2</sup> School of Science, University of Tübingen, Tübingen, Germany*

Depending on their motivation, individuals prefer different group contexts for social interactions. The present research sought to provide more insight into this relationship. More specifically, we tested how challenge/threat and a promotion/prevention focus predict attraction to groups with high- or low-power. As such, we examined differential outcomes of threat and prevention focus as well as challenge and promotion focus that have often been regarded as closely related. According to regulatory focus, individuals should prefer groups that they expect to "feel right" for them to join: Low-power groups should be more attractive in a prevention (than a promotion) focus, as these groups suggest securityoriented strategies, which fit a prevention focus. High-power groups should be more attractive in a promotion (rather than a prevention) focus, as these groups are associated with promotion strategies fitting a promotion focus (Sassenberg et al., 2007). In contrast, under threat (vs. challenge), groups that allow individuals to restore their (perceived) lack of control should be preferred: Low-power groups should be less attractive under threat (than challenge) because they provide low resources which threatened individuals already perceive as insufficient and high-power groups might be more attractive under threat (than under challenge), because their high resources allow individuals to restore control. Two experiments (*N* = 140) supported these predictions. The attractiveness of a group often depends on the motivation to engage in what *fits* (i.e., prefer a group that feels right in the light of one's regulatory focus). However, under threat the striving to *restore control* (i.e., prefer a group allowing them to change the status quo under threat vs. challenge) overrides the fit effect, which may in turn guide individuals' behavior in social interactions.

**Keywords: regulatory focus, threat, challenge, groups, social power**

# **Introduction**

Imagine you have the possibility to join a new team. You could become a member of a high-power group (e.g., a team of sports referees or a group of supervisors) that is relatively independent and has access to many resources. You could also join a low-power group (e.g., a group of sports players who depend on the referees' decisions, or a team of subordinates continuously evaluated by their supervisors) that is more constrained and depends on others when doing their work. The group's social power has clearly an impact on social interactions within and beyond that group. Under which conditions would these groups appear attractive for you to join?

#### *Edited by:*

*Eva Jonas, University of Salzburg, Austria*

#### *Reviewed by:*

*Winter Mason, Stevens Institute of Technology, USA Daniela N. Kayser, University of Potsdam, Germany*

#### *\*Correspondence:*

*Annika Scholl, Department of Social Psychology, Institute of Psychology and Education, University of Ulm, Helmholtzstraße 18, 89081 Ulm, Germany a.scholl@iwm-tuebingen.de*

#### *Specialty section:*

*This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology*

> *Received: 16 October 2014 Accepted: 20 March 2015 Published: 07 April 2015*

#### *Citation:*

*Scholl A, Sassenrath C and Sassenberg K (2015) Attracted to power: challenge/threat and promotion/prevention focus differentially predict the attractiveness of group power. Front. Psychol. 6:397. doi: 10.3389/fpsyg.2015.00397*

Low-power is often less attractive than high-power (e.g., Sachdev and Bourhis, 1985, 1991; Bourhis, 1994; but see Schmid Mast and Hall, 2003), because it implies lower resource control (Keltner et al., 2003). On the one hand, this lack of control will seem especially inattractive when the demands of a situation seem excessive, such as under threat. However, on the other hand, being in a position of low control allows individuals to keep a "low profile" and to simply follow others' lead. This might appear relatively more attractive under different conditions. Thus, we argue that the degree to which a person is attracted to a low- and a high-power group, respectively, crucially depends on that person's *motivation*.

In this sense, we investigate the impact of two sets of distinct motivational states on the attraction to group power: a prevention vs. a promotion focus (Higgins, 1997, 1998), and a threat vs. challenge state (Blascovich and Tomaka, 1996). Previous research treated these motivational states as being closely related to each other or even overlapping, due to their focus on potential losses (in case of both a prevention focus and threat) or gains (for both a promotion focus and challenge). Indeed, research shows that a prevention focus facilitates threat reactions and that a promotion focus elicits challenge responses (e.g., Keller, 2007, 2012; Keller and Bless, 2008; Seery et al., 2009; Sassenrath et al., unpublished). Inversely, threat elicits behavior in line with a prevention focus (i.e., avoiding errors), whereas challenge evokes behavior in line with a promotion focus (i.e., using opportunities for gains; Seibt and Förster, 2004; Derks et al., 2006; Oyserman et al., 2007).

Despite these commonalities, the present research seeks to provide evidence that prevention and threat, as well as promotion and challenge, do *not* lead to the same but different evaluations of self-relevant social targets. We provide evidence that preventionfocused individuals evaluate low-power groups as being relatively more, and high-power groups as being less, attractive than promotion-focused individuals. In contrast, individuals under threat evaluate low-power groups as being even less and highpower groups as being even more attractive than when they are challenged. The current research has two goals. First, it seeks to contribute to an understanding of how these two sets of motivational states—despite their commonalities—differ in terms of their outcomes in social context. Second, it aims at highlighting how threat affects the attraction to and potentially also the choice of social settings compared to other motivational states.

# **Regulatory Focus and the Attraction to Group Power**

Regulatory focus theory (Higgins, 1997, 1998) posits two distinct motivational systems, which can vary chronically and situationally: The *promotion focus* refers to the regulation of the needs for nurturance, gain, and achievement. This leads to applying eager, approach-related strategies, such as ensuring "hits" and approaching ideal states. In contrast, the *prevention focus* is concerned with regulating the needs for security and safety, which results in the application of cautious strategies, such as ensuring "correct rejections" and avoiding to miss one's obligations. In a promotion focus, events are perceived as gains vs. non-gains. In a prevention focus, events are perceived in terms of non-losses vs. losses. In short, promotion-focused individuals strive toward

taking advantage of opportunities, whereas prevention-focused individuals strive toward not making mistakes.

This preference for self-regulatory strategies influences the evaluation of targets (e.g., events, behaviors, or social groups). Individuals find targets that are compatible with their regulatory focus more attractive (Higgins, 2000; see for example Higgins et al., 2003). This suggests that evaluations of groups should likewise depend on how well individuals expect the group to fit their regulatory needs: Groups that individuals expect to fit their regulatory needs should "feel right" and, thus, be more attractive to join. What does this imply for the attractiveness of a high- and a low-power group?

Social power is defined as asymmetric control over one's own and others' outcomes (Fiske and Berdahl, 2007). This means that *high-power groups* have control over resources (e.g., information, food, money, social appreciation) which others depend on. Highpower groups have a higher capacity to modify another (i.e., a low-power) group's state than *vice versa*. This provides those high in power with relative freedom to interact with others in a way they prefer while being independent from them; in contrast, *lowpower groups* are per definition relatively dependent on others (i.e., on high-power groups; Fiske and Dépret, 1996; Keltner et al., 2003; Fiske and Berdahl, 2007). A low-power group depends more strongly on others to get access to the required resources. Examples for typical high- vs. low-power groups refer to the groups of professors/students at the university, referees/players in sports, or leaders/subordinates in organizations.<sup>1</sup>

In this regard, a *high*-power group allows individuals to engage in behaviors that fit the regulatory needs of nurturance and striving for gain (i.e., taking advantage of opportunities), and individuals are well aware of this: Those high in power likely display promotion-oriented behaviors (Keltner et al., 2003), and individuals likewise anticipate high-power groups to provide them with the opportunity to display promotion-focused behavior (Sassenberg et al., 2007). In other words, high-power groups not only provide individuals with resource control once they have become a group member, but individuals (i.e., non-group members) also *expect* high-power groups to provide such control. In contrast, a *low*power group has limited access to resources and faces more social constraints. As a result, those low in power act more carefully, in a more prevention-oriented way (e.g., taking care not to make a mistake or a negative impression on the high-power group; Keltner et al., 2003). Conversely, individuals also *expect* low-power groups to allow them to demonstrate such prevention-focused behavior (Sassenberg et al., 2007).

To conclude, individuals hold *expectations* with regard to which behavior will likely be demonstrated in high- and low-power groups. As a result, regulatory focus should make groups with a different amount of power more attractive: *High*-power groups

<sup>1</sup>Please note that high-power—as control over own and others' outcomes—not only implies opportunities, but also a certain responsibility for those others' situation (see also Chen et al., 2001; Fiske and Berdahl, 2007; Sassenberg et al., 2012). However, at least in Western cultures, individuals spontaneously associate high-power with the opportunities (rather than the responsibilities) it implies (Zhong et al., 2006; Sassenberg et al., 2012). Hence, in the present research, we focus on this predominant understanding of power as providing especially freedom and opportunity (i.e., means to restore control).

are expected to provide the opportunity to apply promotionoriented strategies which better fit a promotion (than a prevention) focus. Consequently, high-power groups should be relatively more attractive to promotion-focused (than prevention-focused) individuals. Similarly, *low*-power groups are assumed to allow for safety- and security-oriented behaviors. These behaviors do not "feel right" in a promotion focus, but better fit a prevention focus. Hence, low-power groups should be evaluated more favorably in a prevention (than in a promotion) focus.

Empirical evidence supports this idea. Indeed, Sassenberg et al. (2007, see also Sassenberg and Scholl, 2013; Sassenberg et al., 2013) demonstrated that *promotion*-focused individuals (explicitly and implicitly) evaluate high-power groups as being more attractive than prevention-focused individuals. This effect disappears if high-power is no longer associated with opportunities—in other words, if the resulting promotion-oriented behaviors in the group become less likely (i.e., rather when the responsibilities which high-power implies are salient; Sassenberg et al., 2012), providing further support that individuals prefer groups that likely fit their regulatory focus. *Prevention*-focused individuals, in contrast, indeed evaluated low-power groups more favorably than did promotion-focused individuals. In these studies, regulatory focus predicted the attraction to group power among non-group members (i.e., when individuals did not yet belong to the group; Sassenberg et al., 2007, 2013). However, this similarly seems to apply to group-members (e.g., when individuals already belong to a high-/low-power group at work; Sassenberg and Scholl, 2013). While these studies clearly supported the predictions concerning the impact of regulatory focus and group power on group attraction, high-power groups were rated as more attractive than lowpower groups in all studies assessing group attraction explicitly.

In short, this research indicates that a promotion (vs. prevention) focus makes high-power groups even more attractive, whereas a prevention (vs. promotion) focus promotes the attractiveness of low-power groups—because individuals expect that belonging to the respective group allows them to *engage in* their preferential self-regulatory behavior. Do threat and challenge, which are often considered closely related to prevention and promotion focus, result in parallel effects? As we will outline in the following, threat and challenge likely produce distinct effects on the attraction to group power—in particular, because the motivation resulting from threat counteracts the fit effect.

# **Threat vs. Challenge and the Attraction to Group Power**

The Biopsychosocial Model of Threat and Challenge (Blascovich and Tomaka, 1996) describes how individuals respond to goalrelevant situations that require active dealing with the demands resulting from the goal at hand (so-called motivated performance situations). Examples include situations in which individuals perform a test or give a speech in front of an audience. By definition, threat and challenge as such arise when individuals perceive a discrepancy to a desired end-state—referring to the examples above, individuals may seek to perform well on the test or to impress the audience. This discrepancy and the resulting tendency to reduce it (i.e., the striving for a change of the status quo) are not

part of the two regulatory foci. This difference is crucial for the following argument.

When perceiving such a discrepancy, individuals evaluate whether their personal resources (e.g., knowledge, skills) match the situational demands (e.g., the effort required for a task). Accordingly, they respond with *threat* when personal resources fall below the demands of the current situation—that is, when individuals perceive chances to be too low to overcome the discrepancy. For instance, threat occurs when individuals perceive that the demands of a math test exceed their personal skills to solve it, or perceive that their skills to deliver a speech fall below the expectations of the audience. In contrast, if they experience personal resources to match (or exceed) the situational demands, individuals respond with *challenge*: They will likely overcome the discrepancy, even though it might be effortful. In sum, challenged individuals see a high likelihood of gain in the given situation, whereas threatened individuals perceive a high likelihood of loss (cf. Tomaka et al., 1993).

Threat and challenge guide subsequent responses: under challenge, resources seem to suffice and individuals feel ready to master the demands at hand. As an outcome, they show more efficient physiological reactions and higher task performance (than threatened individuals; e.g., Tomaka et al., 1993; Seery et al., 2010). In contrast, under threat individuals feel they cannot cope with the demands at hand and thus experience a lack of control. Consequently, they seek for means to *restore* their control (cf. Staw et al., 1981; Kamphuis et al., 2011; see also Sherman and Cohen, 2006; Sherman and Hartson, 2011 for a similar argument). These means of control restoration can directly address the source of threat (e.g., produce defensive responses toward the "threatening" test). Importantly, however, when such direct means are not available, individuals apply indirect or palliative control restoration (see Jonas et al., 2014). That is, when being threatened in one context (e.g., one's anticipated low performance on an upcoming test), individuals seek to restore control preferably in the context in which the threat comes up. If, however, threat cannot be regulated in such a functional way, individuals switch to *another* context or domain, for example by reaffirming their values in a context or domain that is *unrelated* to the one in which the threat originally came up (e.g., one's skills in another field or one's relationships with friends and family). Reminding oneself of such competencies and values that do not directly address the source of the threat at hand can (at least for some time) reduce the perceived threat (even though it is not functional in the sense that it removes the source of threat; for an overview see Sherman and Hartson, 2011). Taken together, in case of those palliative responses, threat in one context affects evaluations in another context.

We propose that this palliative regulation of threat also applies to the evaluation of high- and low-power groups: under threat, individuals perceive a lack of resources to master the current discrepancy. This elicits efforts to restore control. High-power groups, by definition, provide relative resource control, whereas low-power groups lack such control and depend on others (Fiske and Berdahl, 2007); individuals are well aware of this (Sassenberg et al., 2007). We thus expected that threat (vs. challenge) in one context would affect as how attractive individuals evaluate a high- and a low-power group. A *high*-power group should be even more attractive under threat (than challenge), because such a group allows individuals to restore control (in another context) and thereby reduce the threat; this is less important under challenge when one already experiences sufficient resources. Conversely, *low*-power groups offer low resources (and even more demands). This is particularly unattractive under threat (when resources already seem low), but less unattractive under challenge (when resources seem sufficient and control restoration is less important). In sum, we expect that high-power groups might be even more attractive to threatened (than challenged) individuals, whereas low-power groups should be even less attractive for threatened (than challenged) individuals.

Importantly, our predictions imply differences in attraction to group power between *prevention focus and threat* as well as between *promotion focus and challenge*—despite the previously outlined similarities of these states. Similarly to prevention, threat addresses potential losses and security needs. Similarly to promotion, challenge is concerned with gains and nurturance needs. However, threat/prevention as well as challenge/promotion also differ in several significant aspects. A prevention or promotion focus does not rely on the perception of a discrepancy (e.g., Crowe and Higgins, 1997; Idson et al., 2000; Grant and Higgins, 2003), but they only imply a heightened preparedness for gain- or losssignals, respectively. Individuals are attracted to opportunities to act in line with this preparedness (according to the regulatory fit hypothesis; Higgins, 2000) because they seek to *sustain their current state*—in other words, engage in what "feels right." However, threat and challenge by definition imply the experience of a discrepancy and, thus the striving to *change* the status quo toward the desired end-state (e.g., performing well on the upcoming test). Taken together, individuals in a prevention/promotion focus are attracted to situations sustaining their state, whereas threatened/challenged individuals strive for a change of the status quo. In particular, the striving to regain control under threat should counteract the selection of fitting group that we expect to result from regulatory focus. Different from a prevention focus, threat (compared to challenge) should lead to a preference for high-power groups.

In line with the idea that the effects of prevention and threat as well as of promotion and challenge differ, Sassenberg et al. (2015) demonstrated such differences on the *attentional level*, more specifically, the processing of negative stimuli. Both preventionfocused and challenged individuals direct their attention more to negative (than neutral) stimuli, which is not the case in a promotion focus or under threat: in a prevention focus, in which individuals are more sensitive to losses, attention is directed more toward targets signaling such potential losses (i.e., here negative stimuli that fit a prevention focus, rather than a promotion focus). In contrast, attention is not directed toward negative stimuli under threat, but rather under challenge under which one perceives sufficient resources to deal with negative targets (here, negative stimuli; Sassenberg et al., 2015). While this study focused on differences between these sets of states concerning their influence on the attention to negative targets, the current research investigates differences for the *evaluation* (i.e., valence) of (social) targets, namely how attractive it seems to join a high- or a low-power group.

# **The Present Research**

To sum up, we predicted that regulatory focus promotes attraction to groups that provide the potential to sustain the preferred self-regulatory strategies: a high-power group is more attractive for promotion- (than prevention-) focused individuals, whereas a low-power group is less unattractive for prevention- (than promotion-) focused individuals. Threat (vs. challenge), however, promotes attraction to groups depending on the groups' potential to restore control: a high-power group might be more attractive under threat (than challenge), whereas a low-power group should be less unattractive under challenge (than threat). Hence, under threat group preference should be determined by the extent to which a group can serve as a resource rather than by its regulatory fit to the currently preferred self-regulatory strategies.

Results supporting this prediction would extend prior research in several ways. First, the findings would contribute to an understanding of how individuals under threat regulate their state in a subsequent social context, here, by preferring a group membership that (more or less) allows for restoring control. Second, the present research extends findings on the conditions under which a low- and high-power group is (relatively) more attractive. Third and most importantly, such findings shed further light into potential differences between the two sets of motivational states—promotion/prevention focus and challenge/threat states—and show that these should be distinguished in the social context.

Two experiments tested our predictions. The procedure for measuring attraction to a low- and high-power group closely followed Sassenberg et al. (2007) and was identical for both studies. Experiment 1A focused on the impact of promotion/prevention focus in order to replicate earlier findings on attraction to a lowand a high-power group (Sassenberg et al., 2007). Implementing the identical group context and group attraction measure, Experiment 1B tested the impact of threat/challenge. Across the two studies, regulatory focus and threat/challenge were manipulated independently of the group context, in order to rule out potential demand effects.

# **Experiment 1A**

# **Material and Methods** Participants and design

Sixty undergraduates (38 female, 22 male; *M*age = 22.45, SD = 2.70; range 18–31 years) participated in an experiment with a 2 (Regulatory focus: promotion vs. prevention) *×* 2 (Group power: high vs. low) design. Regulatory focus was manipulated between participants, whereas Group power was a repeated measures factor. Participants took part in this study on campus in exchange for some candy.

# Procedure

Participants completed a paper–pencil questionnaire, supposedly comprising "pretest materials" for two unrelated studies. In fact, the questionnaire included (1) the regulatory focus manipulation (Friedman and Förster, 2001) and (2) the measure of attraction to a powerful and a powerless group (Sassenberg et al., 2007). To manipulate situational regulatory focus, participants lead a mouse through a maze, either toward a mouse hole to prevent it becoming an eagle's prey (i.e., security-related strategies were activated; prevention focus condition; *N* = 30) or toward a large piece of cheese (i.e., nurturance-related strategies were activated; promotion focus condition; *N* = 30).

As "unrelated second part" of the study, participants completed questions on the "perception of groups." Here, they were informed that the researchers were interested in their spontaneous evaluation of a group. More specifically, they read that though it can be difficult to make a general judgment of a group and its members, individuals usually have a rough idea of how others think and act. They were then instructed to think of and afterward indicate their spontaneous evaluation of a group that included "members of a large company who have high-power and average status" (highpower group) and a group that included "members of a large company who have low-power and average status" (low-power group). The order of the two groups was counterbalanced. There were no interactions with order (i.e., high-power vs. low-power group presented first; all *F*s *<* 1.22, *p*s *>* 0.275); hence, we do not further discuss this factor in the following. The exact same power manipulation has been used before (Sassenberg et al., 2007, 2012). In those articles, it produced the same results than other power manipulations and can therefore be considered as well validated.

Two items assessed attraction to each group, respectively ("If you imagine being a member of this group, how do you feel?"; 1 = *bad* to 7 = *good*; "How attractive is this group to you?"; 1 = *not at all* to 7 = *very much*; *r*(60)high-power-group = 0.73, *p <* 0.001; *r*(60)low-power-group = 0.77, *p <* 0.001). One item served as manipulation check for perceived power of each group ("Which of the following attributes fits to this group?" on a 8-point semantic differential from "*not at all powerful*" to "*very powerful*"). Finally, participants indicated their demographics, were thanked, debriefed, and compensated.

### **Results**

#### Checks

Results yielded an effect of Group power on perceived power, indicating that participants rated the low-power group as being less powerful (*M* = 2.83, SE = 0.24) than the high-power group (*M* = 7.35, SE = 0.25), *F*(1,58) = 103.95, *p <* 0.001, η 2 <sup>p</sup> = 0.64. There was no interaction of Group power with Regulatory focus and/or Order on perceived power (all *F*s *<* 1, *p*s *>* 0.408). Hence, the manipulation of Group power was successful.

#### Attraction to group power

We predicted that prevention-focused participants would find the low-power group relatively more attractive than promotionfocused participants, whereas promotion-focused participants would find the high-power group even more attractive than prevention-focused participants.

A 2 (Regulatory focus: promotion vs. prevention) *×* 2 (Group power: low vs. high) mixed model analysis of variance with repeated measurement on the last factor yielded a main effect of Group power, *F*(1,58) = 23.81, *p <* 0.001, η 2 <sup>p</sup> = 0.29, but not of Regulatory focus (*F <* 1). Overall, participants evaluated the high-power group as more attractive (*M* = 5.08, SE = 0.23) than the low-power-group (*M* = 3.38, SE = 0.20). This effect was qualified by the predicted Regulatory focus *×* Group power interaction, *F*(1,58) = 7.18, *p* = 0.010, η 2 <sup>p</sup> = 0.11.

The high-power group was more attractive than the lowpower group in the promotion focus condition, *F*(1,58) = 28.57, *p <* 0.001, η <sup>2</sup> = 0.33, but not (significantly so) in the prevention focus condition, *F*(1,58) = 2.42, *p* = 0.125, η <sup>2</sup> = 0.04. More importantly, as expected, the low-power group was more attractive to prevention-focused (*M* = 3.82, SE = 0.28) than promotionfocused participants (*M* = 2.95, SE = 0.28), *F*(1,58) = 4.86, *p* = 0.031, η <sup>2</sup> = 0.08, MD(mean difference) = 0.87, CI95%[0.08, 1.65]. In contrast, the high-power group was more attractive to promotion-focused (*M* = 5.58, SE = 0.33) than preventionfocused participants (*M* = 4.58, SE = 0.33), *F*(1,58) = 4.59, *p* = 0.036, η <sup>2</sup> = 0.07, MD = 1.00, CI95%[0.07, 1.93], see **Figure 1**.

#### **Discussion**

This experiment supported our assumptions and is in line with earlier research (Sassenberg et al., 2007) that prevention-focused individuals are comparatively more attracted to low-power groups than promotion-focused individuals, whereas the reverse is true for high-power groups. Hence, although the high-power group was—in the present research as well as in earlier research using a similar paradigm (Sassenberg et al., 2007, Studies 1–3)—perceived as more attractive than the low-power group, this especially applied to promotion-focused individuals. Importantly, the findings also indicate that prevention-focused individuals do not find low-power groups as unattractive as promotion-focused individuals do, presumably because these groups fit their security-oriented strategies (for evidence, see Sassenberg et al., 2007). Notably, the sample size of the current study is not ideal and, thus, the study is somewhat underpowered. However, this might be considered as a less severe limitation, as the current effect replicates earlier findings (Sassenberg et al., 2007, 2013; Sassenberg and Scholl, 2013).

To extend these findings, Experiment 1B examined how the experience of threat (vs. challenge)—rather than a prevention or promotion focus—predicts individuals' attraction to group power. Following the design from Experiment 1A, we thus again measured attraction to a high- and low-power group with the

identical procedure as before, but this time after inducing a threat vs. challenge state.

# **Experiment 1B**

## **Method**

#### Participants and design

Eighty undergraduates (51 female, 29 male; *M*age = 22.06, SD = 2.00; range 18–27 years) participated in an experiment with a 2 (threat vs. challenge) *×* 2 (Group power: high vs. low) design. The first factor was manipulated between participants, whereas the second factor constituted a repeated measures factor. Participants were compensated on campus with a candy.

# Procedure

The procedure of this experiment (from recruiting and location over materials to compensation and debriefing) was exactly the same as in Experiment 1A, except for the fact that the regulatory focus manipulation was replaced by a threat vs. challenge manipulation.

As challenge/threat manipulation, participants completed an open-ended questionnaire on exam preparation. They imagined preparing for an exam in a class that they had already performed once before. They either read that they were dissatisfied with their prior result and were now preparing for a "free-shot" (i.e., the possibility to improve their prior grade; *challenge* condition; *N* = 40) or they read that they had failed the class and were now preparing for their "final chance" (i.e., if they failed again, they would be excluded from the study program; *threat* condition; *N* = 40).<sup>2</sup>

Order of group power was again counter-balanced. There were no interactions with order (i.e., high-power vs. low-power group presented first; all *F*s *<* 2.17, *p*s *>* 0.145). The two measures for group attraction showed good internal consistency [2 items each; *r*(80)high-power-group = 0.61, *p <* 0.001; *r*(80)low-power-group = 0.58, *p <* 0.001].

#### **Results** Checks

We first tested if the low-power group was indeed perceived as less powerful than the high-power group. Participants rated the low-power group as having less power (*M* = 3.24, SE = 0.18) than the high-power group (*M* = 7.46, SE = 0.15), *F*(1,77) = 232.23, *p <* 0.001, η 2 <sup>p</sup> = 0.75. Varying degrees of freedom result from missing data from one participant. There were no interactions of Group power with threat/challenge and/or with Order on perceived power (all *F*s *<* 1, *p*s *>* 0.383), indicating that the manipulation of Group power was successful.

#### Attraction to group power

We expected that the *low*-power group should be evaluated as even less attractive under threat than challenge, whereas the *high*-power group might be more attractive under threat than challenge.

A 2 (threat vs. challenge) *×* 2 (Group power: low vs. high) mixed model analysis of variance with repeated measurement on the last factor showed a main effect of Group power, *F*(1,78) = 56.94, *p <* 0.001, η 2 <sup>p</sup> = 0.42, but not of threat/challenge state (*F <* 1). The high-power group was again perceived as more attractive (*M* = 5.28, SE = 0.20) than the low-power group (*M* = 3.60, SE = 0.14). This effect was qualified by the expected threat/challenge state *×* Group power interaction, *F*(1,78) = 8.03, *p* = 0.006, η 2 <sup>p</sup> = 0.09. The high-power group was overall more attractive than the low-power group in the challenge, *F*(1,78) = 11.11, *p* = 0.001, η <sup>2</sup> = 0.13, and the threat condition, *F*(1,78) = 53.84, *p <* 0.001, η <sup>2</sup> = 0.41. Supporting our predictions, however, the high-power group tended to be even more attractive under threat (*M* = 5.61, SE = 0.29) than under challenge (*M* = 4.95, SE = 0.29), *F*(1,78) = 2.70, *p* = 0.104, η <sup>2</sup> <sup>=</sup> 0.03, MD <sup>=</sup> 0.66, CI95%[*−*0.14, 1.47], though this effect did not reach conventional levels of statistical significance. In contrast, as predicted, the low-power group was rated as even less attractive under threat (*M* = 3.30, SE = 0.19) than under challenge (*M* = 3.90, SE = 0.19), *F*(1,78) = 4.82, *p* = 0.031, η <sup>2</sup> = 0.06, MD = 0.60, CI95%[0.06, 1.14], see **Figure 2**.

#### Additional analyses combining data from Experiments 1A and 1B

Taken together, the findings indicated that challenge and prevention focus, as well as threat and promotion focus, yield a similar pattern regarding the attraction to group power. This would result in the prediction of a three-way interaction between the respective Motivational State (regulatory focus vs. challenge/threat),

<sup>2</sup>Notably, the threat manipulation refers to "failure" or "loss," and the challenge manipulation (indirectly) refers to "gains." To rule out the possibility that the threat/challenge manipulation had unintentional side effects on participants' regulatory focus, we pretested the manipulations with a non-overlapping undergraduate sample (*N* = 49; 36 female, 13 male; *M*age = 22.18, SD = 2.94; range 18–31 years). The threat/challenge manipulation neither affected selfreported prevention focus (*M*challenge = 4.58, SD = 1.40; *M*threat = 4.67, SD = 1.23), *t*(47) = 0.24, *p* = 0.405, *d* = 0.07, nor self-reported promotion focus (*M*challenge = 4.99, SD = 0.76; *M*threat = 5.13, SD = 0.52), *t*(47) = 0.79, *p* = 0.216, *d* = 0.22, but it affected self-reported threat (*M*challenge = 4.00, SD = 1.96; *M*threat = 5.00, SD = 1.96; all on 7-point Likert scales), *t*(47) = 1.79, *p* = 0.040, *d* = 0.51 (all one-tailed). Additionally, the threat/challenge manipulation did not affect the number of words participants wrote down in the openended questionnaire on exam preparation (*M*challenge = 44.67, SD = 24.63; *M*threat = 42.64, SD = 19.81), *t*(47) = 0.32, *p* = 0.752, *d* = 0.09. This indicates that participants in both conditions exerted a similar level of effort and elaboration while completing the manipulation.

the respective Gain-/Loss-Focus (promotion/challenge: gains; vs. prevention/threat: losses), and Group power (low vs. high). Such an interaction would further support the assumption that in the case of attraction to group power, challenge/promotion focus and threat/prevention focus can have opposite implications, respectively.

To test this, we combined the two data sets from Experiments 1A and 1B, which comprised of the identical dependent measure. A 2 (Motivational State: regulatory focus vs. threat/challenge) *×* 2 (Gain-/Loss-Focus: gains under promotion/challenge vs. losses under prevention/threat) *×* 2 (Group power: low vs. high) mixed model analysis of variance with repeated measurement on the last factor was conducted. Please note that in this analysis, the factor Motivational State represents the two separate experiments (Experiment 1A: regulatory focus vs. Experiment 1B: threat/challenge). This analysis showed a main effect of Group power, *F*(1,136) = 72.82, *p <* 0.001, η 2 <sup>p</sup> = 0.35, which was qualified by the expected three-way interaction, *F*(1,136) = 15.59, *p <* 0.001, η 2 <sup>p</sup> = 0.10. We explored this 3-way interaction with two separate 2 (Motivational State: regulatory focus vs. threat/challenge) *×* 2 (Group power: low vs. high) ANOVAs for Gain-/Loss-Focus, respectively.

Indeed, the Motivational State *×* Group power interaction was significant for gains (promotion vs. challenge), *F*(1,68) = 6.83, *p* = 0.011, η 2 <sup>p</sup> = 0.09, and also for losses (prevention vs. threat), *F*(1,68) = 9.16, *p* = 0.003, η 2 <sup>p</sup> = 0.12 (see **Table 1**, for an overview of all means and SDs).

There were no interactions between Motivational State and Group power on the attraction to group power or on the power manipulation check, ruling out any differences in the effect of group power between the two experiments. In sum, this indicates that challenge vs. a promotion focus, and threat vs. a prevention focus, differentially predict attraction to a group, depending on this group's power.

#### **Discussion**

Replicating the previous study, the high-power group was overall more attractive than the low-power group. However, as predicted this pattern depended on individuals' motivational state: Under challenge, the low-power group was relatively more attractive than under threat. Combined with the results from Experiment 1A, this in sum indicates that for both prevention-focused and challenged individuals, low-power groups are relatively more attractive than for promotion-focused and threatened individuals.

Moreover, the high-power group—on a descriptive level appeared more attractive under threat than under challenge. Though this pattern has to be treated with caution, it may imply that a high-power group is particularly attractive to promotionfocused (see Experiment 1A) and threatened individuals (more so than prevention-focused and challenged individuals).

# **General Discussion**

Previous research indicates that threat/challenge and prevention/promotion focus, respectively, are related in several outcome areas, such as avoiding errors/taking advantage of opportunities for gains (Seibt and Förster, 2004; Derks et al., 2006; Oyserman



et al., 2007). Based on the example of attraction to group power, the present research indicates that these sets of concepts do, however, differ in the level of evaluating social targets (i.e., evaluating a potential membership in social groups). We predicted this difference between the two sets of states as an exemplary outcome of one fundamental motivational difference between them: the tendency to *engage in* what "feels right" under regulatory focus (Higgins, 2000) and thus sustain the current state, vs. the tendency to overcome a discrepancy by *restoring control* under threat (vs. challenge). On a cautionary note, we investigated these effects in separate studies (rather than one overall study). However, procedures of the studies are highly similar and results from an analysis across both studies confirmed the predicted pattern. Therefore, it is, in our opinion, appropriate to interpret comparisons of results across both studies.

## **Implications for Threat/Challenge and Regulatory Focus**

Our findings are in line with earlier research on the differential effects of promotion and threat as well as prevention and challenge on attention to negative stimuli (Sassenberg et al., 2015). *Prevention* and *challenge* can lead to similar outcomes (i.e., evaluation of low-power groups and attention to negative stimuli), but other research has also shown that prevention and *threat* are closely related (e.g., Seibt and Förster, 2004; Oyserman et al., 2007). An essential difference between these sets of findings—showing differences or similarities between these motivational states—lies in the *self-relevance* of the assessed outcome.

Studies finding a close relation between prevention and threat (as well as promotion and challenge) mostly test their effect on cognitive styles while processing *neutral* material. Here, threat and a prevention focus result in more local, thorough, and careful information processing. In contrast, challenge and a promotion focus result in more global, flexible, and risky information processing (e.g., Blascovich et al., 1999; Friedman and Förster, 2001). Accordingly, threat/prevention focus and challenge/promotion focus should produce similar responses when *neutral* (i.e., not self-relevant) outcomes are investigated.

In contrast, studies investigating *self-relevant* outcomes—such as perceived valence or the processing of targets otherwise relevant to the self (e.g., group membership, social roles, impressions made on others, performance feedback) appear to produce differential results. Here, promotion and prevention focus direct attention to and foster a positive evaluation of targets that are *congruent* to the judge's focus: gain-related stimuli in a promotion focus and loss-related stimuli in a prevention focus (see regulatory fit principle; Higgins, 2000). In contrast, threatened individuals direct attention to and foster a positive evaluation of targets that are suitable for (palliative) control restoration: individuals prefer stimuli *opposing* their threat/challenge state (i.e., positive stimuli under threat and negative stimuli when being challenged). Challenged individuals want to master the situation while perceiving sufficient control. Hence, they are ready to "fight barriers" in coping with the demands at hand. Threatened individuals want to restore control and thereby reduce the threat. Thus, resources and opportunities to regain control are particularly attractive. In that sense, the striving to reduce threat is stronger than the search for fitting social context that allow for effortless sustaining of the current strategy. Regulatory focus leads to the striving for fitting behavioral opportunity, whereas threat (and challenge) render the outcomes of choices such as control more relevant.

These explanations on a process level are certainly based on the current data—they are yet speculative and require further research. Nonetheless, the current findings provide evidence for the general idea that promotion/challenge and prevention/threat do not coincide in their effects when it comes to valence outcomes (rather than information processing styles or performance).

# **Implications for Threat in Social Interactions**

What do the results imply for individual responses to threat in social context? Previous research showed that threat can enhance group cohesiveness (e.g., Sherif and Sherif, 1953; Sherif et al., 1961) and lead to avoidance or rigidity ("freezing"; see Scherer et al., 2004; Mendes et al., 2007), for instance, in terms of adjusting from initial anchors or decisions (Staw et al., 1981; Kassam et al., 2009; Kamphuis et al., 2011; de Wit et al., 2012). Similarly, a related line of research demonstrates that under threat, individuals cling to a high-power *in-group* as a means to restore their control (Fritsche et al., 2008). Extending these findings, our results indicate how individuals under threat (vs. challenge) are attracted to group membership in the first place—that is, to groups they are not yet a member of: threatened individuals strive less for belonging to low-power groups (and potentially more for belonging to high-power groups) than do challenged individuals. They prefer contexts in which they, rather than others, would have control within the social interaction.

Depending on whether the control provided by a high-power group is relevant to the threat at hand or not, striving for member-

# **References**


ship in such a group could be a comparatively effective strategy to restore control: it could better reduce the stress resulting from the threat than "being frozen" at the status quo. Indeed, the findings contribute to an understanding how the possibility of group membership may serve as a means to regain perceived control under threat in social contexts.

Threat, however, may not only arise on the *individual* level (e.g., in anticipation of an upcoming test), as considered here. It can also result from the *social* context itself—such as when comparing one's performance with that of superior others, which can threaten one's self-esteem. This type of threat may likewise determine which group individuals are especially attracted to.

On a more abstract level, the present research indicates that motivational states shape subsequent cognitive evaluations of the social context—here, the valence of group power. As to the impact of threat on social interaction, the current research implies that threatened individuals may seek social roles and social groups that provide them with the means to restore control again. By becoming a member of a high-power group, a threatened individual regains control through being in the dominant social role. Similarly, threat might also lead to a preference for a stable (rather than potentially changing) social role (appointed on a long-term basis rather than elected for a limited time) or for working on an independent (rather than an interdependent) task. These are fruitful approaches for future research which could also be extended beyond the (university) context studied here (e.g., employee samples). In short, threatened individuals (but not those in a prevention focus) seem to prefer social interactions that are predictable (i.e., come with high control).

To conclude, individuals at times face the choice of joining a low-power or a high-power group. While high-power is commonly more attractive than low-power, the respective preference also depends on individuals' motivation. More specifically, the attraction to a high- and low-power group can depend on whether the group offers the potential to engage in one's preferred strategies (i.e., fits one's regulatory focus) or the potential to restore control (i.e., especially under threat) to master the situation at hand and reach a desired end-state. The findings thereby contribute to an understanding why one and the same group may be differentially attractive to be joined, depending on an individual's current motivational state.


*Psychology*, Vol. 7, eds W. Stroebe and M. Hewstone (New York, NY: Wiley), 31–61.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2015 Scholl, Sassenrath and Sassenberg. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Threat to Freedom and the Detrimental Effect of Avoidance Goal Frames: Reactance as a Mediating Variable

#### Daniela Niesta Kayser<sup>1</sup> \* † , Verena Graupmann<sup>2</sup>† , James W. Fryer<sup>3</sup> and Dieter Frey<sup>4</sup>

<sup>1</sup> Division of Social Psychology, University of Potsdam, Potsdam, Germany, <sup>2</sup> Department of Psychology, DePaul University, Chicago, IL, USA, <sup>3</sup> Department of Psychology, State University of New York at Potsdam, Potsdam, NY, USA, <sup>4</sup> Department of Social Psychology, Ludwig Maximilian University of Munich, Munich, Germany

#### Edited by:

Christina Steindl, University of Salzburg, Austria

#### Reviewed by:

Eric Donald Wesselmann, Illinois State University, USA Rex A. Wright, University of North Texas, USA Louisa Pavey, Kingston University, UK

> \*Correspondence: Daniela Niesta Kayser niesta@uni-potsdam.de

†These authors have contributed equally to this work.

#### Specialty section:

This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology

Received: 29 January 2015 Accepted: 15 April 2016 Published: 18 May 2016

#### Citation:

Niesta Kayser D, Graupmann V, Fryer JW and Frey D (2016) Threat to Freedom and the Detrimental Effect of Avoidance Goal Frames: Reactance as a Mediating Variable. Front. Psychol. 7:632. doi: 10.3389/fpsyg.2016.00632 Two experiments examined how individuals respond to a restriction presented within an approach versus an avoidance frame. In Study 1, working on a problem-solving task, participants were initially free to choose their strategy, but for a second task were told to change their strategy. The message to change was embedded in either an approach or avoidance frame. When confronted with an avoidance compared to an approach frame, the participants' reactance toward the request was greater and, in turn, led to impaired performance. The role of reactance as a response to threat to freedom was explicitly examined in Study 2, in which participants evaluated a potential change in policy affecting their program of study herein explicitly varying whether a restriction was present or absent and whether the message was embedded in an approach versus avoidance frame. When communicated with an avoidance frame and as a restriction, participants showed the highest resistance in terms of reactance, message agreement and evaluation of the communicator. The difference in agreement with the change was mediated by reactance only when a restriction was present. Overall, avoidance goal frames were associated with more resistance to change on different levels of experience (reactance, performance, and person perception). Reactance mediated the effect of goal frame on other outcomes only when a restriction was present.

#### Keywords: freedom restriction, goal frames, avoidance, approach, reactance, self threat, change

# INTRODUCTION

The antagonism between the inevitability of change and resistance to change is deeply ingrained in human thinking and acting. A general aversion to change has been attributed to a fundamental motivation to favor previously made choices and to attach ourselves to courses of action, once they are initiated (Cialdini et al., 1995). In everyday life, we often encounter demands for change of our behavior, e.g., adapting the syllabus of a research methods course to follow new department guidelines. The motivation and likelihood with which we are going to implement the requested change to our syllabus depends upon the degree to which we perceive such a demand as a threat to our freedom. Our inclination to make the change may also be influenced by whether the demand has been communicated as aimed at achieving better learning outcomes or avoiding negative learning outcomes. Does communication style additionally influence the perceived threat

to freedom in the request? And if so, are there requests that are more restricting than others? The present experiments were designed to examine whether and how communication style framed with approach and avoidance goals affects the impact of freedom restrictions.

# Change and Psychological Reactance

Within the field of social psychology the concept of freedom is mostly looked at in the context of the individual's control and choice. Reactance theory (Brehm and Brehm, 1981) in particular emphasizes the importance of individual freedom and behavioral choices and defines conditions under which people react against attempts to control their behavior and eliminate their freedom of choice. Freedom in terms of reactance theory is defined as a person's belief to be able to engage in a certain behavior and to decide on the type of behavior, as well as how the behavior is performed and when. Thus, reactance theory proposes that when we believe to be free to choose a course of action, we experience reactance if that freedom is eliminated or threatened. According to Brehm (1966), psychological reactance is an aversive motivational state directed toward the re-establishment of freedom, even if that resistance is not associated with optimal outcomes for the person. The potentially negative consequences of reactance are mirrored in the fact that the term is adapted from the field of electrical engineering and means 'blind resistance.' Reactance manifests itself in an increased desire to engage in the restricted behavior or actual attempts to engage in it, or an active refusal to engage in the prescribed behavior, e.g., students may decrease their own effort, patients may not adhere to treatment plans, or employees may resist the implementation of a new strategy. When Amy asks her friend Rachel to accompany her to the concert of their favorite band and uses the message "you must come with me," reactance is likely aroused by forcing a desired outcome on the person; likewise, reactance is aroused as well by eliminating access to a desired outcome such as when parents forbid their daughter to attend the concert: "you must not go to this concert" (Wicklund, 1974). Where freedom is threatened by social pressure, reactance evokes the tendency to reassert the lost freedom by the individual to resist that pressure, e.g., by reducing one owns efforts to slow down the proposed goal or by disagreeing with the communicator to weaken the implementation of a project (e.g., Dillard and Shen, 2005; Silvia, 2006). Reactance occurs in response to a message that implies a forceful attitude and makes the person want to approach a desired goal, which will lead the person to resist this imposition or the way the request is imposed upon them. Likewise, reactance occurs when the message contains a perceived threat and prohibits a specific behavior or demands the person to stay away from a desired goal. This will likely lead the person to resist this prohibition or the way the prohibition is phrased.

In its origin, reactance theory examined how people can be persuaded, attitudes can be changed, and consumer behavior can be influenced. The theory extends to other aspects of individual behavior that involve motivation following uncontrollable events, in particular, achievement motivation and task performance. In contrast to Seligman's (1975) theory of learned helplessness, which predicts a decrease in subsequent motivation and performance a loss of control over outcomes is experienced, whether or not it is aversive, reactance theory predicts increased motivation and performance. In order to accommodate these seemingly opposing predictions, Wortman and Brehm (1975), proposed an integrative model: When perceived control over an outcome is high, psychological reactance results in enhanced motivation to achieve and perform well when having to overcome resistance. However, when participants' expectation of control is low, subsequent performance should deteriorate when facing resistance. In line with this model and based on the work by Folkman and Lazarus (1985), experienced stress induced for example by particular demands of a task, can be evaluated as challenging or threatening. Perceptions of challenge enhance performance, whereas perceptions of threat inhibit performance. Moreover, introducing change in the rules of the task at hand has shown to have a negative effect on participants' performance. In a study by Drach-Zahavy and Erez (2002), threat was operationalized in terms of negative outcomes and an interview with the five worst performers, a procedure that set a focus on failure (as opposed to the challenge condition, where the focus was set on success). This procedure deemed to be particularly detrimental in a change situation in which the participants had to use a different strategy at round 2 compared to round 1.

Particularly owing to the work on stereotype threat, the link between existence of a threat and depletion of self and lowered performance (e.g., Baumeister, 1984; Steele and Aronson, 1995; Frederickson et al., 1998) is well documented. Similarly, a threat induced by a demand to change should consume resources, divide attention away from the task, and therefore disrupt and diminish an individual's mental activity. A change request in the academic context, we argue, can be experienced as a threat, and if it is, it should impair intellectual performance.

Reflected in the current literature on social influence, requests to change are especially pertinent to questions of persuasive communication. It might represent a threat to Rachel, when Amy communicates the message that they 'must' visit a concert of a band that they both like, even if Rachel is a fan of this band herself. In fact, one of the basic claims of reactance theory is that high-pressure communicators are likely to be seen as threatening to personal freedom (Wicklund, 1974; Brehm and Brehm, 1981). In a study requesting the change for the use of washing detergents from containing phosphates to not containing phosphate, one indication of consumers' reluctance to participate in this change of habit was expressed in their beliefs that washing detergents without phosphate would be less efficient (Mazis et al., 1973).

Research examining the role of state-(Rains and Turner, 2007) trait-(Pavey and Sparks, 2009) reactance in how persuasive information on health risks is evaluated and affects intentions to engage in health behaviors has shown a negative relationship between reactance and the intended effect of such messages. Here reactance has also been associated with a perception of threatened freedom, further supporting the original theory in the domain of persuasion.

Derogating the object is one possible avenue to restore freedom, while derogating the source of threat (Kohn and Barnes, 1977) is another possible outcome of reactance albeit indirectly. This is consistent with the initial theoretical reasoning by Brehm

(1966) arguing that reactance could not be measured directly, but instead be inferred by its effects. Since the beginning of the work on reactance, reactance has been implicitly and explicitly defined in many different ways. Among other accounts, reactance has been viewed as cognitive (e.g., Petty and Cacioppo, 1986), i.e., measurable through thoughts that can be listed in selfreport techniques and operationalized in terms of counterarguing. Since reactance evokes anger and is associated with the experience of hostile feelings (Wicklund, 1974), it has also been considered an emotion (e.g., Dillard and Mejenders, 2002).

Finally, whether the motivation to restore freedom following an uncontrollable event is set in the domain of achievement or persuasion, reactance as a motivated psychological state will likely guide attention, influence thought process (Derryberry and Tucker, 1994), stimulate feelings, and direct behavior (Kuhl, 1986). Change induced in an aversive manner may therefore be reflected in reactance shown in cognitive, affective, and behavioral outcomes.

# Change and Goal Framing

Change implies a new direction and new goals that can imply the elimination of options and the reduction of a set of perceived freedoms. A restricted freedom may yield greater aversion when pursued with a certain type of goal. Avoidance goal frames focus on trying to avoid or stay away from a negative outcome or a negative psychological situation. Examples of avoidance goal frames are "Try to avoid doing poorly on a test," "Try not to be disloyal to friend," and "Try to avoid smoking a cigarette." Approach goals, on the other hand, use positive, desired possibilities such as "Try to do well on a test," "Try to be a loyal friend," and "Try to become smoke-free," which typically leads to favorable psychological processes and outcomes, such as perceptions of personal progresses or competence in goal pursuit (Elliot and Sheldon, 1997; Elliot et al., 1997). The approachavoidance distinction is integral to the understanding of how individuals deal with changes, as requests to change always also imply a change in motivational tendency to initiate a shift in a person's course of action. The type of goal that is pursued in light of a restriction leads to a number of important implications for the experience of threat.

# Goal Framing and Reactance

Looking at goal framing and how it relates to a person's sense of freedom and threat to freedom, the experience of the different goal frames in terms of reactance should vary according to the respective goal frame: Approach goal frames suggest a change in course of action that can be a potential gain to the status quo, i.e., no change in course of action will lead to a neutral outcome, whereas a change leads to a positive outcome, and therefore change can be interpreted as optional. In this vein, a sense of behavioral freedom is maintained and approach goal frames associated with a restricted freedom should therefore yield less psychological reactance.

Avoidance goal frames, on the other hand, suggest a change in course of action that is required in order to avoid experiencing a deterioration of the status quo, i.e., no change in course of action will lead to a negative outcome, and therefore the change is not optional. Therefore setting an avoidance goal frame when told to change has the potential to increase the perception of restriction to behavioral freedom. A change embedded in an avoidance goal frame might thus be associated with more psychological reactance than when embedded in an approach goal frame. Moreover, the inherent focus on negative possibilities in avoidance goal frames has shown to be associated with a host of aversive psychological processes, including perceptual, attentional, mental control, emotional and behavioral processes, for example experiencing fear of failure, anxiety, and wanting to escape from the goal-relevant situation (for example, Wegner, 1994; Elliot and McGregor, 1999). This link between avoidance goal frames and aversive psychological processes can be interpreted as a threat to a central self-motive (Graupmann et al., 2013), which may result, among others in a tendency to react against the implied restriction of freedom.

Research has yet to directly examine the link between the restriction of a freedom, avoidance goal pursuit and the emergence of reactance. If freedom was threatened by implication, the emergence of reactance when avoidance but not approach goal frames are salient would suggest that this type of goal framing is experienced as a self-threat. Along with this idea, research on the self posits that self-regulation draws on a limited common pool of resources (Baumeister, 1998). Since regulating the self is difficult, subsequent acts of self-regulation entail a state of ego-depletion (Baumeister, 1998). An individual that engages in goal-directed behavior and that monitors his/her goal progress expends resources. Furthermore, research on goal pursuit documents that the pursuit of some types of goals is more depleting than others (Oertig et al., 2013). In particular, avoidance compared with approach goal frames have been shown to deplete self-regulatory resources, which is manifested in perceptual, attentional, emotional, and behavioral deficits. This grounding in negative possibilities may produce aversive psychological processes that have negative consequences such as resistance, impairment of performance, and disagreement with social influence; processes much as experienced following a perceived elimination of freedom.

According to this goal-pursuit-reactance approach to selfthreat, an avoidance frame in a restricting situation should not only be detrimental to the willingness to comply with the request, but also be more self-impairing than an approach frame: avoidance goal frames have been repeatedly found to lead to a less deep processing of tasks, to preparing things in a more disorganized fashion, to worrying more about the own competence and to elicit more negative emotional reactions (Elliot and McGregor, 1999).

People are generally motivated to approach positive outcomes and avoid negative outcomes. However, to the best of our knowledge, only limited research has been conducted on the interplay of the impact of goal framing on performance and perception processes when a freedom is restricted. Integrating predictions from reactance theory and approach avoidance accounts into a larger theoretical framework enables us to test the impact of an avoidance goal frame when a person is told to change and, more importantly, through the experience of reactance of

this person and her subsequent task performance and person perception processes.

Combining these two lines, perceptions of and requests to change are a threat due to (1) the potential to restrict one's freedom as predicted by reactance theory and (2) the threatening nature when the change is framed in terms of avoidance goals.

# Overview of the Present Research

In the present research, we investigate the link between goal framing and psychological reactance as an indication of threat to self. On the basis of both theory and prior empirical work on goals and resistance to change, we hypothesize that change requests presented with an avoidance goal frame will be associated with (a) worse performance and (b) more negative evaluations of competence (own and communicator's) than change requests presented with approach goal frames. We further hypothesize that change requests presented with an avoidance (as compared with approach) goal frame will be associated with (c) more experience of reactance and in turn negatively affect the outcome. We have conducted two studies designed to test this set of hypotheses. In Study 1, a change request implied a restriction by asking participants to switch from a preferred working strategy to a new one that they previously did not chose and the goal frame was varied between approach and avoidance. In Study 2, we systematically varied the presence of a freedom restriction, as well as the goal frame in a message given to student participants about a proposed change in their study program. In both studies all procedures were in accordance with the local IRB regulations and the Declaration of Helsinki.

# STUDY 1

A goal is a cognitive representation of a possible state or outcome that an individual seeks to attain (Austin and Vancouver, 1996; Elliot and Thrash, 2002) and goals focus on either a positive or a negative possibility. Given that the way in which a goal is worded corresponds to the way in which the goal is represented in memory (Elliot and Friedman, 2007), we systematically varied the frame of the goal in which the change request was phrased.

# Method

We conceptualized a change request presented as a restriction framed with an approach goal as something that the participants "must do" versus a restriction framed with an avoidance goal as something that the participants "must not do."

With the goal of 50 participants, we began collecting data from a college student sample during the spring semester and terminated data collection when the academic year ended leaving a final N = 56.

#### Participants and Design

Fifty-six (39 women) undergraduates at the University of Rochester participated in the experiment entitled "Applied performance and problem solving" in return for course credit. The mean age of participants was 20.00 years (range = 18–24). Three participants who failed to engage in the puzzle-solving task were excluded from the analyses. Participants were randomly assigned to one of two between-subjects goal frame conditions: the approach goal frame condition (n = 29) or the avoidance goal frame condition (n = 24). The approach goal frame condition was manipulated by the wording "For the following task, you must choose a different strategy than the one that you just used" in half of the cases. The avoidance goal frame condition was manipulated by the wording "For the following task, you must not choose the same strategy that you just used" in the other half. The experimenters in this and the subsequent experiment were blind to participants' condition, and remained unaware of the hypotheses being tested throughout data collection.

#### Procedure

Upon arrival for the experiment, participants were greeted and presented with a computer-based problem-solving task on a screen (adopted from Förster et al., 1998, Study 3; anagram test). The screen displayed a collection of the letters d, p, q, b, r, and g presented in a rectangular box with rows and columns filled with letters where each letter was presented exactly 114 times<sup>1</sup> . The participants received the written instruction that completing the task consisted of counting how many times the letter p was presented on the screen and that they were under no time limit.

Following the presentation of the letters with an open time frame, participants were free to choose one among four possible strategies to complete the initial task for practice purposes: One strategy involved searching letters row by row, a second one involved searching letters column by column, a third involved searching letters grid by grid and the fourth involved searching the screen as a whole. After these strategies were explained to participants, they selected the strategy that suited them the most. They completed the task again with no time limit which consisted in counting how many times the letter p was presented and in entering the correct number in a field on the bottom of the screen online. After the participants finished solving the initial task, they were told to complete a second problem-solving task (very similar to the first task), hereby inducing the change. The change consisted in the explicit instruction to use a different search strategy than the one they had just adopted for the initial task, hereby manipulating the respective goal frame condition. This constituted the second round of solving the puzzle. The second task used the letters v, k, w, x, y, and z (to avoid habituation); participants were asked to search for the letter v (again same amount of letters for each letter category).

After completing the tasks, the participants were administered the measures and then given a final questionnaire assessing their basic demographics. Finally, participants were debriefed about the purpose of solving the puzzle and informed that the session was now over.

#### Measures

#### **Accuracy of solving the puzzle**

A difference measure of how many letters were correctly identified in each of the two rounds was calculated to represent

<sup>1</sup>An image of each puzzle option is displayed in the section for Supplementary Materials.

the overall accuracy of solving the puzzle and the performance of the participants. Similar to previous studies in which the absolute difference scores were computed (see Drach-Zahavy and Erez, 2002: high difference scores represented low performance), we took the absolute number of correctly identified letters at round 1 and subtracted it from the absolute number of correctly identified letters at round 2. If the resulting number was positive, the participants correctly identified more letters at round 1 compared with round 2. Conversely, if the participants identified more letters at round 2 compared with round 1, the resulting number was negative. In this manner, we were able to determine the accuracy of solving the puzzle and the performance accounting for the change from round 1 to round 2. Hence, not the performance per se at time 2, but the difference between the two rounds contingent on the respective goal frame condition was measured. Unlike previous research, we did not assess stress appraisals of challenge as opposed to threat, but manipulated the threat by restricting the strategy that the participants could select for the second task.

#### **Reactance**

Experience of threat to freedom in the form of psychological reactance was assessed with nine items used in previous research (Jonas et al., 2009). Items include "How reasonable did the request to change the strategy appear to you?" and "How restricted did you feel in your freedom to choose the strategy you wanted to use"? (1 = not at all, 10 = extremely) Scores were averaged to form a composite index (α = 0.83).

#### **Additional measures**

Two one-item measures for perceived task difficulty ("How difficult do you think the task was") and perceived competence ("How competent do you think you were in solving the puzzle"; 1 = not at all, 10 = extremely) were adapted from existing measures (Elliot and Church, 1997) for performance-avoidance goal frames.

# Results

#### Performance

A one-sample t-test on mean performance of participants between the two rounds showed that performance did not improve in general between round 1 (M = 7.90, SD = 6.49) and round 2 (M = 7.05, SD = 7.49), t(52) < 0.50, p > 0.64). Additionally, an independent sample t-test on performance in round 1 showed that it was similar in both goal framing conditions (Diff = −2.99, SD = 1.82), t(51) = −1.64, p = 0.11. Next, we tested whether performance differed as a function of condition and option selected by the participants at round 2. There were no significant main effects nor interaction, all Fs < 1.62, all ps > 0.199.

According to our hypothesis, a change request in an avoidance frame corresponds to an increase in perceived freedom restriction (i.e., a threat) which translates into increased reactance which, in turn impairs performance. We tested this mediational hypothesis using the bootstrapping procedure and corresponding SPSS macro process of Hayes (2012) developed for mediation and moderation analysis. It allowed us to test for indirect effects, regressing performance onto goal frame (dummy-coded: approach = 0/avoidance = 1), with reactance as the proposed mediator. One thousand bootstrap resamples were performed. As expected, we found an overall effect for goal frame and reactance on performance, R 2 adj. = 0.11, F(2,50) = 3.18, p = 0.048. First, a direct effect of goal frame on performance shows that those who were asked with an avoidance frame (compared to an approach frame) to switch their strategy found 4.79 more letters at round 1 compared with round 2, t(51) = −2.07, p = 0.036, 1 − β = 0.708, with a 95% confidence interval excluding zero (0.146–9.441), hence doing worse than those asked with an approach framed goal. Moreover, those who were asked with an avoidance frame were 0.93 relatively more reactant, t(51) = 2.51, p = 0.015, 1 – β = 0.669, with a 95% confidence interval excluding zero (0.185–1.675) than those asked with an approach frame, a score resulting from the difference in reactance following a one-unit change in goal frame. Additionally, those who felt relatively more reactant found on average 1.68 less letters at round 2, t(51) = −2.04, p = 0.047, 1 − β = 0.712, with a 95% confidence interval excluding zero (−3.337 to −0.028). Most importantly, the indirect effect of goal frame on performance through reactance is negative and statistically different from zero as evidenced by a 95% bias-corrected bootstrap confidence interval that is entirely below zero (−4.11 to −0.19). Suggesting that those participants confronted with an avoidance framed change request performed 1.57 times worse on average than those confronted with an approach framed change request, as a result of the mediation by reactance (see **Figure 1**, for a graphical depiction of the mediation model).

#### Reactance and Additional Dependent Variables

First, we tested the impact of goal framing on reactance, task difficulty, and self-perceived competence to compare the differential effect of approach and avoidance goal framing. Avoidance as compared with approach frame increased reactance, t(51) = −2.51, p = 0.015, d = −0.67, but had no differential impact on task difficulty or competence, ts ≤ −0.72, ps ≥ 0.48; see **Table 1** for means and standard deviations. Next, we used a moderated regression analysis to explore the relationship between reactance, task difficulty, and self-perceived competence moderated by condition. While reactance had an overall negative impact on task

∗∗p < 0.01.


TABLE 1 | Means and standard deviations for reactance, task difficulty and self-perceived competence in Study 1.

difficulty, β = 0.291, t(50) = 2.17, p = 0.035, it had no effect on competence, β = 0.168, t(50) = 1.22, p = 0.227, and no effect on either outcome variable when moderated by condition<sup>2</sup> .

Finally, we tested whether task difficulty or competence mediated effects on performance, finding a mediation only for competence, R <sup>2</sup> = 0.124, F(2,50) = 3.55 p = 0.036, showing that more competence valuation yields a smaller number of mistakes when searching for the correct letters (β = 0.296, p = 0.032)<sup>3</sup> .

# Discussion

In line with our hypotheses, when a change was communicated with an avoidance goal frame as compared to an approach goal frame, participants in Study 1 showed impaired task performance. They also indicated to experience more threat to freedom, i.e., psychological reactance, when avoidance compared to approach was the goal frame. Importantly, reactance mediated the effect of goal frame on the outcome: Avoidance (vs. approach) goal frame increased perceived threat to freedom, which in turn impaired task performance. This finding supports our theoretical claim that individuals who focus on a negatively phrased restriction or undesired option, here a prohibition to do something, face a greater aversive motivational state than individuals who focus on a positively phrased restriction or desired option, here an order to do something, even when the content of the task itself remains almost identical. This is also in line with previous research on approach and avoidance goal frames in the achievement literature which documents that individuals are more disorganized and challenged when they set avoidance framed performance goals (Elliot and McGregor, 1999). Only those participants in the avoidance (compared with the approach condition) who scored high in the perceived difficulty of the task showed a negative relationship between reactance and perceived competence. This finding resonates with empirical evidence on the effects of threat versus challenge in a task situation, in which a more difficult task (task difficulty perceived to be high) was more likely to produces a threat (Förster et al., 1998). For the purpose of our task, feedback was not mentioned and not given, so in neither of the conditions appraisal of own competence were possible.

Having obtained initial evidence that a perception of threat to freedom is associated with more negative outcomes in task performance when a change request is framed with an avoidance goal, we intended to more directly test the role of freedom restriction, introducing a condition that does not imply a restriction in Study 2. Here we examined the impact of goal frame in interaction with the presence versus absence of an actual restriction to freedom. Furthermore, we intended to see whether the effect of goal frame would replicate in a situation of persuasive influence, looking at communicator variables and persuasion in Study 2. Finally, we explored the effect of goal frame and restriction on affect.

# STUDY 2

The cognitive response approach (Petty et al., 1981) assumes that the impact of a message on attitudes is mediated by cognition. In hearing or reading a persuasive message, individuals generate cognitions that can be in agreement or disagreement with the message. Dillard and Shen (2005) contend that it is plausible that individuals respond to freedom-threatening messages with unfavorable cognitions about the message and about the communicator.

Accordingly, we expect to find that individuals confronted with a potential change presented with an avoidance compared to an approach frame will experience the avoidance frame as more threatening to their freedom which in turn will lead to a lower agreement with and more counterarguing regarding the proposed change. We further expect to find that individuals in the avoidance compared with the approach frame condition will evaluate the communicator of a threatening message more negatively. This negative evaluation will be shown on dimensions such as the perceived trustworthiness and competence of the communicator. We further expect to find that those individuals who are confronted with a restriction framed with an avoidance goal will experience the strongest reactance and that reactance will act as a mediator between the goal frame and the outcome, here the agreement with the proposed change. In Study 2, we also assessed differences in experienced emotions.

# Method

Our heuristic approach was to match participant numbers of previous studies on persuasive threat attempts and resistance to attitude change, which obtained medium to large effect sizes (e.g., Miller et al., 2007). With the goal of 100 participants, we

<sup>2</sup>Running correlations for each goal framing condition separately, however, the approach condition did not reveal a significant association between reactance and task difficulty in the approach condition (p ≥ 0.25) supporting the moderated regression analysis. However, in the avoidance condition, reactance is positively correlated with task-difficulty (r = 0.46, p = 0.023) suggesting that when more threat was underway – as induced by the avoidance framing of the change request-, more experience of threat to freedom was associated with a greater sense of challenge.

<sup>3</sup>When including task difficulty as mediating variable the overall model did not reach significance, R <sup>2</sup> = 0.06, F(2,50) = 1.58, p = 0.217, indicating that the effect of goal frames on performance cannot be explained with a greater perceived task difficulty in the avoidance (M = 4.13, SD = 1.45) as compared with the approach condition (M = 4.07, SD = 1.44), t(51) = −0.141, p = 889.

began collecting data from a college student sample at a German university during the summer semester and terminated data collection when the academic year ended leaving a final N = 105.

#### Participants and Design

fpsyg-07-00632 May 13, 2016 Time: 15:0 # 7

One hundred and five (79 women) participants volunteered toward the end of an undergraduate lecture to evaluate what was presented as the integration of a new program in return for course credit. The mean age of participants was 20.90 years (range = 19– 43). Participants were randomly assigned to conditions in a two (restriction: yes vs. no) by two (goal frame: approach vs. avoidance) factorial design.

## Procedure

First, the participants read a cover story in which a professor described and advertised a new concentration named 'Urban Design' as allegedly being implemented to the psychology program in the near future. They were informed that as a consequence of this addition a total of 13 instead of (now) 12 concentrations would be part of the program. The text varied between four scenarios, namely whether the addition would entail restrictions in terms of limited access to existing seminars (restriction present) versus no limitations (restriction absent) for the students and whether the addition of the new concentration was framed in terms of an approach or avoidance goal. In the approach goal frame conditions the change was described as helping to improve the university's high ranking whereas in the avoidance goal frame condition the change was described as helping to avoid a decrease in the ranking.

After reading the text participants responded to questions regarding their perception of the message (message evaluation: agreement with change), communicator (communicator evaluation: trustworthiness, competence), cognitive response: counterarguing, their perceived threat to freedom (reactance) and their affective (positive, negative) state. Next, participants were given a final questionnaire assessing their basic demographics. Finally, participants were debriefed about the purpose of the survey's content and informed that the session was now over.

## Measures

#### **Message evaluation**

To assess participants' agreement with integration of new concentration we used a five-item measure by Miller et al. (2007), e.g.: "How much would you support the implementation of such a program." (α = 0.89).

#### **Communicator evaluation**

To assess participants' perception of the communicator's competence we used a three-item measure of perceived competence of the communicator (Dillard and Shen, 2005), e.g.: "How qualified does the person who communicated the topic appear to you." (α = 0.91). To assess perceived trustworthiness, we used a two-item measure used in previous research on the impact of reactance on person perception (Silvia, 2006), e.g.: "How trustworthy does the person who communicates the topic appear to you?" (Spearman–Brown ρ = 0.88).

#### **Cognitive response**

To assess counterarguing we used a three-item measure adapted from Rains and Turner (2007), e.g., "Did you develop counterarguments against the here presented position? (α = 0.91).

#### **Reactance**

Threat to freedom in the form of psychological reactance was assessed exactly as in Study 1, and scores were averaged to form a composite index (α = 0.81).

# **PANAS-X**

A self-report affect scale (PANAS-X; Watson et al., 1988) was used to assess the affective consequences of the goal framing and restriction conditions. This scale consists of 20 words and phrases that describe different positive, negative and neutral feelings and emotions. The participants had to indicate to what extent "they felt this way right now." Scores were averaged to form a composite index for positive (six items; α = 0.84) and negative (nine items) mood (α = 0.78).

All items were presented with a 10-point scale reaching from 1 (not at all) to 10 (absolutely).

# Results

We ran a 2 (goal frame: approach versus avoidance) × 2 (restriction: present versus absent) between-subjects multivariate analysis of variance (MANOVA) on the composite scores of all dependent variables.

#### Reactance

There was a main effect of goal frame, F(1,97) = 9.30, p = 0.003, η <sup>2</sup> = 0.09, 1 − β = 0.85 with participants reporting greater experience of reactance in the avoidance (M = 3.63, SD = 2.29), relative to the approach condition (M = 2.73, SD = 1.48). This finding confirms our assumption that regardless of any other consideration, individuals would be more reactant when the addition to the program was presented in terms of negative possibilities. In addition, we found a main effect of restriction, confirming that the presence of a real restriction to the participants program in their concentration elicited a greater response in perceived threat when a restriction was present (M = 4.17, SD = 2.12) than when it was not (M = 2.20, SD = 0.95), F(1,97) = 44.39, p < 0.001, η <sup>2</sup> = 0.31, 1 − β = 1.00. These main effects were qualified by a significant interaction, F(1,97) = 5.72, p = 0.02, η <sup>2</sup> = 0.06, 1 − β = 0.66. Means and standard deviations are presented in **Table 2**.

# Post hoc Comparisons for the Experience of Reactance

Next, we conducted Bonferroni-corrected post hoc comparisons and found that within the approach goal frame conditions, participants showed more reactance in the restriction as compared with the no restriction condition, p = 0.013, d = 0.92. Similarly, within the avoidance goal frame conditions, participants showed more reactance in the restriction as compared with the no restriction condition, p < 0.001, d = 1.26. Looked at from a different angle, participants in the freedom restriction conditions showed the most reactance when they were in the avoidance as compared with the approach condition,


TABLE 2 | Means and standard deviations for reactance, agreement, trust, competence, counterarguing, positive affect, and negative affect in Study 2.

p = 0.005, d = 0.69. There was no significant difference between approach and avoidance in the no restriction condition, p = 0.553, d = 0.33.

# Message Evaluation, Communicator Evaluation, Cognitive Response, PANAS-X

We found a significant main effect for goal frame on all other dependent variables with the exception of counter-arguing (p > 0.1, η <sup>2</sup> = 0.026; more counter-arguing for avoidance frame) and positive mood (p > 0.4, η <sup>2</sup> = 0.007). There was less agreement with the proposed change (Mavoi = 5.80, SD = 1.75; Mappr = 6.97, SD = 1.74), F(1,97) = 10.52, p = 0.002, η <sup>2</sup> = 0.10, 1 − β = 0.89, less perception of communicator competence (Mavoi = 5.96, SD = 1.75; Mappr = 6.91, SD = 1.63), F(1,97) = 8.03, p = 0.006, η <sup>2</sup> = 0.08, 1 − β = 0.80, and less trust (Mavoi = 5.53, SD = 1.96; Mappr = 6.59, SD = 1.75), F(1,97) = 8.18, p = 0.005, η <sup>2</sup> = 0.08, 1 − β = 0.81, when it was presented with an avoidance frame as compared to an approach frame. Furthermore they experienced more negative mood (Mavoi = 1.53; SD = 0.52, Mappr = 1.34; SD = 0.41) in the avoidance frame condition, F(1,97) = 4.61, p = 0.034, η <sup>2</sup> = 0.05, 1 − β = 0.57.

As for restriction, we also found a marginally significant main effect on trust, indicating that people perceived the communicator as less trustworthy when she emphasized the restrictions for the program related to the change (M = 5.73; SD = 1.90) compared with the absence of a restriction (M = 6.45; SD = 1.90), F(1,97) = 3.79, p = 0.054, η <sup>2</sup> = 0.04, 1 − β = 0.49; see **Table 2** for all means and standard deviations.

#### Mediational Role of Reactance

Since the interaction effect of restriction and goal frame suggests that the framing affected reactance only when a restriction was present, we tested the meditational role of reactance here including restriction as a moderator, looking at the behaviorally most relevant outcome variable: agreement. To examine whether the agreement with the proposed change was mediated by the experience of threat to freedom in the form of reactance in dependence of the restriction condition, we conducted a moderated mediation analysis using the PROCESS tool by Hayes (2012). Examining the relationship between the goal frame condition and our central outcome variable 'agreement,' bootstrapping techniques employed to test conditional indirect effects confirmed the mediating role of reactance in Study 2 only when an actual restriction was present. The indirect effect of reactance on agreement was significant and positive (0.71) with a 95% confidence interval excluding zero (0.1653 to 1.3147), indicating significant mediation only when a restriction was present: The avoidance frame was associated with more reactance which in turn led to less agreement with the proposed change (see **Figure 2**). However, the interaction term qualifying this mediation did not reach significance (p = 0.127)<sup>4</sup> .

# Inter-correlations of Dependent Variables

Previously, the above six theoretically derived variables have been found to predict decision making following persuasion attempt. To explore these relationships further, correlational analyses were carried out. The majority of the variables correlated significantly with agreement. Most importantly, agreement is negatively associated with reactance r = −0.52, p < 0.001, while reactance is negatively correlated with both perceived

<sup>4</sup>Finding the goal frame affecting agreement via reactance only when a restriction was present on the one hand, and the main effect for goal frame on agreement on the other hand, suggest that framing has the same effect on agreement in the presence or absence of a restriction, however, for different reasons. One candidate mechanism to mediate the effect of goal frame on agreement in the absence of a restriction could be trust, since it represents the only other variable besides reactance to be affected by both factors (goal frame and restriction). In order to explore the meditational role of trust on reactance, we conducted a moderated mediation analysis with restriction as a moderator. Examining the relationship between the goal frame condition and agreement, bootstrapping techniques employed to test conditional indirect effects also here suggest a mediating role of trust in Study 2 only when there was no actual restriction. The indirect effect of trust on agreement was significant and negative (−0.66) with a 95% confidence interval excluding zero (−1.1653 to 0.1359), indicating mediation when a restriction was absent: The avoidance frame was associated with less trust which in turn led more reactance. However, this conditional effect was only slightly greater than the non-significant indirect effect of trust on reactance when there was a restriction (−0.50; confidence interval including zero −1.3681 to 0.0529), which is reflected in the fact that here the interaction term qualifying this mediation did not become significant (p > 0.6).

F(2,102) = 13.48, p < 0.001. #p = 0.37, n.s.; ##p = 0.13, n.s.; <sup>∗</sup>p < 0.05, ∗∗p < 0.001.

competence, r = −0.47, p < 0.001, and trustworthiness of communicator r = −0.53, p < 0.001, and is positively correlated with counterarguing, r = 0.60, p < 0.001 and negative mood r = 0.25, p = 0.009. So, the more reactant the person felt the less competent s/he evaluated the communicator. Additionally, the other aspect of how the communicator was perceived (trustworthiness of communicator) and the cognitive response (counterarguing) correlate substantially with both reactance and agreement with proposed change. As predicted, the greater the reactance, the less trustworthy the communicator appears and the more counter-arguing takes place. Conversely, the less trustworthy the participant evaluated the communicator and the more counterarguments s/he elaborated pertaining to the change the lower was her agreement with the message (see **Table 3** for all correlations).

# Discussion

In Study 2, we replicated the finding that goal frame influences how people respond to a proposed change through the experience of freedom restriction. While a freedom restriction was implied in the change request in Study 1, here we varied the explicit presence or absence of a restriction in a communication that was approach or avoidance framed. Consistent with reactance theory participants' felt threatened in their freedom when the communication indicated a restriction to their current range of choice for seminars. The restriction, however, led to even more reactance when it was presented in terms of an avoidance frame as compared to an approach frame. So, the presence of an actual restriction played an important role, however, when a restriction was present, the goal frame still significantly affected the degree of reactance. Goal frame, but not restriction, also influenced the evaluation of the communicator such that an avoidance frame led to less positive perceptions of the communicator and worse evaluations of the message. Moreover, the message presented with an avoidance frame as compared with an approach frame led to more counterarguing and less agreement with the proposed change. Looking at the meditational role of reactance here, we find that reactance mediates the impact of the goal frame on agreement with the communicated change only when an explicit freedom restriction was present. When there was no restriction the impact of goal frame was instead mediated, in part, via trust in the communicator, as our explorative mediation analyses suggested (see footnote 4). This suggests, that even messages that do not necessarily restrict the audience may yield disagreement elicited by reduced trust in the communicator when an avoidance goal is the frame.

Specifically, the above findings replicate and extend those reported in a meta-analysis conducted by Rains (2013), where counter-arguing and negative affective responses played a strong direct role in explaining the emergence of reactance. The present study extends these findings by relating the experience of a threat to freedom to goal framing.

# GENERAL DISCUSSION

In two experiments, when restricting a person's freedom avoidance goal frames were associated with more resistance to change on different levels of experience (reactance, performance, communication) and across different domains (achievement, person perception) than approach goal frames. Consistent with reactance theory, participants showed an especially aversive motivational state when a change was phrased in terms of a prohibition. The particular association between psychological reactance and goal frame in explaining the reception of a proposed change suggests that in some circumstances change can be experienced as more of a threat to freedom and therefore as a greater self-threat, particularly when it is frame as an avoidance goal.

Understanding change as a potential threat to self, contingent on the motivational and the situational context, emphasizes the complexity of the psychological processes that are involved. The appraisal of change is not limited to cognitive restructuring and integrating of new information or behavioral scripts. It can also be experienced as a self-relevant message that requires accommodation not only on the behavioral level, but that impacts an individual's sense of self-determination. As such, the goalframing of change has implications for theory and practice.

# Theoretical Implications

Extending reactance theory, the present research differentiated the induction of reactance. Across two experiments we differentiated between two outcomes of social pressure: We induced threat to freedom by requesting a person to change strategy ("you must," Study 1) or by suggesting that a change would lead to a restriction in choice options ("limited access to seminars," Study 2), while varying the goal frame, with the


Frontiers in Psychology | www.frontiersin.org May 2016 | Volume 7 | Article 632 |

assumption that avoidance frames imply freedom restriction and may thus increase reactance. Drawing on one part of the underlying axioms of reactance theory, namely, on the effect of impositions and prohibitions (see Wicklund, 1974), leads to the assumption that both the change requests and the restriction of choice may be perceived as threats to freedom and should hence, induce similar amounts of reactance.

The present findings, however, show that a restriction, when presented with an avoidance goal frame is perceived to be more threatening. To our knowledge, no other study has yet examined these specific message features in their implication for important every day decisions on how to phrase a request for change. In communicating change that implies a freedom restriction, emphasis on whether the change is associated with an additional positive outcome or whether a negative outcome can be avoided has a direct effect on the extent to which it is perceived as a restriction to personal freedom: Additional positive outcomes suggest a choice for the individual to decide, if these are wanted, or if the status quo is satisfying enough. Potential negative outcomes put the status quo at risk, which suggests more need to change an existing course of action. Our findings suggest that it is partly the implied restriction of freedom conveyed through goal framing that can make an actual freedom restriction to be responded to more or less in terms of a threat to self – here: expressed reactance and the associated variables (impaired performance, negative evaluation of a communicator, disagreement with message).

This interpretation of our findings resonates well with two prominent theoretical accounts. First, it draws on research that demonstrates on how the framing of a goal corresponds to the way in which the goal is represented in memory (Elliot and Friedman, 2007). The negative and undesirable dimension appears to be more salient in memory. Cognitive representations or schemas, when made salient or primed, are more easily retrieved and the likelihood that the forbidden option is selected or undesirable behavior is shown increases. Research in education and health (e.g., Granpre et al., 2003), focus on the question on when children and young adults react most favorably to antismoking, anti-drinking or anti-drug campaigns. These findings are supportive of the idea that it is a more adaptive strategy for teachers and health counselors to offer and name a desirable outcome than to focus on the undesirable outcome. Also, research conducted in the context of drug abuse prevention, points at the potential controlling nature of warnings and prohibitions (see, self-determination theory, Ryan and Deci, 2000) and contends that warning of negative outcomes may increase perceived threat to self and thus impair self-regulation. Therefore, instead of communicating prohibitions that are perceived to be controlling, phrasing a request in a manner that enables the individual to seek a solution in a more autonomous way allows for an easier integration of the change into a self-relevant course of action.

Second, it taps into the work by Higgins (1998) on the regulatory focus, which contends that not only chronic states, but also momentary situations – such as induced by message framing – can temporarily yield either a promotion or a prevention focus. For example, feedback messages or task instructions can communicate gain/non-gain information (promotion focus) or non-loss/loss information (prevention focus). Despite the seemingly close affinity between the concepts of approach-avoidance goal frames and self-regulatory focus, these two concepts are theoretically different. The approachavoidance distinction is rooted in the hedonic principle that contends that individuals strive to attain pleasurable and to stay away from painful outcomes. Self-regulatory focus theory distinguishes between two kinds of goal attainment that vary in chronic focus: attainment of aspirations and accomplishments (promotion focus) and attainment of responsibilities and safety (prevention focus). In combining both theoretical concepts, the approach-avoidance distinction could be viewed as a unifying conceptual thread used to organize and integrate various levels of investigation as it is applicable to dispositional, domainspecific, and situation-specific levels of analysis. For example, work by Higgins (1998) tested whether participants' motivation for approach versus avoidance was influenced by their regulatory focus and found that participants primed with promotion focus ideals recalled situations better in which they had to approach a match to a goal. The reverse applied to participants primed with a prevention focus.

Across both studies, the present findings show that the type of goal frame directly influences the outcome: When a change message was framed in terms of positive outcomes it led to less reactance, which in turn, yielded a lower (selfperceived) task difficulty, performance or person perception. The explaining role of reactance when a restriction to freedom was implied is consistent with recent findings by Reinhart et al. (2007), who found that gain-framed messages produce more positive reactions toward organ and tissue donation, and lower psychological reactance than loss-framed messages. In this research, following freedom restriction, psychological reactance, and perceived manipulative intent were found to mediate the relationship between framing and message reactions.

# Practical Implications and Future Directions

Research has supported the idea that resistance in the context of organizational change is often attributed to the situation specific to a change (e.g., Burke, 2008). Resistance to change comes from experiencing a lack of choice (i.e., the imposition of change) or from being forced to move away from a known state of being and acting (i.e., the deprivation of stability). The attenuation of aversive affect is critical to the success of a change request. It is important to factor in how changes and change requests are communicated. Understanding the importance of how to communicate change, so that people can engage with it constructively is of high relevance in the context of health communication, organizational change, social and economic justice movements, as well as in educational settings. When people see change as an opportunity to improve the status quo rather than a necessity to maintain it, they feel less threatened in their self-determined action and are more likely to integrate the proposed change, as opposed to showing resistance against change to protect an ego-motive. In this context it could be of interest to – in addition to reactance – explore

autonomy, which has shown differential relationships with reactance on how influence is interpreted depending on the source of communication (Pavey and Sparks, 2009)– the frame of the message might add to understanding of the complexity of persuasive communication.

Apparently, the pursuit of some types of goals is more depleting than others. One underexplored issue concerns the process that mediates the link between avoidance framed goal pursuit and goal progress. The presented studies show that avoidance goal frames possess a number of features that are detrimental in the process of regulation (see also Elliot and Sheldon, 1997). Future research may need to further address the question whether and why an avoidance framed goal is perceived to be more threatening to self and why prohibitions more than orders seem to elicit more negative affect. Mediational work on avoidance goal frames remains relatively sparse (see Elliot and Thrash, 2002): several processes appear to account for avoidance goal frames effects such as worry, stress generation, and poor goal progress. Reactance promises to be a new and intriguing mediator to study more in detail.

Furthermore, recent work seems to support our findings in that they point at the threatening nature of avoidance goal frames: avoidance as compared with approach goal frames deplete self-regulatory resources (Oertig et al., 2013). In the context of this research, importantly, the degree of threat was measured directly and compared between the two different goal frame conditions, a measure that the recent work has failed to assess. Therefore, it is only speculative if avoidance versus approach goal frames are experienced as more threatening in our research herein, but it is reasonable to argue so. Additionally, the cumulative findings of two studies appear to be consistent on the notion of the avoidance– reactance–compliance with change link and yield first empirical evidence that the focus on a negative outcome can lead to a more pronounced perceived threat of the self-motive freedom. This threatened self-motive then negatively affects the compliance with the change. Reduced levels of persistence in the avoidance as compared with the approach condition may be an alternative account, which would be worth investigating in future research.

Finally, future research may investigate the process behind the link between avoidance and the emergence of reactance. It is possible that a request to change is perceived to be less legitimate when negative instead of positive outcomes are emphasized (Elliot and Friedman, 2007).

To further enhance the generalizability of our findings, future research may need to implement more distinct state-and trait measures and different time frames. Previous work has discussed state and trait levels of avoidance goal frames (Fryer and Elliot, 2007) and state and trait levels of reactance (Dowd et al., 1991; Shoham et al., 2004).

In sum, given the negative implications of avoidance goal pursuit, a practical approach in social interactions would rely on a message framing that shifts the individual toward the pursuit of approach framed goals. Communicators might be welladvised to avoid high threat messages. Resistance in the context of organizational change is often attributed to the situation specific to the change at hand. Each request to change holds an implicit threat and may thwart the person's motive of need for control. The thwarted need, in turn, may lead to more resistance to follow the request (Rothbaum et al., 1982). One method to provide control for those whose need was thwarted may consist in attempting to predict events in order to avoid disappointment even if it means a worse outcome compared to the status quo. According to Shelly Taylor's prominent work on adjustment to threatening events, a person's thwarted need of control following a change request may be restored in providing meaning in the (loss) experience and providing an opportunity to regain mastery over the event and the possibility to enhance one's self-esteem (Taylor, 1983).

# CONCLUSION

The present research shows that communication style in terms of goal frame influences how a request to change is perceived via the experience of threat to freedom. It therefore allows theoretical insights into the process underlying the impairing effect of avoidance goal frames and helps to further our understanding on when and why a request leads to compliance. Understanding more specifically how the frame of a communicated change can affect self-related notions of freedom should ideally help to circumvent blind resistance in communication while allowing to focus on the actual implications of the change.

# AUTHOR CONTRIBUTIONS

For the present manuscript, DK and VG have contributed equally to study design, data collection, data analysis, interpretation and the theoretical integration of the research. JF has contributed to the Study design and data collection of Study 1, DF has contributed to the conduction of Study 2 and the theoretical integration of the research. All authors have contributed to the preparation of the manuscript with DK and VG taking the lead. All authors have approved the manuscript and take responsibility for all aspects of the work.

# ACKNOWLEDGMENT

We thank Professor Andrew J. Elliot, Ph.D. and Prof. Dr. Maria Agthe for their helpful comments on a previous version of this paper. We acknowledge the support of the Open Access Publishing Fund of the University of Potsdam.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpsyg. 2016.00632

# REFERENCES

fpsyg-07-00632 May 13, 2016 Time: 15:0 # 12



Wortman, C. B., and Brehm, J. W. (1975). "Responses to uncontrollable outcomes. An integration of reactance theory and the learned helplessness model," in Advances in Experimental Social Psychology, Vol. 8, ed. L. Berkowitz (New York, NY: Academic Press).

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Niesta Kayser, Graupmann, Fryer and Frey. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Legitimate vs. illegitimate restrictions – a motivational and physiological approach investigating reactance processes

#### *Sandra Sittenthaler1\*, Christina Steindl2 and Eva Jonas2*

*<sup>1</sup> Division of Economic and Organizational Psychology, Department of Psychology, University of Salzburg, Salzburg, Austria, <sup>2</sup> Division of Social Psychology, Department of Psychology, University of Salzburg, Salzburg, Austria*

Threats to our freedom are part of our daily social interactions. They are accompanied by an aversive state of motivational arousal, called reactance, which leads people to strive to reestablish their threatened freedom. This is especially the case if the threat seems to be illegitimate in nature. However, reactance theory suggests that reactance should also be aroused when people are exposed to legitimate freedom threats. In this article we first aim to show that both illegitimate and legitimate freedom threats evoke reactance. Second, we aim to extend past work on reactance by exploring the underlying process of experiencing a legitimate vs. an illegitimate restriction. In the current study (*N* = 57) participants were restricted in an illegitimate (unexpected and inappropriate) or legitimate (unexpected but appropriate) way, or were not restricted at all. We assessed participants' experience of reactance, their behavioral intentions to restore their freedom, their approach motivational states, as well as their physiological arousal (heart rate). Results indicated that when restricted in an illegitimate or a legitimate way, participants indicated the same amount of reactance as well as anger. However, when looking at people's physiological reactions, important differences between illegitimate and legitimate restrictions become apparent. Illegitimate restrictions led to an immediate arousal, whereas legitimate restrictions led to a time delayed arousal. This suggests that illegitimate restrictions lead to a sudden increase in aversive arousal. Legitimate restrictions, however, seem to be associated with a more cognitive process in which people first need to structure their thoughts and reflect upon the situation before getting into the feeling of reactance in a physiologically arousing sense. Moreover a mediation analysis could show that behavioral intentions to regain one's freedom result in positive and negative approach motivation. In sum we propose a combined dual-process and intertwined-process model explaining people's reactions to legitimate vs. illegitimate restrictions.

Keywords: reactance, restrictions, threat, anger, motivation, social interaction, physiological arousal

# Introduction

Imagine the following scenario: you are a future University student and very excited about beginning your studies next month. Browsing through the rental offers in the internet,

#### *Edited by:*

*Snehlata Jaswal, Indian Institute of Technology Jodhpur, India*

#### *Reviewed by:*

*Elena Bessarabova, University of Oklahoma, USA Soledad De Lemus, Universidad de Granada, Spain*

#### *\*Correspondence:*

*Sandra Sittenthaler, Division of Economic and Organizational Psychology, Department of Psychology, University of Salzburg, Salzburg 5020, Austria sandra.sittenthaler2@sbg.ac.at*

#### *Specialty section:*

*This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology*

*Received: 29 November 2014 Accepted: 30 April 2015 Published: 18 May 2015*

#### *Citation:*

*Sittenthaler S, Steindl C and Jonas E (2015) Legitimate vs. illegitimate restrictions – a motivational and physiological approach investigating reactance processes. Front. Psychol. 6:632. doi: 10.3389/fpsyg.2015.00632* you come across an advertisement of a lovely flat right next to the University. You are totally excited about the flat meeting your expectations and you call the number given in the ad. The landlord answers your questions, gives you some information on the flat and the rental contract, and finally invites you to a viewing. When you mention that you are a University student, the landlord interrupts you and says: "No, I won't rent my flat to a student!" and hangs up.

What's going through your mind thinking of the described scenario? Do you feel restricted in your freedom? Do you suddenly become angry? Do you maybe recognize a slight tension in your body? Why do you feel this way? Would you feel differently if the landlord had explained to you that his past tenants had been students who had trashed and damaged the flat before leaving? Would you still be angry? Would an additional justification like this override the clear restriction you faced as a student trying to secure accommodation?

In the current study we investigate how freedom threats in social interactions shape people's subsequent reactions. Furthermore, we are interested in investigating how people react when being restricted in an illegitimate vs. legitimate way. By analyzing people's experience of threat, their behavioral intentions, their self-reported affect and motivation, as well as their physiological arousal, we aim to explore how reactance processes in social interactions emerge. Thereby we aim to complement Dillard and Shen's (2005) description of reactance as an intertwined affective and cognitive phenomenon by adding the important factors of motivation and physiological arousal to increase our understanding of reactance processes.

#### Reactance Theory

In his publications on psychological reactance, Brehm contends that restrictions lead to reactance, a so called "motivational arousal" to re-establish the threatened freedom (Brehm, 1966, 1972; Brehm and Self, 1989). The desire of being 'free' again seems familiar to almost all of us. It is about daily moments (e.g., dress regulations) of perceived and actual freedom restrictions that end up making us feel unwell, nervous, and angry. According to the degree of the perceived restriction, to the context of the restriction, and to the meaning of the restriction for our future, we basically react differently (Brehm, 1966; Brehm and Brehm, 1981).

Restrictions often arise in a social interaction context, for example in our introduction example the restriction happens in the interaction between the landlord and the student. Restrictions to people's freedoms in social interactions have shown to trigger different reactance effects, like counter arguing, source derogations, aggression, or changes in attractiveness (e.g., Brehm, 1966; Clee and Wicklund, 1980; Brehm and Brehm, 1981; Fitzsimons, 2000). Dillard and Shen (2005) have summarized these reactions in their intertwined-process model, proposing that reactance is an intertwined process of negative cognitions (i.e., expression of disagreement with the restriction), and anger affect (i.e., feeling irritated, angry, annoyed, and aggravated). Testing competing conceptualizations of reactance, Rains (2013) confirmed the proposed intertwined-process model in a meta-analysis of 20 studies. Compared to single process models, a linear affective-cognitive model, and a dualprocess model, the intertwined-process model best fitted the data.

However, what exactly is reactance if it is an intertwined combination of anger affect and cognition? If we think of dualprocess models and their distinction between a more impulsive, affect driven, and a more cognitive dominated reflective modus of social behavior (e.g., Strack and Deutsch, 2004, for overview Gawronski and Creighton, 2013) – is this distinction meaningless in the context of reactance? Or can we distinguish between more impulsive vs. more reflective reactance processes? Furthermore, as reactance is defined as a motivational state (Brehm, 1966) we also need to better understand the role of motivational variables in addition to cognition and affect within the reactance process. Physiological arousal can be seen as a factor connected to motivational processes (e.g., Zanna and Cooper, 1974; Baum et al., 1986). Therefore, in the current research we compare physiological arousal following different kinds of freedom restrictions (illegitimate vs. legitimate restrictions) which should, according to Brehm (1966), both induce reactance.

Brehm (1966) suggested, for reactance behavior to emerge, the perceived restrictions do not solely have to stem from illegal acts. Restrictions that have legitimate justifications or even lead to a positive outcome can trigger reactance as well. This means that a loss of freedom, no matter how well justified should still arouse reactance. However, while illegitimate behaviors are unexpected, inappropriate, improper, and unjust, legitimate behaviors are unexpected but appropriate, proper, and just (Tyler, 2006; Zhang and Sapp, 2013). In this connection, fairness research suggests that decisions by authorities who are perceived of being legitimate (i.e., appearing fairer and appropriate) are more likely to be accepted by the people affected from the decision (Tyler and Huo, 2002).

To come back to our landlord example, one might ask whether the student may experience the same amount of reactance or less reactance depending on whether the landlord gives a justification of his behavior (legitimate restriction) or does not give such a justification (illegitimate restriction). In addition, when looking at physiological and motivational measures, there might be different kinds of processes when people are confronted with different restrictions to their freedom.

#### Physiological Arousal and Motivation

Cardiovascular responses such as heart rate (HR) are believed to be the best indicator of effort effects (see Wright, 2008) and in general to be a particularly promising method for showing physiological arousal in human beings (Brehm et al., 1964; Croyle and Cooper, 1983; Elkin and Leippe, 1986; Losch and Cacioppo, 1990; Harmon-Jones et al., 1996; Robinson and Demaree, 2007; Butler et al., 2009). Additionally, research showed that effort influence on the cardiovascular system is mediated by sympathetic nervous system activity, which is especially well reflected inter alia in heart contractility changes (Papillo and Shapiro, 1990; Berntson et al., 1993; Brownley et al., 2000). Similarly, Baum et al. (1986) already showed that reactance is associated with an increased activity of the sympathetic nervous system. Other studies related to learned helplessness demonstrated that the factor decisive for the increased physiological arousal, which is associated with one's motivational intensity to take action, is one's expected coping potential (Wortman and Brehm, 1975; Brehm and Self, 1989; Brehm, 1999). In studies testing Wortman and Brehm's model on reactance and helplessness (Pittman and Pittman, 1979; Mikulincer, 1988), a threat led to depression if people did not expect to be able to take action but aroused anger if people expected to be able to take action.

Consistent with these findings, research has shown that the possibility for coping with a situation is also related to approach motivation (Harmon-Jones et al., 2003). Approach motivation is a state in which people are motivated to move toward something and as such is related to not only positive but also negative emotions (Harmon-Jones et al., 2013). In several studies, Harmon-Jones et al. (2009) demonstrated that anger, a negative affective state, is related to approach motivation (e.g., Harmon-Jones and Allen, 1998; Harmon-Jones et al., 2003; Harmon-Jones, 2004). However, this was found only for conditions in which people were given the opportunity to resolve the angerarousing event (Harmon-Jones et al., 2003, 2006). If reactance is aroused one believes that he or she could engage in behaviors to restore freedom and thus, approach motivation should emerge. Additionally, the emotional state of anger is a central component of reactance (Dillard and Shen, 2005; Rains, 2013) and thus, reactance should lead to a motivational state of approach. This approach motivational state should then result in people's reactant behavior, which means that they try to approach the restoration of their freedom.

# The Present Research

In the present study we aim to investigate differences in illegitimate and legitimate restrictions by showing that different processes are activated in people experiencing those different restrictions. In line with Brehm (1966), we predict that both, individuals who experience illegitimate as well as individuals who experience legitimate threats should experience more reactance than a control group in which no threat occurs (hypothesis 1). Given Dillard and Shen's (2005) intertwined-process model in which anger plays a central role in all reactance processes, and the literature on anger being associated with approach motivation (e.g., Harmon-Jones et al., 2003), we further hypothesize that both restrictions evoke a negative approach motivational state, namely anger (hypothesis 2). So far, we assume that illegitimate and legitimate freedom threats are approximately similar concerning their outcomes. However, is there any difference regarding the physiological process underlying people's responses to freedom threats? Building on the dualprocess model by Strack and Deutsch (2004), we hypothesize that illegitimate and legitimate threats differ in the way physiological reactions are aroused: as an illegitimate restriction does not give any justification for the threat, people should immediately feel the urge to restore their freedom. This should be reflected in an increase in physiological arousal immediately after experiencing the restriction. Therefore, an illegitimate restriction

may trigger a fast process immediately leading to the motivational state of reactance accompanied by strong approach motivation. In contrast, as legitimate restrictions usually give justifications for the threat, they first may not be experienced as arousing as illegitimate restrictions. People might first need some time to reflect upon the restriction and its justification and to find counter-arguments (Dillard and Shen, 2005). We thus predict that legitimate restrictions are accompanied by a more cognitive process in which people first think about the restriction leading not to a similar immediate increase in physiological arousal as predicted for illegitimate restrictions. However, as both legitimate and illegitimate restrictions are predicted to lead to reactance (Brehm, 1966; Miron and Brehm, 2006) we propose that legitimate restrictions are also arousing but only after a delay of thinking and counter-arguing (Dillard and Shen, 2005). We hypothesize that while illegitimate restrictions trigger a more emotional, faster process and thus, an immediate physiological arousal, legitimate restrictions trigger a more cognitive, slower process and thus, a delayed physiological arousal (hypothesis 3). However, both restrictions should result in increased anger (and negative thoughts) as predicted by the intertwined model by Dillard and Shen (2005). Anger can be seen as an indicator of approach motivation (Harmon-Jones et al., 2013). As any attempt to solve a threat is an attempt to restore one's agency and thus, one's approach motivation (Jonas et al., 2014), we should find an approach motivational state at the end of the reactance process. As approach motivation is independent of valence we predict that behavioral intentions to regain one's freedom should result in the experience of a positive and negative approach motivated state (hypothesis 4).

# Materials and Methods

# Participants and Design

In this laboratory study, 57 students (41 female and 16 male) with a mean age of 22.51 years (SD = 4.94, two missing data points) of the University of Salzburg, Austria, voluntarily participated. The students had been recruited in several psychological lectures and at the University campus where they were asked to participate in a physiological laboratory study. Students were randomly assigned to one of the three experimental conditions (illegitimate vs. legitimate vs. control).

## Experimental Procedures

Participants in the experiment were asked to participate in a paper-and-pencil study for approximately 25 min, while we recorded physiological measurements. The experimenter explained that she was interested in what happens physiologically when students read about typical situations in students' daily life. The students were asked to complete all questions honestly and silently and were informed about the voluntary nature of participation as well as confidential use of data.

The questionnaire started with some demographic questions about sex, age, and field of study. After filling out these questions, the experimenter attached three "sensors" to the participants' fingers on the non-dominant hand to measure skin conductance (SC) and HR during the study. The fingers of the participants were washed with alcohol before attaching NeXus 10 sensors (two measuring SC fixed with Velcrostrip and one single finger clip measuring HR) on the first, second and third fingers. Participants were told to try holding their hand as still as possible while filling out the questionnaire. First a 3-min baseline measure ensured participants would fully focus on the study. They were told to calm down and try to relax for the next 3 min looking at the black screen of the computer. Then participants were asked to carefully read and imagine the reactance-arousing scenario (3-min period). In the illegitimate condition, they were asked to imagine that they were going to start studying at the Paris-Lodron-University in Salzburg the following semester, and were therefore looking for an apartment near the University. In a press advertisement they found an appropriate 1-room-apartment downtown. They called the landlord about viewing the apartment. When the landlord asked them for their profession, they stated they would be a student in Salzburg next semester. Before they could say anything else the landlord interrupted them and stated: "No, you are a student, you won't get this apartment" and broke off the call *(illegitimate condition)*. By contrast, in the *legitimate condition* the volunteers were asked to think of the same situation described above, but after the landlord had interrupted them he explained his behavior why he did not like students to rent his apartment. He explained that he was very sorry for his behavior but that he had bad experiences with student tenants in the past. In the *control condition* the students were asked to imagine that s/he was able to rent the apartment without experiencing any restrictions.

After participants had read the apartment search scenario, we assessed participants' state reactance. The items were arranged into two different scales: experience of reactance (α = 0.89, seven items, e.g., "To what extent do you perceive the reaction of the landlord as a restriction of freedom?" and "How much pressure do you feel as a result of his reaction?," adapted from Jonas et al., 2009), and behavioral intentions including evaluation items (α = 0.84, 10 items, e.g., "To what extent would you describe this man as incompetent to other students?"; "Would you like to ruin this landlord's reputation by publishing a negative review on a respected internet site?"). Answers were given on a 5-point Likert-type scale from 1 (*not at all*) to 5 (*very much*) 1 .

For the dependent measure of the physiological arousal in this study we used SC2 (in micro-mho) and HR (in beats per minute). Concerning future calculations with the physiological arousal we differentiated an immediate response to the threat of freedom while reading the scenario and a delayed response while answering the reactance items, which followed the apartmentsearch scenario. We used the difference values between both the immediate response and the baseline measure (ir-bm), and the delayed response and the (dr-bm) for further calculations. These difference values served as our measures of physiological arousal.

We assessed people's approach motivation by using the PANAS (Positive and Negative Affect Schedule consisting of a negative and a positive emotions scale; Watson et al., 1988). However, based on prior research showing that some of the PANAS items are related to approach motivation (Harmon-Jones et al., 2009; Steindl et al. in preparation) we created a measure for negative approach motivation, namely anger (upset, hostile, irritable; α = 0.85), and a measure for positive approach (active, attentive, inspired, alert, interested, strong, determined; α = 0.73)3 .

At the end of the study, participants were debriefed and thanked for their participation, and received course credits if desired4 .

2For the measure of SC, we found neither a significant interaction (*F*(2,54) *<* 1, *<sup>p</sup>* <sup>=</sup> 0.703, <sup>η</sup><sup>2</sup> <sup>ρ</sup>= 0*.*01) nor a significant main-effect of the restriction manipulation *<sup>F</sup>*(1,54) *<sup>&</sup>lt;* 1, *<sup>p</sup>* <sup>=</sup> 0.420, <sup>η</sup><sup>2</sup> <sup>ρ</sup>= 0*.*03; point of measure x restriction). Thus we did not include the measure in further analyses. Solely, we found a main-effect for the within factor (immediate vs. delayed response), *F*(1,54) = 31.98, *p <* 0.001, η2 <sup>ρ</sup>= 0*.*37. Participants showed less SC at the second point of measure (delayed response, *M* = 0.35, SD = 0.53) compared to the first point of measure (immediate response, *M* = 0.80, SD = 0.90). However, this is not surprising, considering the fact that we used two sensors of NeXus 10, which only could be fixed with Velcrostrip on the fingers of the participants. This method is much more errorprone than the use of real electrodes, such as Beckman Ag/AgCI electrodes used by other biofeedback-instruments (Losch and Cacioppo, 1990; Eisenberg et al., 1991). However, this is not surprising considering the fact that we used two sensors of NeXus 10, which only could be fixed with Velcrostrip on the fingers of the participants. This method for measuring SC is much more error-prone than the use of real electrodes, such as Beckman Ag/AgCI electrodes used by other biofeedback-instruments (Losch and Cacioppo, 1990; Eisenberg et al., 1991).

3The scale is based on Egloff et al. (2003), who originally proposed three positive factors: activation (active, attentive, inspired, alert), interest (interested, strong, determined), and joy (enthusiastic, excited, proud). However, due to a reliability of the interest factor of only α = 0.50 we combined the two factors activation and interest and termed it positive approach factor. Of the remaining three items of the positive PANAS scale we built the factor joy (enthusiastic, excited, proud; α = 0.84). Of the remaining seven items of the negative PANAS we built the factor negative emotions (distressed, scared, jittery, guilty, ashamed, nervous, anxious; α = 0.75). However, as approach motivation is a crucial concept in this article we only focus on the negative approach factor (anger) and the positive approach factor.

4For explorative analyses we measured the subscales "Victim Sensitivity" and "Observer Sensitivity" from the "Justice Sensitivity Inventory" by Schmitt et al.

<sup>1</sup>Items of the Experience of Reactance Scale: to what extent do you perceive the reaction of the landlord as a restriction of freedom? How much do you feel being put under pressure by his reaction? Are you frustrated about the reaction of the landlord? How inadequate do you think is his reaction? How much does his reaction bother you? How illegitimate do you think is the landlord's reaction? To what extent are you offended by his reaction?

Items of the Behavioral Intentions and Evaluation Scale: would you ever consider renting a flat of this landlord in the future? How much would you try to describe this man as incompetent to other students? How important would it be for you to argue against the landlord's reaction? How strong is your wish to complain about his reaction at the professional association for tenant's interests? How much would you advise other students against this landlord? Would you like to severely criticize the landlord in a daily newspaper? Would you like to ruin his reputation by publishing a negative review on a respective internet site? Do you think that this

landlord could also have prejudices against foreigners? Do you think that the landlord also shows discriminatory behavior in other areas? How likely do you think it is that this man takes advantage of other people?

We also measured the attractiveness of the apartment with one item and the sympathy for the landlord but did not use those two single items for further analyses. Participants in the legitimate (*M* = 1.94, SD = 0.64) and the illegitimate condition (*M* = 2.50, SD = 1.19) rated the apartment as less attractive compared to participants in the no restriction control group (*M* = 3.63, SD = 0.96), *F*(2,54) = 14.78, *p <* 0.001, η<sup>2</sup> <sup>ρ</sup>= 0*.*35. Furthermore participants showed less sympathy for the landlord in the two experimental conditions (illegitimate: *M* = 1.65, SD = 1.04; legitimate: *M* = 1.61, SD = 0.61) compared to the control group (*M* = 3.68, SD = 1.00), *F*(2,54) = 32.08, *p <* 0.001, η<sup>2</sup> <sup>ρ</sup>= 0*.*54.

# Results

#### Reactance Measures

To test hypothesis 1 that both illegitimate and legitimate restrictions lead to reactance we conducted univariate analyses of variance for the experience of reactance and the behavioral intentions measures separately. Means and error bars (95% CI) are displayed in **Figure 1**.

#### Experience of Reactance

As expected, participants in the illegitimate (*M* = 3.84, SD = 0.58) and the legitimate condition (*M* = 3.78, SD = 0.50) scored higher on the experience of reactance measure than participants in the control condition (*M* = 2.19, SD = 0.84), *F*(2,54) = 38.94, *p <* 0.001, η<sup>2</sup> <sup>ρ</sup>= 0*.*59. Subsequent *post hoc* analyses showed significant differences between both the illegitimate and the legitimate condition compared to the control condition (*p*<sup>s</sup> *<* 0.001). There was no significant difference between the illegitimate and the legitimate condition (*p* = 0.761).

# Behavioral Intentions

As expected, participants in the illegitimate (*M* = 3.20, SD = 0.64) and the legitimate condition (*M* = 3.14, SD = 0.60) scored higher on the behavioral intentions measure than participants in the control condition (*M* = 2.09, SD = 0.57), *<sup>F</sup>*(2,54) <sup>=</sup> 20.38, *<sup>p</sup> <sup>&</sup>lt;* 0.001, <sup>η</sup><sup>2</sup> <sup>ρ</sup>= 0*.*43. Subsequent *post hoc* analyses showed that both the illegitimate and the legitimate conditions differed significantly from the control condition (*p*<sup>s</sup> *<* 0.001). There was no significant difference between the illegitimate and the legitimate condition (*p* = 0.779).

# Approach Motivation – Anger and Positive Approach

To investigate hypothesis 2, stating that both, illegitimate and legitimate restrictions evoke approach motivation, we conducted a multivariate analysis of variance (MANOVA) with the PANAS scales' negative approach motivation (anger) and

<sup>(2005)</sup> as well as the "Collectivism and Individualism" scales by Triandis and Gelfand (1998).

positive approach factor. We found a significant effect for anger, *<sup>F</sup>*(2,54) <sup>=</sup> 9.56, *<sup>p</sup> <sup>&</sup>lt;* 0.001, <sup>η</sup><sup>2</sup> <sup>ρ</sup>= 0*.*26. Participants in the illegitimate (*M* = 3.35, SD = 0.96) and legitimate condition (*M* = 3.19, SD = 1.10) scored higher on anger than participants in the control condition (*M* = 2.00, SD = 1.07), *p <* 0.001, and *p* = 0.001, but participants in the illegitimate and legitimate condition did not differ from each other, *p* = 0.628. However, we did not find a significant effect for people's positive approach, *<sup>F</sup>*(2,54) <sup>=</sup> 2.32, *<sup>p</sup>* <sup>=</sup> 0.108, <sup>η</sup><sup>2</sup> <sup>ρ</sup>= 0*.*08, indicating that participants in the illegitimate (*M* = 2.98, SD = 0.46), legitimate (*M* = 2.76, SD = 0.76), and control condition (*M* = 3.23, SD = 0.59) showed about the same amount of positive approach.

# Physiological Measures Heart Rate

To test hypothesis 3, whether illegitimate restrictions lead to an immediate physiological response whereas legitimate restrictions lead only to a delayed physiological response, we performed a mixed-model ANOVA for the time of measurement (immediate physiological response vs. the delayed physiological response) as within-subject factor and the kind of restriction (illegitimate vs. legitimate vs. control) as between-subject factor. We found a significant main-effect of the restriction manipulation, *<sup>F</sup>*(2,54) <sup>=</sup> 3.40, *<sup>p</sup>* <sup>=</sup> 0.041, <sup>η</sup><sup>2</sup> <sup>ρ</sup>= 0*.*11. *Post hoc* analyses indicated that participants in the illegitimate condition (*M* = 3.72, SD = 0.76) showed a greater physiological response compared to participants in the control condition (*M* = 0.90, SD = 0.78), *p* = 0.012 but not compared to participants in the legitimate condition (*M* = 2.56, SD = 0.80), *p* = 0.296. Participants in the legitimate and control condition did not differ either, *p* = 0.142. We further found a maineffect for the within factor (immediate vs. delayed response), *F*(1,54) = 13.37, *p* = 0.001, η<sup>2</sup> <sup>ρ</sup>= 0*.*20. Participants showed more physiological arousal at the second point of measurement (delayed response, *M* = 3.02, SD = 4.21) compared to the first point of measurement (immediate response, *M* = 1.81, SD = 3.29).

Furthermore and most importantly, we found the predicted interaction (point of measure × restriction), *F*(2,54) = 3.67, *<sup>p</sup>* <sup>=</sup> 0.032, <sup>η</sup><sup>2</sup> <sup>ρ</sup>= 0*.*12. Simple effects within the illegitimate [*F*(1,54) <sup>=</sup> 0.43, *<sup>p</sup>* <sup>=</sup> 0.517, <sup>η</sup><sup>2</sup> <sup>ρ</sup>= 0*.*01] and control condition [*F*(1,54) = 1.96, *p* = 0.168, η<sup>2</sup> <sup>ρ</sup>= 0*.*04) showed that participants' physiological arousal did not differ in the immediate and delayed response. However, within the legitimate condition [*F*(1,54)<sup>=</sup> 17.55, *<sup>p</sup> <sup>&</sup>lt;* 0.001, <sup>η</sup><sup>2</sup> <sup>ρ</sup>= 0*.*25], participants showed much more physiological arousal in the delayed compared to the immediate response condition. Moreover, within the immediate response condition, legitimately restricted participants showed about the same level of physiological arousal as participants in the control condition, *p* = 0.427. Participants in the illegitimate conditions displayed higher physiological arousal values compared with participants in the control (*p* = 0.003) and the legitimate condition (*p* = 0.029). Following the delayed response condition, participants in the illegitimate (*p* = 0.068) and legitimate condition (*p* = 0.053) tended to display higher arousal values compared with participants in the control condition. Participants in the illegitimate condition displayed about the same level of physiological compared to participants in the legitimate condition, *p* = 0.950. These results support our hypothesis. Means and SD as well as error bars (95% CI) are displayed in **Table 1** and illustrated in **Figure 2**. Moreover, correlations with all measured variables are displayed in **Table 2**.

#### Approach Motivation

Hypothesis 4 predicted that the attempt to restore freedom leads people to come back into an approach state. Using the software Process 2.11 (Hayes, 2013, model 6), we performed two serial multiple mediation analyses, one with positive approach and one with negative approach (anger) as a dependent variable. First, we employed Contrast A (illegitimate vs. control condition, legitimate condition as a covariate) and Contrast B (legitimate condition vs. the control condition, illegitimate condition as a covariate) as the independent variables, positive approach as the dependent variable, and reactance experience and behavioral intentions as the two mediators. We used a 95% bias corrected bootstrap confidence interval (95% BC CI) and 5000 bootstrap samples. We found that the illegitimate threat did not have a significant total effect on positive approach, *b* = −0.25, SE = 0.21, *t*(54) = −1.16, *p* = 0.250. The legitimate condition, however, had a negative effect on positive approach, *b* = −0.47, SE = 0.22, *t*(54) = −2.15, *p* = 0.036, indicating that the legitimate restriction led to a lower positive approach. The effects for both restrictions was non-significant when the potential mediators experience of reactance and behavioral intentions had been added to the prediction, *b* = −0.31, SE = 0.31, *t*(54) = −1.01, *p* = 0.320 and *b* = −0.52, SE = 0.31, *t*(54) = −1.72, *p* = 0.091. The bootstrapped indirect effects of the illegitimate and the legitimate freedom threat via experience of reactance and behavioral intentions was significant in a positive direction, *b* = 0.36, SE = .19, BC CI [0.03, 0.80] and *b* = 0.34, SE = .19, BC CI [0.04, 0.82] (for the path coefficients see **Figure 3**). Thus, both restrictions which first arouse an experience of reactance that further led to behavioral intentions to restore one's freedom resulted in a final positive approach state suggesting that approach had been restored.

TABLE 1 | Means and SD for the ANOVA as a function of the experimental manipulation: immediate response and delayed response concerning heart rate (HR).


<sup>a</sup>*Immediate response* <sup>=</sup> *HR reading the scenario – baseline measure (difference value);* <sup>b</sup>*Delayed response* <sup>=</sup> *HR answering the reactance items – baseline measure (difference value).*

Second, we performed the same analyses but this time with negative approach (anger) as the dependent variable. We found that both, the illegitimate as well as the legitimate threat had a total effect on anger, *b* = 1.35, SE = 0.33, *t*(54) = 4.05, *p <* 0.001 and *b* = 1.19, SE = 0.34, *t*(54) = 3.46, *p* = 0.001. The effects for both restrictions were reduced when the potential mediators experience of reactance and behavioral intentions had been added to the prediction, *b* = −0.70, SE = 0.29, *t*(54) = −2.39, *p* = 0.020 and *b* = −0.77, SE = 0.29, *t*(54) = −2.67, *p* = 0.010. The bootstrapped indirect effects of the illegitimate and the legitimate freedom threat via experience of reactance and behavioral intentions on anger were significant, *b* = 0.74, *SE* = 0.20, BC CI [0.32,1.44] and *b* = 0.71, *SE* = 0.22, BC CI [0.27,1.51], indicating that both restrictions first arouse an experience of reactance that further led to behavioral intentions to restore one's freedom and finally resulted in a negative approach state, namely anger (for the path coefficients see **Figure 4**). Furthermore, the indirect effects of both threats on anger via experience of reactance were significant as well, *b* = 1.20, SE = 0.29, BC CI [0.55, 2.15] and *b* = 1.15, SE = 0.28, BC CI [0.54, 2.13], indicating experience of reactance and behavioral intention as mediators for both restrictions on anger.

These results support our hypothesis that behavioral intentions that aim to regain one's freedom should result in approach motivation, positive or negative.

# Discussion

#### Discussion of the Results

In the current research we explored effects of illegitimate and legitimate freedom threats in social interactions on people's subsequent reactions to those threats. We hypothesized that although both illegitimate and legitimate freedom threats arouse the experience of reactance, they should differ in their physiological arousal effects. Supporting our hypotheses and confirming


<sup>a</sup>*Immediate response* <sup>=</sup> *HR reading the scenario – baseline measure (Difference value);* <sup>b</sup>*Delayed response* <sup>=</sup> *HR answering the reactance items – baseline measure (Difference value);* ∗*p < 0.05, two-tailed.* ∗∗*p < 0.01, two-tailed.*

Brehm's (1966) assumption, results indicated that both kinds of restrictions evoked experienced reactance and behavioral intentions to restore one's freedom, for example by thinking about ruining the threatening man's reputation or describing him as incompetent to others. Moreover, both restrictions led to the same amount of experienced anger, which is a key component of psychological reactance (Dillard and Shen, 2005).

However, with regard to people's physiological arousal, our findings point to different underlying processes. While illegitimate freedom threats led to an immediate increase in HR, legitimate freedom threats led to a delayed increase in HR. We assume that illegitimate restrictions may be followed by a more automatic arousal. Legitimate restrictions, however, may first be followed by a more cognitive process and only after a delay be followed by a physiological arousal. Thus, people in the illegitimate condition may immediately feel their emotions boiling over. They are physiologically aroused and activated to "fight" for their freedom. People in the legitimate condition seem to need

some more time to recognize the landlord's unobvious restriction. After this delay in time they are able to enter the arousal state. *Dual process theories* might help to further understand the underlying mechanisms between these different processes. Dual process models distinguish between more impulsive, automatic and more reflective, controlled processes. As stated by Bargh (1994) automatic processes are typically unintentional and require little amounts of cognitive resources. Conversely, controlled processes are intentional and require considerable amounts of cognitive resources (see Gawronski and Creighton, 2013). Thinking of the landlord scenario, people in the legitimate condition may need more cognitive resources for understanding the restriction as a limitation of their freedom and for arguing against the restriction. Building on Strack and Deutsch's (2004) reflective–impulsive model (*RIM*) of social behavior we thus propose that illegitimate restrictions are associated with the impulsive system, and legitimate restrictions are associated with the reflective system. However, both processes seem to be related to anger and counter arguing. In both restriction conditions participants show experience of reactance and behavioral intentions – as the intertwined model by Dillard and Shen (2005) suggests. If people are confronted with restrictions to their freedom, they feel the motivation to restore that freedom – no matter if the restriction has been illegitimate or legitimate. However, the process how to regain freedom might differ. Being confronted with an illegitimate restriction, participants immediately feel an aversive arousal. They feel their anger, experience reactance, and show behavioral intentions to restore their freedom. This may further lead to approach motivation and restore people's agency. With regard to legitimate restrictions, however, the picture seems to be less clear. People do not like the restriction but there is an obvious and justified reason for the restriction. This could mean that people first have to reflect upon and argue against the restriction before getting into the same arousal state as people of the illegitimate restriction. When they are given the opportunity to restore their freedom, i.e., they intend to behave in a reactant way, positive approach motivation emerges. However, they experienced lower positive approach than people in the illegitimate condition. Yet, as this positive approach state has been achieved in both restrictions only after intending to behave in a freedomrestoring way, reactance seems to be functional in both restriction situations. These findings suggest that only after reactance has been aroused and demonstrated by behavioral intentions, one experiences a feeling of regained approach motivation. Deriving out of this, defensive behaviors may be important in order to experience one's agency again (Jonas et al., 2014).

# Implications, Limitations, and Future Research

In the current study we were able to show that illegitimate and legitimate freedom threats share their outcomes – they both arouse similar amounts of experience as well as behavioral intentions. Moreover, they both evoke negative emotions (anger) and positive emotions with an approach motivational character. Most importantly, our findings highlight the usefulness of exploring people's physiological reactions. By looking at people's HR, we found important differences between the two threats. Those would not have shown by looking at behavioral indicators only. As Wright (2008) stated, physiological arousal is a good indicator of effort effects and as such predicts people's motivation to achieve their goals (Brehm and Self, 1989). As we found a physiological difference between the two restrictions we also assume that people's effort to restore their freedom differs. However, we did not test whether there are differences in people's actual efforts but only measured intentional behavior. Future research might consider assessing real behavior in order to investigate people's efforts to restore their freedom.

One further limitation of this study is that we did not include more of the relevant measures in a time sequence. For example, we did not measure negative (and positive) affect immediately after the freedom restriction but only following the reactance measures. Moreover, we did not incorporate a manipulation check for (il)legitimacy after both scenarios. For future research this would be useful to get a better insight into the timing of both cognitive and affective responses. Following the intertwined model research, counter arguing is one important part of the reactance process (Dillard and Shen, 2005; Rains, 2013). Therefore, the perception of (il)legitimacy could have changed during the reactance process. We suggested that participants in the legitimate restriction condition first had to think about the given situation embedded in the social interaction. The landlord did not want to let the flat to students because he had made bad experiences with students. However, participants could have tried to argue against the landlord's reason not to hire the flat to students thinking of being a very proper and correct student himself/herself. Therefore it could be that those participants also had experienced a timedelayed arousal feeling of "illegitimacy" in the legitimate restriction5 .

One might also ask whether the delayed physiological response to the legitimate restriction could be partly explained by the fact that more information had to be processed in the legitimate condition when thinking about the reasons provided for the landlord's behavior. As this study is only a first step in understanding the underlying mechanisms and the considerations of the dual-process vs. intertwined model, future research should focus on shedding further light on the underlying mechanisms in illegitimate and legitimate restrictions. Strack and Deutsch (2004) show that both, the impulsive and reflective system, operate in parallel. However, the impulsive system enjoys priority over the reflective system because the latter only operates under conditions of sufficient cognitive capacity. The information processing in the impulsive system is assumed to be independent of resources. Consequently, we would predict that for example a cognitive load task (e.g., remembering a seven digit number) may influence legitimate (more "reflective") but not illegitimate ("more impulsive") restrictions of freedom.

When looking at the intertwined model of reactance (Dillard and Shen, 2005) in which reactance is defined as a combination of affect and cognition, one might be surprised that their definition misses a crucial factor: motivation. Brehm stated that reactance is "a motivational state and as such is assumed to have energizing and behavior-directing properties" (Brehm and Brehm, 1981, p. 98). Therefore, motivation is a very important factor that cannot be ignored. By adding the motivational components of physiological arousal and approach, our research attempts to better grasp the phenomenon of reactance.

During experiencing reactance and behaving in a reactant way, approach motivation seems to play an important role (see also Steindl et al., in preparation). It is defined as an "impulse to go toward" (Harmon-Jones et al., 2013, p. 291) and has been found in states sharing a motivational but not necessarily an emotional direction. Thus, anger (negative valence) and determination (positive valence) share the motivation to approach rather than avoid (e.g., Harmon-Jones et al., 2009, 2011). Believing in

one's ability to take action is a prerequisite for approach motivation to emerge (Harmon-Jones et al., 2003, 2006). As reactance motivates people to take action on restoring their freedom, they should feel capable to do so. Without the belief in one's coping abilities, helplessness rather than reactance would emerge and thus, avoidance motivation would be more likely. As approach motivation is often assessed by frontal alpha asymmetry in EEG studies (e.g., Harmon-Jones et al., 2010), one could elaborate on the findings of this study and explore frontal alpha asymmetry in reactance processes. This would be especially interesting with regard to the question if illegitimate and legitimate freedom restrictions differ.

One critical limitation of our research is that we only used a scenario for the experimental manipulations to evoke reactance. One might question to what extent it is possible for participants to solely imagine how they might feel if they are restricted in their freedom. However, the findings of the current study reveal that in addition to indicating a strong experience of reactance and behavioral intentions to restore their freedom, participants even showed a physiological response – although they were confronted only with a scenario. Therefore, we assume that we were able to arouse reactance with scenarios only. However, it would be useful for future studies to replicate the present findings using real illegitimate vs. legitimate restrictions.

# Conclusion

In sum, the present study helped to get first interesting new insights into understanding the underlying mechanisms of illegitimate and legitimate restrictions of freedom. To better understand reactance processes, we added the important factors of motivation and physiological arousal (already mentioned by Brehm, 1966) to Dillard and Shen's (2005) description of reactance as an intertwined affective and cognitive phenomenon. First, we showed that both illegitimate and legitimate freedom threats evoke similar amounts of reactance. Second, extending past research on reactance theory, we found that illegitimate restrictions lead to an immediate physiological arousal and legitimate restrictions lead to a time delayed physiological arousal. However, in order to provide a complete picture of the underlying processes of illegitimate and legitimate reactance processes and how these restrictions affect interaction processes, more research is needed. It is crucial to investigate the underlying processes of different forms of threat (illegitimate vs. legitimate) to enhance our understanding of social interaction processes.

# Acknowledgments

We are grateful to Haliemah Mocevic and Judith Rachl for their help in the data collection. We would like to thank Nicholas Levy and Anna-Maria Muck and two anonymous reviewers for their helpful comments on an earlier version of this manuscript. The second author of this article was financially supported by the Doctoral College "Imaging the Mind" of the Austrian Science Fund (FWF-W1233).

<sup>5</sup>Following Jonas et al. (2009) our measure of reactance also included one item measuring participants' feeling of illegitimacy. When we ran a factor analysis for the seven experience of reactance items of our reactance measure we found that the "illegitimate"-item highly loaded on the extracted experience of reactance scale (0.73). Therefore, it is not surprising that following the time delay in which participants had the opportunity to counter-argue the freedom restriction they experienced about the same amount of "illegitimacy" in both restriction conditions.

# References


Rains, S. A. (2013). The nature of psychological reactance revisited: a meta-analytic review. *Hum. Commun. Res.* 39, 47–73. doi: 10.1111/j.1468-2958.2012.01443.x

Robinson, J. L., and Demaree, H. A. (2007). Physiological and cognitive effects of expressive dissonance. *Brain Cogn.* 63, 70–78. doi: 10.1016/j.bandc.2006.08.003

Schmitt, M., Gollwitzer, M., Maes, J., and Arbach, D. (2005). Justice sensitivity: assessment and location in the personality space. *Eur. J. Psychol. Assess.*21, 202–211. doi: 10.1027/1015-5759.21.3.202


Strack, F., and Deutsch, R. (2004). Reflective and impulsive determinants of social behavior. *Pers. Soc. Psychol. Rev.* 8, 220–247. doi: 10.1207/s15327957pspr0803\_1


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2015 Sittenthaler, Steindl and Jonas. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# The Dynamic Reactance Interaction – How Vested Interests Affect People's Experience, Behavior, and Cognition in Social Interactions

*Christina Steindl\* and Eva Jonas*

*Department of Psychology, Social Psychology, University of Salzburg, Salzburg, Austria*

In social interactions, individuals may sometimes pursue their own interests at the expense of their interaction partner. Such self-interested behaviors impose a threat to the interaction partner's freedom to act. The current article investigates this threat in the context of interdependence and reactance theory. We explore how vested interests influence reactance process stages of an advisor–client interaction. We aim to explore the interactional process that evolves. In two studies, participants took the perspective of a doctor (advisor) or a patient (client). In both studies we incorporated a vested interest. In Study 1 (*N* = 82) we found that in response to a vested interest of their interaction partner, patients indicated a stronger experience of reactance, more aggressive behavioral intentions, and more biased cognitions than doctors. A serial multiple mediation revealed that a vested interest engendered mistrust toward the interaction partner and this mistrust led to an emerging reactance process. Study 2 (*N* = 207) further demonstrated that doctors expressed their reactance in a subtle way: they revealed a classic confirmation bias when searching for additional information on their preliminary decision preference, indicating stronger defense motivation. We discuss how these findings can help us to understand how social interactions develop dynamically.

Keywords: social interaction, vested interest, mistrust, experience of reactance, aggressive behavioral intentions, biased cognitions, information search

# INTRODUCTION

Buying a coffee at the bakery, chatting with a colleague at the office, meeting a friend, talking to the neighbors*...* Every single day we interact with people. These interactions require at least two persons who act, react, and in doing so, influence each other. Most of our daily interactions are easy-going. Imagine for example that Mr. Smith is ill and consults Dr. Boston for advice about possible treatment options. The doctor informs Mr. Smith about the advantages and disadvantages of possible medications and listens carefully while he describes his situation and states his needs and fears. Mr. Smith gets a prescription for his preferred medication, thanks the doctor for her advice and feels understood and appreciated. This example demonstrates how positive actions lead to positive responses and further positive actions. Consequently a favorable and trustful relationship-loop can develop.

However, interactions do not always run that smoothly. For various reason they sometimes develop negatively and the positive relationship-loop changes into a conflict-loop. Getting back

*Edited by:*

*Eddy J. Davelaar, Birkbeck, University of London, UK*

#### *Reviewed by:*

*Cindy Harmon-Jones, University of New South Wales, Australia Ursula Athenstaedt, University of Graz, Austria*

*\*Correspondence:*

*Christina Steindl christina.steindl2@sbg.ac.at*

#### *Specialty section:*

*This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology*

*Received: 29 November 2014 Accepted: 02 November 2015 Published: 27 November 2015*

#### *Citation:*

*Steindl C and Jonas E (2015) The Dynamic Reactance Interaction – How Vested Interests Affect People's Experience, Behavior, and Cognition in Social Interactions. Front. Psychol. 6:1752. doi: 10.3389/fpsyg.2015.01752*

to our doctor–patient example, imagine that Dr. Boston financially benefits from selling a specific kind of medication, e.g., a depot injection. She overrides the patient's wishes and needs, talks him into believing that the injection is best for him and reacts with harsh rejection to any other decision. As the doctor is only interested in his own financial benefit, the patient might fear that he may get a wrong medication and may therefore experience a realistic threat, such as an actual danger to his health (see Stephan and Stephan, 2000). However, the doctor's vested interest can pose an additional threat because the patient may experience a threat to his freedom to decide for himself. The experience of a freedom threat can also happen without any realistic threat present. Imagine for example that not the doctor but the patient would financially benefit from being sold the depot injection because he collaborated with the company which produced the depot. Although the doctor would neither be threatened economically, nor in his health or safety, he could experience the patient's vested interest as a threat to his freedom. This was demonstrated in an experiment by Wicklund et al. (1970). Participants were asked to rate six pairs of sunglasses and to choose one pair to model in front of a television camera. The chosen pair could then be purchased for half price. Before participants tried the glasses on, the salesperson either said "If you want to buy a pair, I'll be glad to handle it, since I get a 50% cut off all orders" (vested interest condition) or "I really don't care whether you buy a pair or not since I don't get anything from it, but I will be glad to handle the order if you do decide to buy them" (control condition). As participants tried the glasses on, the salesperson pressured them toward a positive evaluation by saying "Those are made for you," "Those are great." Findings indicated that the sunglasses were rated lower in the vested interest compared to the control condition. The authors argued that the self-interested behavior of the salesperson created a pressure to buy and thus, as a consequence a desire to maintain the freedom not to purchase any pair arose.

The consequences of social influence attempts like those have been well-explored in research on reactance theory. The theory demonstrates that threats to our freedom to decide for ourselves can lead to a state of motivational arousal with the aim to restore the freedom. This state is known as psychological reactance and has been found to result in different reactance effects. Thus, people show for example aggressive behaviors or biased cognitions such as a devaluation of the imposed object and the threatening person (for an overview on reactance theory, see Brehm, 1966; Brehm and Brehm, 1981; Miron and Brehm, 2006; Steindl et al., 2015).

The aim of the current paper is to improve our understanding of how threats in social interactions lead to conflict loops, i.e., why social influence attempts resulting from vested interests are perceived as threats to freedom and how they lead to the experience of reactance resulting in cognitive and behavioral consequences. We propose a "Loop2Loop model of social interactions" (Jonas, 2015; Jonas and Bierhoff, in press; Jonas and Steindl, 2015), which builds on interdependence theory (Thibaut and Kelley, 1959; Kelley and Thibaut, 1978; Kelley et al., 2003; for an overview see Van Lange and Rusbult, 2011; Van Lange, 2012) and subdivides the interaction process into its single stages. In the present studies we investigate how vested interests in social interactions affect people's perception of the interaction partner and their subsequent reactions with regard to: (a) their experience of threat, (b) their behavioral intentions, and (c) their cognitions. By analyzing the single stages of the social interaction, which are characteristic for reactance processes, we aim to explain how vested interests shape people's sequential series of reactions to the threat. In the following paper we refer to this as the "dynamic interaction" between two individuals. This means that in an interaction, person A's behavior affects person B's reactions, those again affect person A's reactions, and so forth.

# Interdependence Theory

Social influence attempts, such as health campaigns or clinical advice, are sometimes perceived as threats to our freedom to decide on our own (e.g., Grandpre et al., 2003; Dillard and Shen, 2005; Silvia, 2005). Those threats often happen in social interactions where one or more persons limit a specific freedom of another person. To understand why and how such negative interactions develop, interdependence theory (Thibaut and Kelley, 1959; Kelley and Thibaut, 1978; Kelley et al., 2003; for an overview see Van Lange and Rusbult, 2011; Van Lange, 2012) may help. Interdependence theory specifies the characteristics of a social exchange situation in which two or more persons are interdependently related to each other. It mainly focuses on the costs and benefits (outcomes) of social interactions.

Interdependence means that the outcomes of the interaction are influenced not only by one's own behaviors (actor control) but also by the behaviors of the interaction partner (partner control) and by the cooperative behavior of the two partners (joint control). In terms of the theory, we can describe an interaction as a situation in which two people depend on each other with regard to the fulfillment of their needs, thoughts, motives, and behaviors. The more dependent person (high partner control) is likely to sacrifice or accommodate. The less dependent person (low partner control) holds greater power over the other person and threats and coercions are possible (see Van Lange and Rusbult, 2011). For example, Dr. Boston is dependent on the patient's final choice because she would receive a financial benefit from selling the specific brand of injection to her patient. However, she is less dependent on the patient than vice versa. This could lead the doctor to force the patient to choose the injection. The patient, on the other hand, probably relies even more on the doctor's behavior (high partner control) than vice versa because he does not have the necessary medical expertise to know which medication is best for him to relieve his symptoms. If the doctor would try "to force" the patient to choose a specific brand of injection, the patient may experience an even stronger feeling of having been threatened, especially if he feels he greatly depends on the doctor's behavior. In this case, the patient may experience an even stronger threat in response to the doctor's attempt to influence him compared to a situation in which the patient was less dependent on the doctor's advice. This asymmetric level of dependence, which means that one interaction partner relies more on the other and thus, is subject to a higher partner control, not only affects people's satisfaction with the interaction (e.g., Kelley et al., 2003) but further affects their emotional and motivational experience, their cognitions, and behaviors. If the doctor would persuade the patient to take the depot injection, how would the highly dependent patient feel in this situation? And how would those feelings further affect his reactions?

# A Dynamic Model of Social Interaction: The Loop2Loop Model

Our "Loop2Loop model of social interaction" (Jonas, 2015; Jonas and Bierhoff, in press; Jonas and Steindl, 2015) describes a dynamic interaction process between two or more persons and how their behaviors mutually affect each other (**Figure 1**). It builds on the SABI model of interdependence theory (Kelley et al., 2003) in which the interaction (I) is a function of two interacting persons' (A and B) needs, thoughts, motives, and behaviors, and the characteristics of the situation (S): I = *f* (A, B, S). The Loop2Loop Model sets a strong focus on people's motives influencing people's motivation for the actual interaction. They give rise to the emerging motivational-affective state, the motivated cognitions, and motivated behaviors. These effects create a dynamic development of social interactions which means that the behavior of one person affects the experience, behaviors, and cognitions of another person which in turn affect the first person and so forth. Therefore, the Loop2Loop model extends the SABI model by explaining how interactions dynamically develop over time. For example, trusting behaviors such as honesty or reliability create a trustworthy atmosphere further arousing a feeling of one's motives being in good hands. The relationship-loop can develop in a favorable way. On the contrary, untrustworthy behaviors such as cheating or lying threaten one's motives, arouse a feeling of being threatened and lead to negative cognitions about the interaction partner which further creates a conflict-loop. Jodlbauer and Jonas (2011), for example, found that clients who recognized self-interested cues of advisors perceived the advisors as less trustworthy which in turn decreased their willingness to accept the recommendation. If we would take any further, this could in turn threaten the advisor's motivation to sell a product and thus, could trigger anger and even aggressive behavior in the advisor. When integrating this example into the Loop2Loop model, we have the advisor's self-interested behavior in one loop and the client's resistant behavior (refusal of the recommendation) in the other loop. What about the motivation and cognition in-between? Why do situations involving self-interested behaviors often elicit resistance?

# Vested Interest, Reactance, and Mistrust

Think back to the doctor–patient example in which Dr. Boston tries to persuade her patient to take the depot injection. How would the patient react? Research in the context of persuasion has shown that persuasive attempts, for example, trying to convince a person of a specific opinion or product often fail to produce their desired effects. Rather, they often lead to the exact opposite attitude or behavior. Several studies found psychological reactance to be a possible explanation for these failures (e.g., Worchel and Brehm, 1970; Heller et al., 1973; Dillard and Shen, 2005; Kim et al., 2013; Rains, 2013): Perceived intent to persuade

is often perceived as a threat to one's freedom to decide for oneself leading to a motivation to restore one's freedom called psychological reactance (Brehm, 1966).

Reactance consists of an affective, as well as a cognitive and a behavioral component. Miron and Brehm (2006) defined the affective component of reactance as one's subjective experience (feeling) that accompanies the urge to restore freedom. This subjective experience was specified as the experience of uncomfortable, hostile, and aggressive feelings (Brehm, 1966; Brehm and Brehm, 1981), but also as anger affect (Dillard and Shen, 2005). Other research combined the extent to which one perceived a situation as a freedom threat with one's emotional experience (e.g., frustrated, annoyed, offended) to assess the subjective experience of reactance (Traut-Mattausch et al., 2008, 2011; Jonas et al., 2009; Sittenthaler et al., 2015; Niesta-Kayser et al., submitted). Dillard and Shen (2005) demonstrated that reactance not only consists of affect but also of cognitions. In their persuasion research, they focused on negative cognitions such as counterarguments or derogation of the source of threat. Another cognitive indicator of reactance used in various reactance studies, is the change in attractiveness which means that people upgrade a restricted but downgrade an imposed message or product (e.g., Brehm, 1966; Bushman and Stack, 1996; Fitzsimons and Lehmann, 2004; Dillard and Shen, 2005; Rains and Turner, 2007; Bijvank et al., 2009; Laurin et al., 2012; Rains, 2013). The behavioral component of reactance is often described as the execution of the restricted behavior. In addition, forcing the threatening agent to remove the threat or behaving in a hostile and aggressive way (Brehm, 1966; Brehm and Brehm, 1981; Miron and Brehm, 2006) represents the behavioral component.

However, these reactance effects only emerge if two conditions are met. First, only if people expect to have a certain freedom of choice, reactance can occur (Brehm, 1966; Brehm and Sensenig, 1966; Clee and Wicklund, 1980). Imagine the patient again, who expects to receive an advice from Dr. Boston but still expects being able to decide what is best for him. Second, only if there is a perceived threat to this freedom of choice, reactance can occur. According to Brehm and Sensenig (1966, p. 703), "When a person is free to choose between two alternatives, any attempt by another person to influence his choice should be perceived as a threat to his freedom." Thus, if the patient notices that Dr. Boston tries to influence him, the patient perceives a threat to his freedom.

When people try to influence someone, they may have manifold reasons for it. One reason for a social influence attempt results from a vested interest. In general, a vested interest is defined as a hedonically relevant attitude object which has important perceived personal consequences for the attitude holder (Crano and Prislin, 1995). The more important the attitude object, the more likely is the attitude expressed in one's behavior. In many cases, this vested interest is of financial nature. A study by Jones et al. (2014) explored how the explicit mentioning of a financial interest of a salesperson influenced the clients' satisfaction with their purchase. They manipulated vested interest by salespersons urging customers to evaluate their purchase most positive in order to minimize the risk of decreasing the salesperson's compensation. If this vested interest was present, participants indicated lower satisfaction with their purchase than if no vested interest was present. Similarly, Wicklund et al. (1970) demonstrated that sunglasses sold by a salesperson with a financial interest were liked less than sunglasses sold by a salesperson with no financial interest. Thus, people's perception of a vested interest of a salesperson resulted in downgrading the purchase. Such a change in the attractiveness of an object, opinion, or message is an important cognitive indicator of reactance (e.g., Brehm, 1966; Brehm and Brehm, 1981; Laurin et al., 2012). In the studies described above (Wicklund et al., 1970; Jones et al., 2014), people showed reactance after they had perceived a vested interest of their interaction partner.

As research on attitude change suggests, people with vested interests are perceived as less trustworthy (for an overview see Eagly and Chaiken, 1993). In two studies, Jodlbauer and Jonas (2011) found that clients, who perceived self-interested intentions of their advisors, trusted them less which in turn decreased their willingness to accept the advice. Thus, we expect mistrust to explain why vested interests arouse a feeling of being threatened and why this feeling results in reactance effects.

# The Present Research

In the present studies we explore how self-interested behavior can affect the dynamics of social interactions. Therefore, we split up the reactance process into its single stages (experience of reactance, aggressive behavioral intentions, and biased cognitions in the form of negative attitudes toward the threatener and a change in attractiveness of the imposed and non-imposed options). In two studies we explored these stages in an advisor– client context by presenting a fictitious doctor–patient paradigm to participants in which they imagined themselves being in the position of either a patient who sought medical advice from a doctor or being in the position of a doctor treating a patient. In the *neutral paradigm* the interaction partner (either the doctor or the patient) was very open to any kind of medication. In the *vested-interest paradigm*, the interaction partner (either the doctor or the patient) had a financial interest and forced the other person to take/recommend the depot medication. We predicted that after reading the vested-interest paradigm, people would show more mistrust, reactance-related experience, behavior, and cognition than after reading a neutral paradigm. In addition, we predicted that people in the role of a client would react with stronger reactance than people in the role of an advisor. In their role as experts advisors are usually less dependent on the clients than the clients are on them. That is clients are subject to a higher partner control and thus, advisors have power over the clients (Thibaut and Kelley, 1959; Kelley and Thibaut, 1978; Kelley et al., 2003). In the example of the doctor–patient interaction, the patient would suffer more than the doctor from a vested interest because the patient's health depends on the doctor's advice. Thus, the patient experiences a realistic threat – an actual danger to his health. A vested interest may additionally be experienced as a threat to one's freedom to decide for oneself. This should result in a higher defensiveness of a patient compared to a doctor. In addition, if advisors do not have any incentive to appear in a positive light, they seem to be more accuracy-motivated and thus, not as biased as clients or personal decision makers (Jonas and Frey, 2003; Jonas et al., 2005). Therefore, a doctor should feel less affected by a patient's attempt to threaten the doctor's freedom.

In both studies, treatment of the participants was in accordance with the ethical standards of the American Psychological Association (APA). Participants were informed that there were no right or wrong answers to the questions, that the data would be treated confidentially and anonymously, and that drawing any personal reference from it would not be possible. In order to assure anonymity, participants were not asked for information that allows inferences to the participants (e.g., names). Participants were aware that they could withdraw from the online-study at any time. Participants were also provided with the name and email address of the responsible investigator. At the end of the survey, participants were thanked for their participation and provided with contact details if they wished to address any questions about the purpose of the study. They also received course credits if desired.

# STUDY 1

In Study 1 we investigated individuals' mistrust toward their interaction partner, their experience of reactance, their aggressive behavioral intentions, and their biased cognitions in a fictitious doctor–patient interaction. We further examined whether mistrust toward their interaction partner causes the emerging reactance process. Therefore, we developed scenarios in which people imagined being in the position of a protagonist whose interaction partner shows self-interested behavior. We address the following assumptions:

# Mistrust and Reactance – Main Effects

Firstly, we predicted that after a vested interest of the interaction partner, one perceives the partner as highly untrustworthy, i.e., s/he experiences strong *mistrust* toward the interaction partner (see Eagly and Chaiken, 1993; Jodlbauer and Jonas, 2011). Second, we hypothesized that a vested interest is experienced as a threat to one's freedom (see Wicklund et al., 1970; Jones et al., 2014). This means that, compared to no vested interest, a vested interest leads participants to *experience more reactance*, reveal more *aggressive behavioral intentions*, and show more biased cognitions, i.e., they have more *negative attitudes* toward their interaction partner and rate the issue that has been imposed on them *less attractive*.

# Reactance – Interaction Effects

As people in the role of clients are first, subject to a higher partner control (see Thibaut and Kelley, 1959; Kelley and Thibaut, 1978; Kelley et al., 2003) and second, have been shown to defend their own position more strongly than people in the role of advisors (see Jonas and Frey, 2003), we predicted that in the current study patients who face a vested interest of their doctor *experience more reactance*, reveal more *aggressive behavioral intentions,* and have more *biased cognitions* than doctors who face a vested interest of their patient.

# Dynamic Development – Mediation Effects

Mistrust should instigate a reactance process to evolve. Therefore, we predicted that people's perception of *mistrust* mediates the relationship between a vested interest and people's *experience of reactance,* which further shades into *aggressive behavioral intentions* and finally results in people's long-term *biased cognitions*. Summarized, the single threat stages should cause the development of the dynamically developing reactance process.

# MATERIALS AND METHODS

# Participants and Design

Eighty-two students (57 women, 24 men, 1 unspecified; *M*age = 23.84 years, *SD* = 7.18) from the University of Salzburg were asked to read one of four paradigms in which they either were asked to empathize with the role of a doctor (advisor) or with the role of a patient (client). They received either a scenario in which their interaction partner showed a vested interest or a neutral scenario. Thus, the experiment was based on a two (interest: vested interest vs. neutral) × two (role: doctor vs. patient) between-subjects design.

# Materials and Procedure

The online questionnaire first gave general information about the study and obtained some demographic information. After reading short information about schizophrenia, participants rated two types of medication (depot injection vs. pills) for the treatment of schizophrenia in their attractiveness [scale from 1 (*not at all attractive*) to 10 (*very attractive*)]. Participants were randomly assigned the role of a doctor or a patient and asked to read one of two paradigms. Afterwards, they answered questions concerning their mistrust toward the interaction partner, their experience of reactance, their aggressive behavioral intentions, and their biased cognitions [scales from 0 (*not at all attractive*) to 10 (*very attractive*)]. At the end of the survey, participants were thanked for their participation and provided with contact details if they wished to address any open questions about the purpose of the study. They also received course credits if desired.

# Paradigms

The paradigms consisted of a consultation situation. One group of participants empathized with the *role of a doctor* who had to decide between recommending either a depot injection or pills for treating their schizophrenia patient M. Schneider. In the *neutral scenario* patient M. Schneider was described as being very friendly and open to information about both types of medication. In the *vested-interest scenario* patient M. Schneider was only interested in being recommended the depot. He mentioned that he collaborated with a pharmaceutical company which produced the depot and that he would therefore have financial advantages. He forced participants in the role of the doctor to recommend the depot. Thus the consulting seemed more like a justification of M. Schneider's already taken decision.

The other group of participants empathized with the *role of the schizophrenia patient* M. Schneider who had to decide which type of medication (depot injection vs. pills) he would choose. In the *neutral scenario* the doctor Dr. Müller was described as very friendly and explained the advantages and disadvantages of both types of medication. In the *vested-interest scenario* Dr. Müller was only interested in recommending the depot. He mentioned that he collaborated with the pharmaceutical company producing the depot and that he would therefore have financial advantages.

### Questionnaire – Reading Check, Mistrust, and Reactance

To check whether people had read the scenarios thoroughly we assessed whether they ascribed to the doctor/patient a personal interest in recommending/receiving the depot or the pills (*reading check – personal interest depot*; "Do you believe the doctor/patient is personally interested in recommending/receiving the depot?"; *reading check personal interest pills:* Do you believe the doctor/patient is personally interested in recommending/receiving the pills?"). After the paradigm we assessed participants' *mistrust* toward their interaction partner (seven items; α = 0.90; e.g., "Do you think that the doctor/patient has made his or her decision already before the consultation?"). Then, participants indicated their *experience of reactance*<sup>1</sup> (Sittenthaler et al., 2015; four items; α = 0.96; e.g., "To what extent do you perceive the behavior as a restriction of freedom?," "How much does the doctor's/patient's behavior bother you?"). We also assessed people's *aggressive behavioral intentions* (three items; α = 0.84; e.g., "How likely are you to describe this doctor as incompetent to other patients?/How likely are you to describe this patient as stubborn to colleagues?"). To assess participants' biased cognitions, we asked them to indicate their attitude toward the interaction partner (*negative attitudes*; two items, r = 0.462 ; e.g., "Do you think that this doctor/patient could have prejudices against mentally ill people?"). Moreover, before and after participants read the scenarios, we assessed the *attractiveness* of the depot and pills medication. For analyzing people's judgment of attractiveness of the imposed and nonimposed medication, we took the difference score between attractiveness of the depot or the pills after minus before reading the scenario with negative scores indicating a lower attractiveness of the medication after the scenario.

# RESULTS

# Reading Check

To assure that participants had read the scenarios thoroughly, we computed a difference score between personal interest in depot and personal interest in pills with positive scores indicating a higher interest in the depot compared to the pills medication. The univariate analysis of variance (ANOVA) with interest (vested interest vs. neutral) and role (doctor vs. patient) as independent variables and the difference score as dependent variable showed a significant main effect for interest, *F*(1,78) = 94.82, *p <* 0.001, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.55. Participants in the vested-interest condition who had been imposed the depot were indeed more strongly convinced that their interaction partner was personally interested in recommending or receiving the depot over the pills (*M* = 6.63, *SD* = 2.68), while participants in the neutral condition were convinced that their interaction partner was more or less equally interested in both medications (*M* = 1.20, *SD* = 2.46). Thus, we can conclude that people realized the threatening partner's intentions. Furthermore, the main effect for role was significant, *<sup>F</sup>*(1,78) <sup>=</sup> 4.63, *<sup>p</sup>* <sup>=</sup> 0.034, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.06, indicating that participants in the role of patients were in general more strongly convinced that the doctor was personally interested in recommending the depot over the pills (*M* = 4.50, *SD* = 3.90) than vice versa (*M* = 3.30, *SD* = 3.52). The interaction between interest and role was not significant, *F*(1,78) *<* 1, *p* = 0.444, η<sup>2</sup> *<* 0.01.

# Mistrust and Reactance

We tested our predictions that first, threatened participants experience strong *mistrust* toward the interaction partner, second, that in general threatened participants show more reactance (experience of reactance, aggressive behavioral intentions, negative attitudes) than non-threatened participants, and third, that especially participants in the role of threatened patients experience the most reactance (experience of reactance, aggressive behavioral intentions, negative attitudes). We ran a multivariate analysis of variance (MANOVA) with interest (vested interest vs. neutral) and role (doctors vs. patients) as independent variables and mistrust, the three reactance scales, and the two attractiveness ratings as dependent variables.

First, for mistrust, the analyses revealed a significant main effect for interest, *<sup>F</sup>*(1,78) <sup>=</sup> 268.95, *<sup>p</sup> <sup>&</sup>lt;* 0.001, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.78. This indicates that participants in the vested-interest group perceived more mistrust toward their interaction partner than participants in the neutral group (*M* = 8.37, *SD* = 0.97 vs. *M* = 4.06, *SD* = 1.56) which supports our hypothesis. Furthermore, the main effect for role was significant, *F*(1,78) = 16.58, *p <* 0.001, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.18, indicating that in general, participants in the role of patients perceived more mistrust than participants in the role of doctors (*M* = 6.74, *SD* = 2.44 vs. *M* = 5.67, *SD* = 2.51). The interaction between interest and role was not significant, *<sup>F</sup>*(1,78) *<sup>&</sup>lt;* 1, *<sup>p</sup>* <sup>=</sup> 0.491, <sup>η</sup><sup>2</sup> *<sup>&</sup>lt;* 0.01.

Second, for reactance, the MANOVA revealed significant main effects for the factor interest on the scales *experience of reactance*, *F*(1,78) = 101.06, *p <* 0.001, η<sup>2</sup> = 0.56, *aggressive behavioral intentions*, *<sup>F</sup>*(1,78) <sup>=</sup> 33.61, *<sup>p</sup> <sup>&</sup>lt;* 0.001, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.30, *negative attitudes*, *<sup>F</sup>*(1,78) <sup>=</sup> 17.76, *<sup>p</sup> <sup>&</sup>lt;* 0.001, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.19, and people's judgment of the *attractiveness of the depot*, *F*(1,78) = 4.23, *<sup>p</sup>* <sup>=</sup> 0.043, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.05. However, people's judgment of the *attractiveness of the pills* was not significant, *F*(1,78) *<* 1, *<sup>p</sup>* <sup>=</sup> 0.520, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.01. Thus, the vested interest evoked more reactance than the neutral condition (*experience of reactance: M* = 6.82, *SD* = 2.67 vs. *M* = 2.36, *SD* = 2.23; *aggressive behavioral intentions*: *M* = 4.20, *SD* = 2.17 vs. *M* = 1.80, *SD* = 1.89; *negative attitudes: M* = 5.98, *SD* = 1.73 vs. *M* = 4.37,

<sup>1</sup>Originally, our reactance measure consisted of 18 items. A study validating this measure (Sittenthaler et al., 2015) reduced the scale to a 10 items measure and confirmed a three-factor structure consisting of *experience of reactance* (four items)*, aggressive behavioral intentions* (three items), and *negative attitudes* (three items). In the current article, we only report the validated scales.

<sup>2</sup>Due to a reliability of <sup>α</sup> <sup>=</sup> 0.58 with three items we excluded the item "Do you think that the doctor/patient shows a similar behavior in other areas?"

*SD* = 1.80; *attractiveness of depot after minus before*: *M* = −0.27, *SD* = 2.68 vs. *M* = 0.88, *SD* = 2.33; *attractiveness of pills after minus before*: *M* = −0.27, *SD* = 2.87 vs. *M* = −0.61, *SD* = 2.17). The results support our hypotheses.

Furthermore, the analyses revealed significant main effects for role on most of the scales, indicating that in general, participants in the role of patients show more *experience of reactance* (*M* = 5.94, *SD* = 3.38 vs. *M* = 3.17, *SD* = 2.61), *F*(1,78) = 39.46, *<sup>p</sup> <sup>&</sup>lt;* 0.001, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.34, more *aggressive behavioral intentions* (*M* = 3.81, *SD* = 2.59 vs. *M* = 2.15, *SD* = 1.75), *F*(1,78) = 16.05, *<sup>p</sup> <sup>&</sup>lt;* 0.001, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.17, and more *negative attitudes* (*<sup>M</sup>* <sup>=</sup> 5.60, *SD* = 1.92 vs. *M* = 4.73, *SD* = 1.88), *F*(1,78) = 5.07, *p* = 0.027, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.06, than participants in the role of doctors. We did not find a significant main effect for role on people's judgment of the *attractiveness* of both medications, *F*s(1,78) ≤ 1.86, *p*s ≥ 0.176, <sup>η</sup>2s <sup>≤</sup> 0.023.

The MANOVA also revealed a significant interaction between interest and role for people's *experience of reactance*, *<sup>F</sup>*(1,78) <sup>=</sup> 4.14, *<sup>p</sup>* <sup>=</sup> 0.045, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.05. Thus, patients in the vested-interest condition experienced the most reactance toward their interaction partner (for detailed results, see **Table 1** and **Figure 2**). However, neither the interactions for people's *aggressive behavioral intentions* and *people's negative attitudes*, *F*s(1,78) *<* 1, *p*s ≥ 0.580, η2s *<* 0.01, nor for their judgment of the *attractiveness* of both medications, *F*s(1,78) ≤ 1.77, *p*s ≥ 0.187, η2s ≤ 0.022, were significant.

# Dynamic Process – Mediation Effects

For testing the assumption that people's perception of mistrust causes a dynamic development of a reactance process, we performed a serial multiple mediation analysis using the software Process 2.11 (Hayes, 2013, model 6). The criterion for detecting a serial multiple mediation was a significant indirect effect which was computed using a 99% bias corrected bootstrap confidence interval (99% BC CI) and 10,000 bootstrap samples (Preacher and Hayes, 2008). Regression analyses revealed that the vested interest had a significant total effect on negative attitudes, *b* = −1.62, *SE* = 0.39, *t*(80) = −4.14, *p <* 0.001. The effect decreased to nonsignificance when the potential mediators mistrust, experience of reactance, and aggressive behavioral intentions had been added to the prediction, *b* = 0.12, *SE* = 0.65, *t*(80) = 0.18, *p* = 0.86. The bootstrapped indirect effect of vested interest on biased attitude via mistrust, experience of reactance, and aggressive behavioral intentions was significant, *b* = −1.04, *SE* = 0.45, BC CI [−2.91, −0.27]. In sum, the perception of mistrust resulting from a vested interest translates into an increased experience of reactance, which in turn leads to more aggressive behavioral intentions and finally results in increased negative attitudes (for the path coefficients see **Figure 3**). This result supports our hypothesis.

# DISCUSSION

In Study 1, our goal was to investigate how individuals in the role of doctors or patients react to vested interests. We examined their mistrust, experience of reactance, aggressive behavioral intentions, and biased cognitions after their interaction partner forced them to choose the depot injection instead of the pills medication. We further aimed to investigate how and why the dynamic in this interaction process develops. First, we found that participants who had been imposed the depot, indicated more mistrust than participants in a control group. Second, a serial multiple mediation revealed that the vested interest caused a chain of reactance reactions because it engendered mistrust toward the interaction partner. This suggests that the perception of mistrust significantly contributes to the person's experience, behaviors, and cognitions and therefore to a dynamically developing conflict-loop.

Maybe even more interesting, patients who had been imposed the depot partly indicated a stronger experience of reactance, more aggressive behavioral intentions, and more biased cognitions than doctors who had been imposed the depot. Derived from interdependence theory (Thibaut and Kelley, 1959; Kelley and Thibaut, 1978; Kelley et al., 2003), we suggest that this result can be explained by the different partner control of doctors and patients. Patients are usually more dependent on the doctors than vice versa. This would mean that patients experience a stronger threat if they perceive the doctor being only interested in his/her own financial benefit but not in the patients' health. Moreover, advisors and clients have different roles. People who are in the role of advisors may not be as concerned with their own decisions but may simply make a professional recommendation for the client (Jonas and Frey, 2003). In doing so maybe they try to focus more on their clients' wishes and needs than on their own. Thus, when threatened by their patients, participants in the role of a doctor might demonstrate less reactance what is important for the further development of a trustful advisor–client relationship. This may be the reasons why participants in the role of advisors seem to behave in a less defensive manner than clients when confronted with a threat.

However, is it really true that participants in the role of advisors are less defensive than participants in the role of clients? Or were our explicit measures just inappropriate to detect their real reactions? What happens if we incorporate a more implicit or indirect measure to detect reactance among participants in the role of an advisor?

# STUDY 2

In Study 2 we aimed to replicate the results of Study 1 indicating that advisors show less reactance than clients when using explicit measures. In Study 1 we could also show that threatened participants downgraded the depot which indicates a lower interest in the imposed medication. This change in people's interest may also affect their kind of information search. Vested interests might also lead people to become interested in learning more about or reading more about the advantages than the disadvantages of the restricted option and they might also become more interested in further information about the disadvantages than about the TABLE 1 | Means, standard deviations, and intercorrelations for participants' mistrust, experience of reactance, aggressive behavioral intentions, and biased cognitions in Study 1.


*Ratings were made on an 11-point scale.* +*The change in attractiveness values are the difference scores between the ratings made after minus before reading the scenario with negative scores indicating a lower attractiveness of the medication after the scenario.* ∗*p < 0.05, two-tailed;* ∗∗*p < 0.01, two-tailed.*

advantages of the imposed option. Thus, Study 2 also included a more indirect measure of reactance, people's *information search*.

Research has found that human information search is often biased (Frey, 1986; Jonas et al., 2001). If people have made a choice, they prefer information supporting their decision

over information opposing their decision, i.e., they reveal a "confirmation bias" (Jonas et al., 2001, p. 557). This phenomenon has been explored in the context of dissonance theory (Festinger, 1957). Research suggests that when people make decisions for themselves, they preferentially search for information supporting their preliminary decision, which serves to reduce post-decisional conflict. With regard to the information search among advisors, research suggests that if advisors are led by accuracy motivation they consider information supporting and contradicting their initial recommendation in a more balanced way (Jonas and Frey, 2003; Jonas et al., 2005). However, the confirmation bias increases for advisors if there is an incentive to justify their position or appear in a positive light to the client (Jonas et al., 2005). If advisors, on the other hand, are motivated to avoid a wrong decision, they reveal another kind of bias: they consider information contradicting their position more carefully and even exhibit a disconfirmation bias (Jonas et al., 2005; Mendel et al., 2009). Investigating the information search of psychiatrists when considering the diagnosis of a patient, Mendel et al. (2009) found that doctors informed themselves more about the risks than about the benefits of available treatment options. This effect can be explained by the doctor's principle of "primum non nocere" which is the motivation of avoiding harm (Mendel et al., 2009) and indeed, the results showed that the more the doctors investigated conflicting information the higher the probability that they made the correct diagnosis, whereas the probability of sticking to their wrong decision was highest when they revealed a confirmation bias (see also Kray, 2000).

We presented participants with information about the facts of the imposed and non-imposed options and with information about the person who showed a vested interest. Firstly, we predicted that due to their professional role, doctors would search for more information overall than patients (Kray, 2000; Mendel et al., 2009). Secondly, we were interested in whether doctors who are in a more powerful role than clients express their experienced reactance in a more subtle way when they have been facing a vested interest of their interaction partner. Thus, we wanted to know if they show a confirmation bias in information search by preferring supporting information about their initially recommended medication and non-supporting information about the imposed medication.

# MATERIALS AND METHODS

# Participants and Design

We had 225 participants who participated for course credit in an online study. However, we needed to exclude the data of seven participants because they correctly inferred the manipulation and eleven because they indicated that they had been unable to identify with the doctor or the patient in the scenario. Our final sample consisted of 207 students of the University of Salzburg (143 women, 62 men, 2 unspecified; *M*age = 25.07 years, *SD* = 6.31). The experiment was based on a two (interest: vested interest vs. neutral) × two (role: doctor vs. patient) betweensubjects design. Participants were instructed to take the role of a doctor (advisor) or a patient (client) and were asked to read one of four consultation paradigms described in Study 1.

# Materials and Procedure

We again assessed participants' experience of reactance, their aggressive behavioral intentions, and biased cognitions using the same items as in Study 1. Additionally to assess whether participants searched for more information supporting vs. conflicting with their initial decision preference, at the beginning of the questionnaire we asked them to indicate their spontaneous decision for one of the two medications (depot vs. pills). After the questions regarding their experience of reactance, aggressive behavioral intentions, and biased cognitions we presented them information concerning the treatment options and participants indicated whether they wanted to read it in more detail or not (see below).

#### Questionnaire – Reactance

Participants answered some items regarding their experience of reactance (four items, α = 0.96), their aggressive behavioral intentions (three items, α = 0.88), and their biased cognitions (negative attitudes: two items3 , r = 0.56; attractiveness of the depot and the pills medication before and after the scenario: each one item). All items ranged from 1 (not at all) to 10 (very much).

<sup>3</sup>Because of a bad overall reliability in Study 1, we deleted the item "Do you think that the doctor/patient shows a similar behavior in other areas?"

### Information Search

Then we presented short statements concerning the advantages and disadvantages of both medications. Each statement was either a factual information or a person-related information about the advantages or disadvantages of the medication (e.g., *factual contra depot*: "A patient survey showed: the depot injection is often perceived as unpleasant and as evoking anxiety."; *person-related (doctor) pro depot*: "You hear that patients' feedback strongly affect Dr. Boston's work and that so far she has always received positive feedback concerning the depot injection."). In sum, each participant read sixteen statements (two factual advantages and two factual disadvantages of the depot, two factual advantages and two factual disadvantages of the pills, two person-related (doctor vs. patient) advantages and two person-related disadvantages of the depot, two person-related advantages and two person-related disadvantages of the pills4 ). After reading each statement they indicated whether they wanted to read it in more detail or not. Thus, the information search consisted of the sum of statements they wanted to read more about.

In the end, doctors gave their final recommendation for a medication and patients their final decision for either the depot injection or the pills. They indicated how well they could identify with the role of a doctor/patient, and indicated the objective of the study.

# RESULTS

# Reactance

Aiming to replicate the results of Study 1, we predicted that individuals, especially patients, who face a vested interest of their interaction partner, reveal more experience of reactance, more aggressive behavioral intentions, and more biased cognitions, than non-threatened individuals. The MANOVA with interest (vested interest vs. neutral) and role (doctors vs. patients) as independent variables revealed significant interactions on all scales except of the attractiveness rating of the pills [experience of reactance: F(1,203) <sup>=</sup> 24.22, p *<sup>&</sup>lt;* 0.001, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.11; aggressive behavioral intentions: F(1,203) = 61.04, p *<* 0.001, η<sup>2</sup> = 0.23; biased cognition – negative attitudes: F(1,203) = 33.20, p *<* 0.001, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.14; biased cognition – change in attractiveness depot: F(1,2025 ) <sup>=</sup> 7.30, p <sup>=</sup> 0.007, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.04; biased cognition – change in attractiveness pills: F(1,203) <sup>=</sup> 1.42, p <sup>=</sup> 0.234, <sup>η</sup><sup>2</sup> *<sup>&</sup>lt;* 0.01]. Thus, we did not only find a significant interaction for experience of reactance like in Study 1, but this time for the other scales as well. Patients who had been imposed the depot indicated the highest reactance. Furthermore, patients also revealed the strongest decrease in the attractiveness of the imposed depot, which confirms our hypotheses (for detailed results, see **Table 2** and **Figure 4**).

# Dynamic Process – Mediation Effects

As the results of the mediation analyses from Study 1 suggest the existence of a dynamic interaction process, we performed a serial multiple mediation analysis (Process 2.11; Hayes, 2013, model 6) also in Study 2. The criterion for detecting the mediation was a significant indirect effect which was again computed using a 99% bias corrected bootstrap confidence interval (99% BC CI) and 10,000 bootstrap samples (Preacher and Hayes, 2008). Regression analyses revealed that the vested interest had a significant total effect on negative attitudes, *b* = −3.77, *SE* = 0.34, *t*(205) = −11.15, *p <* 0.001. The effect decreased to nonsignificance when the potential mediators experience of reactance and aggressive behavioral intentions had been added to the prediction, *b* = −0.68, *SE* = 0.41, *t*(205) = −1.64, *p* = 0.102. The bootstrapped indirect effect of vested interest on biased attitude via experience of reactance and aggressive behavioral intentions was significant, *b* = −1.36, *SE* = 0.34, BC CI [−2.32, −0.53]. Moreover, the bootstrapped indirect effect of vested interest on biased attitude via experience of reactance alone was also significant, *b* = −1.71, *SE* = 0.48, BC CI [−3.01, −0.53]. In sum, the experience of reactance from a vested interest translates into aggressive behavioral intentions and finally results in increased negative attitudes (for the path coefficients see **Figure 5**). This result again supports our hypothesis of a dynamic development of the reactance process.

# Information Search

First, we tested the prediction that due to their professional role, doctors search for more information than patients. The ANOVA with role (doctors vs. patients) as independent variables and the sum of all information as dependent variable revealed a significant main effect for role on people's general interest in information, *<sup>F</sup>*(1,205) <sup>=</sup> 11.23, *<sup>p</sup>* <sup>=</sup> 0.001, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.06, with doctors searching for more information than patients (*M* = 10.67, *SD* = 4.05 vs. *M* = 8.86, *SD* = 3.72).

Second, we asked whether doctors who are confronted with a vested interest of their patient express their reactance in a subtle way by showing a confirmation bias. According

<sup>4</sup>In a pretest we aimed to find clear statements for the dimensions pro depot, contra depot, pro pills, contra pills. Therefore, we asked 30 participants to rate 24 short factual statements, and 20 short person-related (either doctor- or patient-related) statements concerning their direction [scale from −5 (*contra depot/pills*) to 5 (*pro depot/pills*)], persuasiveness, and relevance [scales from 0 (*not at all*) to 10 (*very much*)]. Based on the means we chose eight factual statements and ran 2 (type of medication: depot vs. pills) × 2(information: pro vs. contra) repeated measures ANOVAs with direction, persuasiveness, and relevance separately. For the factual statements, we neither found significant effects for direction, *F*s(1,29) ≤ 2.78, *p*s ≥ 0.107, η<sup>2</sup> ≤ 0.09, nor for persuasiveness, *F*s(1,29) *<* 1, *p*s ≥ 0.583, η<sup>2</sup> ≤ 0.01. For relevance we found a marginal significant effect for information, *<sup>F</sup>*(1,29) <sup>=</sup> 3.67, *<sup>p</sup>* <sup>=</sup> 0.065, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.11, indicating that participants ascribing a higher relevance to the pro (*M* = 7.48, *SD* = 0.24) than to the contra information (*M* = 7.10, *SD* = 0.23). For the person-related statements, we neither found significant effects for direction, *F*s(1,13) ≤ 2.74, *p*s ≥ 0.122, <sup>η</sup><sup>2</sup> <sup>≤</sup> 0.17, nor for persuasiveness, *<sup>F</sup>*s(1,13) <sup>≤</sup> 3.18, *<sup>p</sup>*<sup>s</sup> <sup>≥</sup> 0.098, <sup>η</sup><sup>2</sup> <sup>≤</sup> 0.20, or relevance, *<sup>F</sup>*s(1,13) <sup>≤</sup> 2.01, *<sup>p</sup>*<sup>s</sup> <sup>≥</sup> 0.180, <sup>η</sup><sup>2</sup> <sup>≤</sup> 0.13. In sum, the statements did not differ from each other indicating that they were evaluated as clearly speaking for or against the treatment with depot/pills (factual information: *M* = 3.42, *SD* = 0.81; patient-specific information: *M* = 2.60, *SD* = 1.22; doctor-specific information: *M* = 2.05, *SD* = 1.10), were clearly evaluated as persuasive (factual information: *M* = 7.32, *SD* = 1.04; patient-specific information: *M* = 5.79, *SD* = 2.01; doctorspecific information: *M* = 5.22, *SD* = 2.26), and were clearly evaluated as relevant (factual information: *M* = 7.29, *SD* = 1.17; patient-specific information: *M* = 5.87, *SD* = 2.06; doctor-specific information: *M* = 4.43, *SD* = 1.96). Thus, we used these eight statements in the main questionnaire.

<sup>5</sup>One missing.

#### TABLE 2 | Means, standard deviations, and intercorrelations for participants' experience of reactance, aggressive behavioral intentions, and biased cognitions in Study 2.


*Ratings were made on a 10-point scale.* +*The change in attractiveness values are the difference scores between the ratings made after minus before reading the scenario with negative scores indicating a lower attractiveness of the medication after the scenario.* ∗*p < 0.05, two-tailed;* ∗∗*p < 0.01, two-tailed.*

to participant's spontaneous decision for one of the two medications in the beginning of the study, we ran two separate repeated measures ANOVAs – one analysis for the sample who chose to take the pills and who thus, may have been most affected by the imposition of the depot (*N* = 71, "imposition group"), and one analysis for the sample who chose to take the depot and who thus, may have been feeling confirmed in its decision (*N* = 136, "confirmation group"). To receive a score for the confirmation bias, we calculated a difference value between people's interest in the number of advantageous and disadvantageous pieces of information chosen.

#### Imposition Group

We inserted interest (vested interest vs. neutral) × role (doctor vs. patient) as the between factors and the differences of the medication (pills vs. depot) × type of information (factual vs. personal) as within factors into a repeated measures ANOVA.

The four-way interaction was significant, *F*(1,67) = 4.46, *p* = 0.038, η<sup>2</sup> = 0.06. Simple effects indicated that doctors in the vested-interest group, i.e., doctors who had been forced to recommend the depot injection (but previously chose to recommend the pills), committed a confirmation bias concerning the factual information: they devalued the imposed depot by showing more interest in the disadvantages than in the advantages of the depot (*M* = −0.73, *SD* = 0.88) but upgraded the non-imposed pills by showing more interest in the advantages than in the disadvantages of the pills (*M* = 0.27, *SD* = 0.70), *p <* 0.001. This upgrading of the pills differed from doctors in the neutral group who devalued the pills by showing a higher interest in disadvantages than in advantages of the pills (*M* = −0.31, *SD* <sup>=</sup> 0.86), *<sup>p</sup>* <sup>=</sup> 0.061 (see **Figure 6**). This means that doctors who had faced a vested interest of their interaction partner more strongly confirmed their decision for recommending the pills than doctors who had not faced a vested interest.

With regard to the patients, both the vested-interest and the neutral group devalued the depot (*M* = −0.46, *SD* = 0.74 and *M* = −0.60, *SD* = 0.51) and upgraded the pills (*M* = 0.25, *SD* = 0.93 and *M* = 0.33, *SD* = 0.49), *p <* 0.001 and *p* = 0.001. This means that, independent of a vested interest of their interaction partner, patients confirmed previously made decisions.

Summarized, a vested interest of the interaction partner only affected doctors but not patients. Doctors reacted to the threat by confirming their previously made decision.

#### Confirmation Group

We inserted interest (vested interest vs. neutral) × role (doctor vs. patient) as the between factors and the differences of the medication (pills vs. depot) × type of information (factual vs. personal) as within factors into a repeated measures ANOVA. The four-way interaction was not significant, *F*(1,132) *<* 1, *p* = 0.366, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.01. However, the analysis revealed two significant threeway interactions.

The three-way interaction between role, medication, and type of information, *<sup>F</sup>*(1,132) <sup>=</sup> 4.34, *<sup>p</sup>* <sup>=</sup> 0.039, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.03, indicated that concerning the factual information, both doctors and patients were more interested in the advantages of the pills (doctors: *M* = 0.02, *SD* = 0.80; patients: *M* = 0.33, *SD* = 0.94) than in the advantages of the depot (doctors: *M* = −0.52, *SD* = 0.93; patients: *M* = −0.67, *SD* = 0.76), *p*s ≤ 0.001. Furthermore, patients showed a higher interest in the advantages of the pills (*M* = 0.33, *SD* = 0.94) than doctors *(M* = 0.02, *SD* = 0.80), *p* = 0.041. Thus, the advantages of the pills were in general more interesting for people than the advantages of the depot.

The three-way interaction between interest, medication, and type of information, *<sup>F</sup>*(1,132) <sup>=</sup> 5.36, *<sup>p</sup>* <sup>=</sup> 0.022, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.04, indicated that concerning the factual information, both the vested-interest and the neutral group were more interested in the advantages of the pills (vested interest: *M* = 0.15, *SD* = 0.92; neutral: *M* = 0.19, *SD* = 0.84) than in the advantages of the depot (vested interest: *M* = −0.44, *SD* = 0.78; neutral: *M* = −0.75, *SD* = 0.91), *p*s *<* 0.001, and that the neutral group (*M* = −0.75, *SD* = 0.91) was even more interested in the disadvantages of the depot than the vested-interest group (*M* = −0.44, *SD* = 0.78), *p* = 0.037. Thus, the advantages of the pills were in general more interesting for people than the advantages of the depot.

Summarized, simple-effects in the interactions indicate that all groups (doctors, patients, vested-interest, and neutral group) that previously chose the depot showed a higher interest in the factual advantages of the pills than in the factual advantages of the depot (for detailed results, see **Figure 7**). Although they had chosen the depot, they were more interested in the advantages of their nonchosen medication, i.e., the pills.6

# DISCUSSION

First, Study 2 showed that patients who had been imposed a medication indicated a stronger experience of reactance, more aggressive behavioral intentions, and more biased cognitions than doctors who had been imposed a medication. These results that partly have also been found in Study 1 suggest that patients are more affected by the vested interest than doctors.

Second, the findings of Study 2 indicate that doctors who have not been imposed a medication, were in general more interested

<sup>6</sup>As with regard to personal information, people were in general more interested in disadvantages of both medications (pills and depot), we do not report the detailed analyses here.

in the disadvantages than the advantages of the medications (see **Figure 6**). Thus, doctors may be careful when it comes to the recommendation of a medication. However, when doctors perceived a vested interest of their patients, they confirmed their previously made decision. They showed a higher interest in the disadvantages of the imposed medication and in the advantages of the non-imposed medication. Thus, doctors behaved like patients who showed this confirmation bias no matter if they perceived a vested interest or not.

Thus, at first sight when only looking at the direct measures of reactance, it seems that doctors who are threatened by a vested interest of their interaction partner are more objective than clients, stay in their professional role, and do not defend themselves. This may be due to the doctors' professional role in which they are able to distance from the decision problem and are therefore not as impulsive as clients who are concerned with their own decision. Interestingly, Study 2 provides a more differentiated picture compared to Study 1: our data on participants' information search suggest that doctors in fact show reactance but in a more subtle manner. If the depot had been imposed they demonstrated a classical confirmation bias, i.e., they were more interested in disadvantages of the depot and advantages of the alternative medication. However, this confirmation bias only emerged concerning the factual information but not concerning the personal-related information about the medication and emerged only in doctors and not in patients. Why? Doctors may be more professional if a person is involved. Thus, when it comes to expressing one's own experience, behaving in a reactant way, or getting informed about the patient, doctors try to remain distanced and professional. Patients, on the contrary, do not have to be professional but can openly express their experience, behavior, and cognition.

# GENERAL DISCUSSION

The current research aimed to describe the dynamical development of interactions between two or more persons. With dynamical we mean subdividing the interaction process into its single stages (motivational-affective states, motivated cognitions, and motivated behaviors, see **Figure 1**) and explaining how people's reactions in an interaction mutually affect each other. The two studies presented here investigated the single stages of an emerging reactance process. After people who imagined being a doctor vs. a patient had been threatened by a vested interest of their interaction partner, we assessed their mistrust toward the interaction partner, their experience of reactance, their aggressive behavioral intentions, and their biased cognitions.

In Study 1 we found that participants who had been imposed a medication indicated more mistrust toward the interaction partner than participants in a control group. A serial multiple mediation revealed that the emerging mistrust triggered a dynamically developing reactance sequence, starting from a high experience of reactance entering aggressive behavioral intentions, and finally ending in people's biased cognitions. As we found patients being more reactant than doctors we conducted Study 2, in which we aimed to explore the advisor's weaker reactions to the threat. To do so, in addition to the classic reactance measures of Study 1, we employed an indirect measure of reactance – people's information search concerning the imposed and the non-imposed medication. The results of Study 2 suggest that doctors show reactance in a subtle way. If the depot had been imposed on them, they were more interested in disadvantages of the depot and advantages of the alternative medication. Patients also showed this confirmation bias but independent of a vested interest. They always confirmed their previously made decisions.

These results suggest that while patients disclose their reactance, doctors show reactance in a more subtle way. The reason for this may be that doctors are less dependent on the patient than vice versa. In terms of interdependence theory (e.g., Kelley et al., 2003), they are subject to less partner control than patients are and thus, have also more power over the patient. The second reason for this may be their professional role, in which they need to remain objective and unbiased. However, especially Study 2 shows us that remaining unbiased after being threatened in one's freedom is not as easy as it seems. If you take a more thorough look, then even doctors who seem to be very objective at first sight are affected by their patient's attempt to threaten their freedom.

# Theoretical and Practical Implications

With the current research we tried to investigate how interactions develop dynamically when a threat happens. Although a bunch of studies investigates how threats in social interactions affect motivation, cognition, and behavior, most of them miss the crucial step of linking them and thus showing how they mutually affect each other. However, if we aim to understand how interactions develop, we need to have a look at how one process influences another one. Inspired by interdependence theory (e.g., Kelley et al., 2003), we split the interaction process into its single stages. Going beyond interdependence theory, we further explored how the single stages mutually influence each other. This approach may help to better comprehend why interactions do not always run smoothly but sometimes provoke conflicts.

Following the recommendation of Miron and Brehm (2006, p. 16) to "consider the various implications of the theory for real world phenomena as well as continue revealing and testing its basic theoretical assumptions," we presented evidence that freedom threats in social interactions may have serious consequences. Therefore, our results are not only important for science but also for our everyday lives. Keeping in mind that threats do not only lead to sudden behavioral reactions but also to long-term biased cognitions, one can better understand own and other's reactions. This can further contribute to taking action resolving own prejudices against others and other's prejudices against oneself. Simply knowing how a threat can escalate into a conflict spiral provoking defensive actions and reactions would be the first step for the development of possible interventions.

Although the patient's behavior is the only part visible for the doctor, the reactant patient additionally has specific cognitions and reasoning processes not visible for the doctor. However, those biased cognitions regarding both, cognitions about the interaction partner and cognitions about oneself and one's attitudes, are important to consider. Therefore, the patient may derogate the doctor and assume that the doctor has prejudices against all mentally ill people. According to reactance theory, the threatened person (the patient), may also downgrade an imposed medication that he previously preferred and upgrade a not imposed medication that he previously rejected (see also, e.g., Bushman and Stack, 1996; Dillard and Shen, 2005; Bijvank et al., 2009). Our results extend reactance theory by showing that biased cognitions can also be found in threatened people's information search. We presented evidence that doctors displayed higher interest in information of the non-imposed medication compared to a lower interest in information of the imposed medication.

# Limitations and Future Research

A limitation of our studies is that we used scenarios in which people were imagining being a doctor or a patient. Such imaginations miss out the important real-life setting in which doctors are committed to their profession and patients suffer from a real disease. Although real-life settings would tell us even more, we believe that these studies are first steps into exploring interaction processes developing out of freedom threats. Thus, we found a chain of emerging stages that influenced each other – a threat of one person to another person led to an experience of mistrust, which further aroused an experience of threat and intended behaviors to restore one' freedom and finally resulted in biased attitudes. Aside from that, it would be difficult and even unethical to manipulate freedom threats in a real doctor– patient interaction. One could only try to observe interactions like these and figure out where threats happen. However, it may be worth to consider carrying out the same studies on a sample of real advisors and real patients and not only on students. Although we found that clients were very reactant if the doctor threatened them in their freedom, this might not apply in all situations. Consider a patient who is really afraid of dying and knows he does not have much medical knowledge. He or she would probably follow the doctor's orders without experiencing reactance.

Although one would assume that doctors could not be threatened in a realistic way because they do not have a direct benefit from prescribing the depot or not, our results show that doctors are threatened as well but they suffer to a lower degree than patients. The reason for this might be their lower dependence on the interaction partner. This means that the doctor is not as dependent on the patient as vice versa. If the patient, however, recognizes that the doctor is not really interested in the patient's health but only in her own financial benefit, the patient experiences a very strong threat. Besides his health being threatened, the vested interest of the doctor can also Steindl and Jonas The Dynamic Reactance Interaction

pose a psychological threat. The patient may also feel threatened in his motive of being free to decide for himself. However, as the doctor is not as dependent on the patient, she might not be as affected. Still, the doctor might be threatened in his motive of being free to advice the medication he wants to advice. The result that doctors experience a threat as well is in line with reactance theory stating that social influence attempts targeting a specific individual can pose a freedom threat (Brehm, 1966; Brehm and Sensenig, 1966; Clee and Wicklund, 1980; Brehm and Brehm, 1981). We believe that the doctors' and patients' different amount of partner control that might cause the different degrees of threat would be a valuable question for future research.

# CONCLUSION

Dealing with freedom threats that happen in our everyday interactions, the present research broadens our understanding

# REFERENCES


of how social interactions develop – from experiencing mistrust toward the interaction partner to experiencing a strong threat, to further being motivated to restore one's scope of action and finally developing biased cognitions. Our results open up possibilities on how social interactions might be viewed in a dynamical sense. Research investigating social interaction may thus pay more attention to this dynamic.

# ACKNOWLEDGMENTS

Christina Steindl was financially supported by the Doctoral College "Imaging the Mind" of the Austrian Science Fund (FWF-W1233). We are grateful to Sandra Sittenthaler for helping in the preparation of the study design and materials and to Johanna Frieler and Christina Brunner for helping in the data collection.

Jonas, E. (2015). "Fairness lohnt sich! Psychologische Facetten von Gerechtigkeit und ihr Beitrag zu Kooperation und Widerstand in sozialen Interaktionen," in *Fairness und Fairplay. Interdisziplinäre Perspektiven*, eds M. Dimitriou and G. Schweiger (Wiesbaden: Springer VS Forschung), 23–49.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2015 Steindl and Jonas. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

**OPINION ARTICLE** published: 03 February 2015 doi: 10.3389/fpsyg.2015.00040

# Motivations for responses to ostracism

#### *Eric D. Wesselmann1 \*, Dongning Ren2 and Kipling D. Williams <sup>2</sup>*

*<sup>1</sup> Department of Psychology, Illinois State University, Normal, IL, USA*

*<sup>2</sup> Department of Psychological Sciences, Purdue University, West Lafayette, IN, USA*

*\*Correspondence: edwesse@ilstu.edu*

#### *Edited by:*

*Christina Steindl, University of Salzburg, Austria*

#### *Reviewed by:*

*Nilüfer Aydin, Alpen-Adria University, Germany Michaela Pfundmair, Ludwig-Maximilians University of Munich, Germany*

**Keywords: ostracism, social exclusion, rejection, pro-social behavior, anti-social behavior**

Ostracism (being ignored and excluded) and other forms of interpersonal rejection threaten individuals' physical and psychological well-being (Williams and Nida, 2011). Researchers often use the terms ostracism, social exclusion, and rejection interchangeably, but there are theoretical and empirical debates about the differential effects of these phenomena (Smart Richman and Leary, 2009; Williams, 2009; Bernstein and Claypool, 2012). We acknowledge these debates but choose to use the term *ostracism* ubiquitously for simplicity because most of the outcomes we discuss are similar across the phenomena. Most individuals experience these threats at least once during their lives, and some individuals experience them daily (Williams, 2009).

Regardless of the mode or source by which the event occurs, ostracism threatens basic psychological needs (belonging, control, meaningful existence, and selfesteem; Nezlek et al., 2012; Wesselmann et al., in press). Williams (2009) posits a temporal structure to ostracism's effects. In Stage 1, ostracism's basic need threat is ubiquitous with few situational or dispositional moderators (Wesselmann et al., in press). Williams's model is motivationfocused; after the initial threat occurs, individuals should be motivated to recover by fortifying their threatened needs. Stage 2 focuses on cognitive and behavioral processes ostracized individuals use to recover. Stage 3 argues that chronically ostracized individuals withdraw socially and experience extreme psychological and physical damage. We will now focus on relevant Stage 2 research and then discuss suggestions for future research on Stages 2 and 3.

### **STAGE 2: REFLECTION AND RECOVERY**

Experimental data suggest that recovery can begin within minutes after ostracism occurs and participants use multiple cognitive and behavioral strategies to recover their thwarted basic needs. Individuals' *cognitive strategies* often focus on attributions for why ostracism occurred and ways to remedy the situation (Williams, 2009). Wirth and Williams (2009) found that attributions influence recovery speed: Individuals attributing ostracism to an experimentally contrived group membership recovered from ostracism quicker than individuals who attributed ostracism to a permanent group membership (i.e., gender; also race in Goodwin et al., 2010; c.f., Masten et al., 2011). Another effective strategy involves encouraging participants to recall ostracism from an outsider's (compared to first-person) perspective (Lau et al., 2009). Also, research demonstrates that priming feelings of physical invulnerability reduces the need for ostracized participants to seek various cognitive and interpersonal recovery options (Huang et al., 2013). Self-construal also facilitates recovery from ostracism: Individuals who have higher interdependent/collectivistic selfconstruals (i.e., define themselves in terms of social relationships) can recover quicker from ostracism compared to individuals who are lower in these construals (Ren et al., 2013; Pfundmair et al., 2015). Finally, reminding someone of positive social relationships, symbolic/parasocial relationships, or religious/spiritual affiliations facilitates recovery from ostracism (Gardner et al., 2005; Twenge et al., 2007; Epley et al., 2008; Derrick et al., 2009; Aydin et al., 2010, 2012; McConnell et al., 2011; Laurin et al., 2014).

Research on *behavioral strategies* focus mostly on pro- or anti-social behaviors and how they facilitate basic need recovery (Williams, 2009). Experimental research demonstrates that ostracized individuals respond more *pro*-socially than included individuals; they attend more to social information relevant to inclusion (Pickett et al., 2004; Bernstein et al., 2008; Böckler et al., 2014), work harder on group tasks (at least among women participants; Williams and Sommer, 1997), focus more on re-inclusion (Maner et al., 2007; Molden et al., 2009), and show increased sensitivity to social influence (Williams et al., 2000; Carter-Sowell et al., 2008; Riva et al., 2014b). Ostracized individuals also respond more *anti*-socially than included individuals. Ostracized individuals respond aggressively toward another person regardless of whether this person ostracized them. Researchers have measured aggression using diverse methods, such as temptations for physical and social aggression, negative evaluations, unpleasant noise, and ostensibly forcing someone to eat hot sauce (Twenge et al., 2001; Buckley et al., 2004; Warburton et al., 2006).

These two behavioral patterns seem contradictory, but Williams (2009) theorizes that each type of behavior should be linked to the specific psychological needs threatened by ostracism. Pro-social behaviors should be more likely to fortify *inclusionary* needs (belonging and self-esteem) because these behaviors are more likely to achieve re-inclusion; antisocial behaviors should be more likely to fortify *power/provocation* needs (control and meaningful existence) because these behaviors will likely provoke acknowledgement from the ostracizers (see also Gerber and Wheeler, 2009). Ostracized individuals should focus on fortifying whichever need group is most salient to them; ostracized individuals who are unlikely (or unable) to be re-included into a group should focus more on fortifying power/provocation needs (via anti-social behavior) because these needs would be easier to fortify than inclusionary needs (Williams and Wesselmann, 2011).

Experimental research supports Williams's need fortification argument, specifically for control and aggression. Warburton et al. (2006) demonstrated that fortifying ostracized participants' control need immediately after ostracism reduced their aggressive responses to the same level as included participants, whereas unfortified ostracized participants continued to aggress. Schoel et al. (2014) found that ostracized individuals' threatened control (but not the other needs) mediated the ostracism → aggression effect. Other research investigates inclusionary needs and pro-social behavior. Ostracized participants afforded re-connection opportunities behaved less aggressively than ostracized participants not afforded this option (Twenge et al., 2007; DeWall et al., 2010). Pfundmair et al. (2014) investigated the interaction between self-construal and oxytocin (a hormone typically linked to pro-social behavior) on ostracized individuals' reactions (specifically, belonging and self-esteem threat). They found that collectivistic-oriented individuals exposed to oxytocin showed reduced need threat compared to individuals exposed to the placebo, suggesting that simple hormonal cues of affiliation can provide temporary relief from ostracism (at least for individuals who emphasize social relationships in their self-concept). Finally, Bernstein et al. (2010) demonstrated that ostracized participants showed more desire to interact with new sources of affiliation than included participants; this pro-social orientation was mediated by ostracized participants' threatened inclusionary needs. Interestingly, they tested both self-esteem and belonging against each other in a multiple-mediation model and found that self-esteem was a stronger mediator than belonging, suggesting potential nuances within each need cluster that future research should investigate.

# **FUTURE RESEARCH QUESTIONS** *Paradoxical Responses?*

Future research needs to address directly these two conflicting behavioral responses. Often, researchers only give participants one behavioral option. Because of this, it is hard to rule out the possibility that ostracized participants are simply responding more extremely than included participants using whatever option they are given because it is the only option they have to fortify some of their basic needs. Some behavioral measures can be interpreted as pro- or anti-social depending on how participants respond (Gerber and Wheeler, 2009). For example, allocating high amounts of hot sauce to someone who hates spicy food fits the conceptual definition of aggression, but the experimenter instructs participants that they *have* to allocate some amount (whether they choose a small or large amount). Thus, participants who give a small amount could be interpreted as either being *less* aggressive or potentially *more* pro-social because they are obeying the experimenter nominally but also not subjecting the target to unnecessary discomfort. Further, how do researchers categorize participants who actively choose not to allocate any hot sauce at all? Is this simply lack of aggression or also an independent pro-social behavior toward the target?

Recent evidence suggests these two responses might coexist: ostracized individuals seek social connections to an interacting partner (pro-social) and devaluate the same target (anti-social) simultaneously (Sommer and Bernieri, 2014). Gerber and Wheeler (2009)'s meta-analysis found that when forced to choose, participants usually favored anti-social (i.e., control-focused) options. Domachowska et al. (2014) found that ostracized participants preferred higher impact responses (either pro- or antisocial), toward new individuals. They only preferred anti-social responses toward the ostracizers. These findings suggest a complex relationship between behaviors, need satisfaction, and contextual factors. Future research should consider merging these findings with other theoretical models. For example, the Meaning Maintenance Model (Heine et al., 2006) argues that when individuals experience threats to their sense of meaning, they seek recovery either through re-affirming meaning in the threatened domain or indirectly by affirming a symbolically-related domain. Ostracism research typically focuses on threats to four basic needs that are conceptually distinct but inter-correlated (Williams, 2009). It is possible that ostracized individuals can focus on fortifying one specific need (or cluster) and indirectly fortify the others by proxy. Thus, anti-social behaviors may fortify control the most but also may fortify the other three needs in smaller degrees.

Other threat-focused models [e.g., rejection-based threats, Smart Richman and Leary (2009), or threats more broadly, Jonas et al. (2014)] may also offer interesting ways of understanding when and why individuals respond pro- or anti-socially. Jonas et al. (2014) argue that there is a temporal structure to psychological reactions to threat, beginning with anxiety and inhibition or avoidance-based behaviors. Their theorized immediate effects converge with Williams's (2009) argument for Stage 1 reactions to ostracism. Jonas et al. (2014) argues that certain reactions move beyond immediate inhibition-based responses and facilitate approach-oriented behaviors; these behaviors can either address the threat directly or symbolically. The research focused on Williams's (2009) Stage 2 can be re-framed within this theoretical context. Both pro-social and anti-social behaviors could be considered approach-oriented; although pro-social behaviors are the most likely to achieve re-inclusion and primarily fortify the inclusionary needs, anti-social behaviors can also facilitate recovery via the power/provocation needs. The research on cognitive strategies can also be considered approach-oriented in that they either actively help individuals recover need satisfaction through attributional reframing or address the threat symbolically via affirming one's other interpersonal or parasocial relationships.

Researchers could also measure theoretically meaningful individual difference variables to test potential moderation of the anti-social/pro-social paradox. For example, ostracized participants who have a higher dispositional need to belong (Leary et al., 2013) should be more likely to favor pro-social responses (linked to inclusionary needs) than anti-social responses (linked to power/provocation needs). Also, ostracized individuals who are oriented more toward long-term rather than shortterm future outcomes respond with higher pro-social behavior than those who favor short-term outcomes instead (Balliet and Ferris, 2013). It is possible that a long-term focus would allow ostracized individuals an edge in overcoming an initial impulse to respond with anger and aggression, thus making inclusionary needs the primary focus. Other research demonstrates that socially anxious individuals recover more slowly (Zadro et al., 2006) and respond less pro-socially when ostracized (Mallott et al., 2009). As social anxious individuals find any social interaction aversive, it is possible these individuals would be less likely to want re-inclusion after ostracism and thus inclusionary needs would be less salient to fortify.

Some individual differences should moderate anti-social responses. Individuals higher in rejection sensitivity (i.e., the tendency to expect and easily perceive rejection; Ayduk et al., 2008) or destiny beliefs (Chen et al., 2012) respond more aggressively than ostracized participants lower in these individual differences. Ostracized individuals high in these latter two variables likely assume that their treatment is consistent with how they will be treated in future interactions; thus the power/provocation need cluster should be most salient to them, facilitating anti-social over pro-social behaviors. Interestingly, ostracized individuals who are high in narcissism also respond more negatively to ostracism than other individuals (Twenge and Campbell, 2003). Narcissistic individuals should expect ubiquitous inclusion because of their inflated self-esteem and may assume the ostracism was anomalous. However, this attribution would not explain their increased aggression using our *future expectations* argument. Wesselmann et al. (2010) offer another possibility: Unexpected ostracism provokes more aggression than expected ostracism, likely because it suggests an inability to read social cues accurately which in turn suggests threats to future social inclusion. Narcissists typically prioritize being admired over being liked (Morf and Rhodewalt, 2001), but they would need to accurately read social cues regardless. Thus, unexpected ostracism would still be more threatening than expected ostracism.

#### *Chronic Ostracism*

Williams (2009) developed Stage 3 by synergizing qualitative interviews of chronically ostracized individuals with other related psychological topics. Chronically ostracized individuals who find their attempts at recovery (Stage 2) continually thwarted should become resigned to their fate and face extreme negative consequences (i.e., feelings of alienation, depression, helplessness, and meaninglessness). Further, chronically ostracized individuals may seek solitude or otherwise disengage from relationships to avoid future ostracism or other unpleasant social interactions (Leitner et al., 2014; Wesselmann et al., 2014). We consider this social withdrawal a *flight* response (compared with *fight* responses, i.e., proand anti-social behaviors; Williams, 2007; see also Smart Richman and Leary, 2009; Pfundmair et al., 2015). Additionally, a *freeze* response (Williams, 2007) may encapsulate the cognitive, physical, and affective numbness that sometimes characterizes chronic loneliness and extreme social pain manipulations (e.g., being told one will spend their entire lives alone; DeWall and Baumeister, 2006; Blackhart et al., 2009; Bernstein and Claypool, 2012; Riva et al., 2014a). Few studies have considered these latter two options regarding psychological need threat and recovery, especially within Williams's (2009) temporal model. This area begs for future research and should consider combining Williams's (2009) model with other threat-based models (e.g., Smart Richman and Leary, 2009; Jonas et al., 2014) to derive predictions regarding *flight* and *freeze* responses.

Chronically ostracized individuals may also be susceptible to recruitment by predatory/extreme groups because these groups may offer a last bastion of inclusion (Wesselmann and Williams, 2010; Williams and Wesselmann, 2011). Because ostracized individuals are susceptible to social influence involving mundane behaviors it may also occur for extreme behaviors. For example, Kruglanski et al. (2009; Kruglanski and Orehek, 2011) argue terrorists interpret their actions as a quest for *meaning*/significance (also Jonas et al., 2014). Individuals (or groups) who believe that they have been humiliated, marginalized, or ostracized by larger communities may be attracted to extremist groups endorsing terrorist actions to regain a sense of significance or *control* (Kruglanski, 2003). Other research examining mass violence in schools suggest that that actual or perceived ostracism was a primary motivator for perpetrators (Leary et al., 2003) so it is reasonable to extend this rationale to understanding violence perpetrated by groups of disaffected individuals such as terrorist organizations. These questions offer exciting research opportunities for understanding how ostracism's psychological threats can inspire the worst in individuals and how society can combat these effects.

#### **REFERENCES**


elicits emotional reactions but neither causes immediate distress nor lowers self-esteem: a metaanalytic review of 192 studies on social exclusion. *Pers. Soc. Psychol. Rev.* 13, 269–309. doi: 10.1177/1088868309346065


of social motivations. *Pers. Soc. Psychol. Rev.* 10, 88–110. doi: 10.1207/s15327957pspr1002\_1


ownership. *J. Pers. Soc. Psychol.* 101, 1239–1252. doi: 10.1037/a0024506


*on Social Isolation, Social Withdrawal, and Being Alone,* eds R. J. Coplan and J. Bowker (Malden, MA: Wiley-Blackwell), 224–241.


*Group Process. Interpers. Relat.* 12, 111–127. doi: 10.1177/1368430208098780

Zadro, L., Boland, C., and Richardson, R. (2006). How long does it last? The persistence of the effects of ostracism in the socially anxious. *J. Exp. Soc. Psychol.* 42, 692–697. doi: 10.1016/j.jesp.2005.10.007

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 29 October 2014; accepted: 08 January 2015; published online: 03 February 2015.*

*Citation: Wesselmann ED, Ren D and Williams KD (2015) Motivations for responses to ostracism. Front. Psychol. 6:40. doi: 10.3389/fpsyg.2015.00040*

*This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology.*

*Copyright © 2015 Wesselmann, Ren and Williams. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# **Victimization experiences and the stabilization of victim sensitivity**

#### *Mario Gollwitzer\*, Philipp Süssenbach and Marianne Hannuschke*

*Department of Psychology, Philipps-University Marburg, Marburg, Germany*

People reliably differ in the extent to which they are sensitive to being victimized by others. Importantly, "victim sensitivity" predicts how people behave in social dilemma situations: Victim-sensitive individuals are less likely to trust others and more likely to behave uncooperatively—especially in socially uncertain situations. This pattern can be explained with the sensitivity to mean intentions (SeMI) model, according to which victim sensitivity entails a specific and asymmetric sensitivity to contextual cues that are associated with untrustworthiness. Recent research is largely in line with the model's prediction, but some issues have remained conceptually unresolved so far. For instance, it is unclear why and how victim sensitivity becomes a stable trait and which developmental and cognitive processes are involved in such stabilization. In the present article, we will discuss the psychological processes that contribute to a stabilization of victim sensitivity within persons, both across the life span ("ontogenetic stabilization") and across social situations ("actual-genetic stabilization"). Our theoretical framework starts from the assumption that experiences of being exploited threaten a basic need, the need to trust. This need is so fundamental that experiences that threaten it receive a considerable amount of attention and trigger strong affective reactions. Associative learning processes can then explain (a) how certain contextual cues (e.g., facial expressions) become conditioned stimuli that elicit equally strong responses, (b) why these contextual untrustworthiness cues receive much more attention than, for instance, trustworthiness cues, and (c) how these cues shape spontaneous social expectations (regarding other people's intentions). Finally, avoidance learning can explain why these cognitive processes gradually stabilize and become a trait: the trait which is referred to as victim sensitivity.

#### *Edited by:*

*Eva Jonas, University of Salzburg, Austria*

#### *Reviewed by:*

*Tanja M. Gerlach, Georg-August-University Goettingen, Germany Michèlle Bal, Leiden University, Netherlands*

#### *\*Correspondence:*

*Mario Gollwitzer, Department of Psychology, Philipps-University Marburg, Gutenbergstrasse 18, 35037 Marburg, Germany mario.gollwitzer@uni-marburg.de*

#### *Specialty section:*

*This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology*

> *Received: 01 December 2014 Accepted: 27 March 2015 Published: 14 April 2015*

#### *Citation:*

*Gollwitzer M, Süssenbach P and Hannuschke M (2015) Victimization experiences and the stabilization of victim sensitivity. Front. Psychol. 6:439. doi: 10.3389/fpsyg.2015.00439* **Keywords: victim sensitivity, personality development, stabilization, cognitive processes, social justice**

# **Introduction**

Imagine the following situation: your colleague asks you to do a favor for her, such as switching shifts with her because she says she needs to see a doctor. You agree and take her early-morning shift. A couple days later, you learn that your colleague never saw a doctor (and never intended to do so); instead, she needed to sleep in that day because she had been partying the other night. What you probably feel in that very moment is a mixture between anger, moral outrage, disappointment, helplessness, and regret. You trusted your colleague, but your trust was betrayed, and you will most probably decide that you will never trust her again—and maybe you will not even trust any other of your colleagues. The incident has probably made you more sensitive to the fact that other people can exploit your goodwill.

Gollwitzer et al. Stabilization of victim sensitivity

Although such incidents of betrayed trust are certainly aversive to everyone, the extent to which people react emotionally to such an incident and ruminate about the injustice that it implies differs considerably between individuals: some people merely feel a sting of anger which quickly dissolves as time goes by. Others experience a powerful and overwhelming range of negative emotions and ruminate for a long time about the incident and what it says about them. The latter kind of individuals can be referred to as having a strong sensitivity to injustice from the victim's perspective (or "victim sensitivity"). Victim sensitivity is a personality trait that has originally been developed to measure individual differences in the justice motive (Schmitt et al., 1995; Schmitt, 1996). Later, it has been conceptualized as one of four perspectives from which people can be sensitive toward injustice (the other perspectives are: observers, beneficiary, and perpetrator; cf. Schmitt et al., 2010). Unlike the other perspectives, victim sensitivity has been found to predict suspicious cognitions, social mistrust, egoism, and uncooperativeness (Fetchenhauer and Huang, 2004; Gollwitzer et al., 2005; Gollwitzer and Rothmund, 2011). According to a model that aims at explaining these effects (i.e., the "sensitivity to mean intentions" or SeMI model; cf. Gollwitzer and Rothmund, 2009; Gollwitzer et al., 2013), victimsensitive individuals can be characterized as harboring a latent fear of being exploited and as being chronically hypersensitive to cues that are associated with untrustworthiness. From this perspective, their antisocial and egoistic behavior can be conceptualized as a defensive reaction to prevent exploitation: victim-sensitive individuals behave uncooperatively toward others because they expect others to behave uncooperatively toward them.

Many empirical findings are in line with that notion: Victimsensitive individuals are more sensitive to even slight cues of untrustworthiness (Gollwitzer et al., 2009, 2012), even if these cues have only limited prognostic validity for a situation in which one might be exploited (Rothmund et al., 2011, 2015). Victimsensitive individuals are more likely to behave aggressively (Bondü and Krahé, 2014) and destructively, especially if they sense a risk of being exploited (Schmitt and Mohiyeddini, 1996; Mohiyeddini and Schmitt, 1997; Schmitt and Dörfel, 1999). They make more egoistic choices in social dilemmas (Fetchenhauer and Huang, 2004), and are less willing to help others in need (Gollwitzer et al., 2005), both in interpersonal and in intergroup situations (i.e., when there is a certain danger that the goodwill of one's ingroup might be exploited by an outgroup; Süssenbach and Gollwitzer, 2015). They are more envious and more jealous (Schmitt et al., 2005), less willing to accept apologies from their partners (Gerlach et al., 2012), and more likely to oppose political reforms because they think that politicians act out of ulterior motives (Agroskin et al., in press).

As any personality trait that deserves this attribute, victim sensitivity remains relatively stable over time: In a representative sample of German adults (mean age: 47.6 years), 60% of the truescore variance in victim sensitivity, measured at three occasions with a time lag of 2 years, can be attributed to a latent trait, whereas only 33% of the true-score variance can be attributed to occasionspecific influences (Schmitt et al., 2005). In line with this finding, several studies have shown that victim sensitivity reliably predicts social behavior in lab experiments even though victim sensitivity was measured weeks or even months before the lab experiment took place (e.g., Gollwitzer and Rothmund, 2011; Gollwitzer et al., 2012). This stability is remarkable, and it demands a psychological explanation. What makes victim sensitivity a stable trait? As we will see, addressing this question requires an elaborate theoretical framework assuming systematic interactions between social experiences, cognitive representations, and learning processes. We will sketch such a theoretical framework in the present article.

The overarching question—how victim sensitivity stabilizes consists of two facets or sub-questions. A first sub-question concerns the "ontogenetic" development of victim sensitivity: when do individuals begin to become victim-sensitive, and what are the psychological processes that catalyze the emergence and stabilization of victim sensitivity during the life course? Our attempt to provide answers to this question bears on insights from lifespan developmental psychology and personality psychology. The second sub-question concerns the "actual-genetic" development of victim sensitivity: how does victim sensitivity stabilize in the course of specific social situations in which justice and trustworthiness are an issue—situations like the one we described at the beginning of this article? How do victim-sensitive individuals perceive and interpret such situations, and how do these perception and interpretation processes contribute to a further stabilization of victim sensitivity? Our attempt to provide answers to this particular question mainly refers to research on associative learning and social cognition. We believe that the General Process Model of Threat and Defense (Jonas et al., 2014) is particularly suitable to explain how victim-sensitive individuals react to cues associated with untrustworthiness in their social worlds.

# **When and How Does Victim Sensitivity Begin to Emerge and Stabilize?**

The "SeMI" model assumes that victim sensitivity is rooted in a specific cognitive dissonance: the dissonance between a need to trust others and a stable expectation that others are not trustworthy (Gollwitzer and Rothmund, 2009). According to the SeMI model, victim-sensitive individuals would love to live in a world in which other people can be trusted, in which the risk of being exploited is close to 0, and in which cooperation is always likely to pay off for everybody in the end. However, at one or several points in their lives, these individuals have experienced that other people are not as trustworthy and as reliable as they had hoped. We assume that such victimization experiences establish the basis for developing victim sensitivity. More concretely, we hypothesize that if victimization experiences constitute "critical" life events and if these events are coped with in a dysfunctional way, victim sensitivity is likely to increase and stabilize. Victimization experiences can have many different faces. Victimization can mean emotional or even physical abuse, betrayal of trust, or social rejection. All of these different experiences have one thing in common: they thwart a particular need, the need to trust.

# **The Need to Trust**

The need to trust other people has been conceptualized as one of the five "core social motives" (Fiske, 2009). To trust means to believe in other people's trustworthiness, that is, in their abilities, their integrity, and—most importantly—their benevolence (cf. Mayer et al., 1995). Trusting others is not only beneficial; it is essential for maintaining relationships and contributing to social groups. Trust helps us master uncertain or novel situations; it is a key component in many social interactions, from bargaining to loving, and it is considered to be at the roots of economic systems, the core of social capital, and the driving machine of democratic societies (Coleman, 1990; Putnam, 2000).

Integrity and benevolence are especially relevant in interdependence situations, that is, when the effect of one's own behavior on the desirability of different outcomes crucially depends on the behavior of other people (Thibaut and Kelley, 1959; Kelley and Thibaut, 1978). One particular type of interdependence situation is the "social dilemma" (cf. Komorita and Parks, 1995), in which one's own willingness to cooperate with others or to contribute to a common good might be exploited by others. Typical social dilemmas are the prisoner's dilemma, the public goods dilemma, or the trust game. The trust game, for instance, consists of two players (cf. Berg et al., 1995). One player, the "truster," can decide to entrust a certain amount of his or her endowment to the other player. This amount is then multiplied by the experimenter and transferred to the other player (the "trustee"), who can then decide to split the total amount or to keep it all for him-/herself. The principal is: trusting one's partner can benefit both players, but only if the "trustee" is cooperative. The situation described at the beginning of this article is a typical "trust game" situation: your colleague asks you for a favor, and your willingness to help her might either be exploited (which was the case in this example) or rewarded because you actually helped her in a difficult situation. Trust is the most important predictor of one's behavior in these kinds of games (e.g., Pruitt and Kimmel, 1977; De Cremer, 1999), and distrust (due to a fear of being exploited) strongly predicts one's unwillingness to cooperate (Coombs, 1973; Orbell and Dawes, 1981; Kerr, 1983). Given that trust is so immensely functional, both on the interpersonal as well as on the intergroup level, it makes sense to assume that trusting others is something that people are motivated to do in general.

Theories of psychosocial development echo the notion that trust is a basic human motive and that the opportunity to lead a happy, healthy life depends on whether people have developed a general sense of trust in their social worlds. Erikson's (1950, 1959) theory of life tasks (and their resolution) assumes that the very first task in life is to develop trust in a caregiver. A toddler whose basic needs (such as food, warmth, and closeness) are thwarted is—according to this theory—likely to develop a deep sense of mistrust, anxiety, and insecurity in later life. In a similar vein, attachment theory (Bowlby, 1982, 1988) also focuses strongly on the infant-caregiver bond and highlights the importance of support and caregiving processes for the development of trust and for the quality of intimate relationships in later life. More precisely, attachment theory posits that early parent–child interactions provide the basis for the development of inner working models (Bowlby, 1982) by forming expectations regarding future interactions. Inner working models correspond to mental representations of oneself, of others, and of relationships in general. These

representations result in attachment patterns, which can be qualitatively categorized into "secure" vs. "insecure" attachment styles (e.g., anxious/ambivalent, anxious/avoidant, and disorganized; Ainsworth et al., 1978). Notably, "insecure" attachment styles are associated with representations of others as being untrustworthy and of oneself as being incapable (and/or unworthy) of obtaining others' cooperation.

Taken together, these theories imply that the capability (or the willingness) to trust others as an adult may depend strongly on the kind of experiences people have had in their childhood. However, this does not necessarily mean that generalized expectations regarding other people's trustworthiness crystallize in early childhood. Empirical findings rather suggest that social trust stabilizes later—especially between early and late adolescence (e.g., Flanagan and Stout, 2010). Thus, adolescence may be considered a critical period in life in which social trust crystallizes and in which people shape their general views about the trustworthiness of other people in accordance with the kind of experiences they had. Additionally, findings from life-span developmental psychology have shown that parental influences on the child's personality development decrease gradually during late childhood and especially during early adolescence, whereas "extra-familial" influences, such as peers, friends, and especially intimate partners, become increasingly relevant (Caspi, 1998).

# **Victimization Experiences**

Social experiences are likely to shape the formation of trust and expectations regarding the trustworthiness of others. The question is which kinds of social experiences have the potential to affect these expectations. We assume that expectations concerning other people's *un*trustworthiness are learned via experiences of victimization (cf. Baumert and Maltese, 2014). These experiences could include *direct* as well as *observed* victimization.

## Direct Experiences of Victimization

Childhood and adolescence are rife with situations that challenge the notion that our fellow humans' intentions are universally good and benevolent. In early adolescence, such victimization experiences can include physical or emotional abuse (Björkqvist et al., 2011), (cyber)bulling (König et al., 2010), or unfair treatment by authorities (Pretsch et al., in press). These situations imply violations of fairness standards—standards of distributive fairness (e.g., equality, equity, or need), of procedural fairness (e.g., the opportunity to "voice" one's opinion), or of interactional fairness (e.g., the right to be treated respectfully). We assume that such violations, especially if they occur repeatedly and if they constitute "critical" life events (see below), contribute to the development and stabilization of victim sensitivity during childhood and adolescence. In addition, experiences of *social rejection*—that is, being excluded from a social relationship or social interaction—are likely to contribute to the development of victim sensitivity as experiences of social rejection can advance generalized negative expectations concerning others' trustworthiness. Relevant experiences of social rejection include parental rejection, but also peer rejection or indirect bullying (cf. Rivers and Smith, 1994; Ettekal and Ladd, 2015).

# Observed Experiences of Victimization

Although some degree of directly experienced victimization is probably necessary for the development of victim sensitivity, experiences of victimization that are observed from a third-party perspective are likely to play a role as well. Observing social rejection, interpersonal transgressions, and violations of fairness standards might be just as suited to form generalized negative expectations concerning others' trustworthiness as actually experiencing them. Drawing on research on vicarious traumatization (McCann and Pearlman, 1990), observed experiences of victimization might be particularly influential under conditions that promote empathy with the victim, for instance, when a family member or one's best friend is bullied, exploited, or otherwise treated badly. Notably, observed victimization of (significant) others may elicit moral outrage and motivate observers to fight against injustice on behalf of the victim—but these observations may nonetheless make observers more sensitive to victimization. A special instance of observed victimization is witnessing injustice in the media. Media consumption can have sustainable effects on normative beliefs, values, and selfas well as world views (Huesmann and Guerra, 1997; Möller and Krahé, 2009). For example, Rothmund et al. (2015) have recently demonstrated that exposure to violent video games at the age of 14 can contribute to a decrease in interpersonal trust 1 year later. These findings suggest that not only directly experienced, but also indirectly experienced confrontations with violence and untrustworthiness (e.g., in the media) can influence adolescents' trustworthiness expectations (see also Rothmund et al., 2013).

#### Victimization Experiences as Critical Life Events

Building on research from life-span developmental psychology, certain victimization experiences—both directly experienced and indirectly observed ones—can be considered "critical" life events. Critical life events are specific kinds of stressors that can be differentiated from "normal" life events by several characteristics (see below; Filipp and Aymanns, 2010). Among these are (1) the extent to which the event is informative about oneself (i.e., relevant for one's self-concept or self-esteem), (2) the extent to which the event interferes with plans and reduces the freedom to act, (3) the unpredictability, and (4) the uncontrollability of the event. The more a victimization experience is self-relevant, goalobstructing, unpredictable, and uncontrollable, the more likely it will have a strong impact on general beliefs about trustworthiness and the stabilization of those beliefs. Again, not only directly experienced instances of victimization, but also indirectly observed instances of victimization can constitute critical life events that can shape a person's dispositional untrustworthiness expectations. For instance, learning that one's best friend had been exploited and cheated upon by his or her partner for years can reduce one's trust into others—maybe even to the same extent as having suffered exploitation oneself can do.

In addition, individual characteristics, vulnerabilities, and resources (e.g., self-concept aspects, individual norms, sensitivities, interpersonal integration, opportunities for social support, etc.) are relevant for how a person copes with the event. The extent to which a particular victimization experience shapes trustworthiness expectations (and, thus, promotes the stabilization of victim sensitivity) thus depends on characteristics of the event itself *in conjunction with* characteristics of the person.

# **Social Information Processing Patterns**

One such person characteristic is how people tend to perceive, interpret, and react to social situations. The social informationprocessing (SIP) model of children's social adjustment (Crick and Dodge, 1994) assumes that these perceptions, interpretations, and reactions to social events are critically influenced by so-called "data base" information stored in memory. This "data base" consists of general social knowledge structures such as inner working models of relationships (Bowlby, 1982), cognitive schemas, selfconcepts, and behavioral scripts (Schank and Abelson, 1977). When confronted with particular social situations, individuals often rely on this social knowledge. Thus, the "data base" critically influences how cues are perceived and interpreted and how people react toward these cues. And, in the sense of a feedback loop, social situations and their outcomes may stabilize and reinforce this social knowledge if the outcomes are consistent with prior expectations.

The notion of a "data base" in the SIP model (Crick and Dodge, 1994) is perfectly compatible with the SeMI model (Gollwitzer and Rothmund, 2009; Gollwitzer et al., 2013). The SeMI model proposes that being confronted with contextual cues associated with untrustworthiness evokes a "suspicious mindset" among victim-sensitive individuals. Past experiences of betrayal, rejection, or unfair treatment (which, according to the SIP model, are stored in a person's "data base") thus contribute to a generalized expectation that people are not trustworthy and unreliable, an attributional bias including a heightened availability of hostile interpretations of others' intentions, and a stabilized behavioral script that favors uncooperativeness in social exchange situations. As we will discuss in Section "How Does Victim Sensitivity Perpetuate Itself Across Social Situations?", the way victim-sensitive individuals perceive, interpret, and react to social encounters in which untrustworthiness cues are present reinforces their cognitive schemas, and thus, their dispositional victim sensitivity even further.

# **Ontogenetic Stabilization Processes**

In the previous paragraphs we have discussed which kinds of victimization experiences—in combination with particular personal characteristics—are likely to contribute to the emergence and stabilization of victim sensitivity during childhood and adolescence. We will now discuss the processes that may be useful to explain how victim sensitivity stabilizes "ontogenetically" over time. First, we will discuss self-stabilization and environment stabilization as two important sources of stabilization according to life-span personality psychology (e.g., Lang et al., 2006). Next, we will discuss person-environment transaction processes and their relevance for the stabilization of victim sensitivity.

#### Self- and Environment Stabilization

Personality theories focus mainly on three different sources for stabilization: (1) an increasing self-stabilization, (2) an increasing stabilization due to a more stable environment, and (3) a stabilizing contribution of the genome.<sup>1</sup> *Self-stabilization* refers to the stabilization of self-relevant knowledge, one's self-concept, over time (Kagan, 1980). Victim-sensitive individuals might develop a "victim self-concept," which includes self-related views such as "I am easy prey" or "I am a person who attracts the attention of bullies;" and the stabilization of such a self-concept may, in turn, increase (and stabilize) one's sensitivity to victimization. *Environment stabilization*, on the other hand, means that social environments become increasingly stable across the life span, which, in turn, also has a stabilizing effect on one's personality. Self- and environment stabilization processes are not independent of each other; nevertheless, personality *→* environment effects can be empirically differentiated from environment *→* personality effects via longitudinal studies (e.g., Asendorpf and Wilpers, 1998). In general, "core" personality traits (such as the "Big Five") have a stronger effect on the environment than *vice versa*, whereas "surface" personality traits (such as self-worth or loneliness; cf. Asendorpf and van Aken, 2003) are more likely to be shaped by environments. For instance, Asendorpf and van Aken (2003) found that extraversion (a "core" personality trait) predicted changes in social relations (e.g., increased support from peers), but not *vice versa*; changes in global self-worth or loneliness (two "surface" traits), however, were predicted by social relations, but not *vice versa*. Victim sensitivity can be conceptualized as having both "core" and "surface" characteristics. Thus, personality *→* environment effects *of* victim sensitivity are likely to be as strong as environment *→* personality effects *on* victim sensitivity.

### Person-Environment Transactions

Dynamic-interactionistic approaches explain the stabilization of personality by an increasing "fit" between persons and the environments they find themselves in (Caspi, 1998). According to Caspi and Roberts (1999, 2001), this increase in fit is a function of four potential "transactions:" (1) reactive transactions, (2) evocative transactions, (3) selective transactions, and (4) manipulative transactions. We will now discuss these transactions—and their relevance for the stabilization of victim sensitivity in particular—in more detail.

*Reactive transaction* refers to the fact that different individuals react differently to the same objective situation. As the SIP model (Crick and Dodge, 1994) as well as social-cognitive personality theories (e.g., Bandura, 1999; Cervone and Shoda, 1999; Shoda and Mischel, 2000; Fleeson, 2001) suggest, cognitive schemas and behavioral scripts shape how a person perceives, attributes, and reacts to social situations (see also Social Information Processing Patterns). In turn, consistently applying these perceptions, attributions, and reactions also reinforces—and, thus, stabilizes—the schema. Consistently attributing "mean intentions" to others reinforces a person's victim sensitivity. In other words, schema-

*Evocative transactions* refer to the processes by which people elicit reactions from others that are consistent with their *a priori* expectations. This stabilizes these expectations. Stated differently, people's behavioral patterns create a consistency in other people's reactions toward them; a "self-fulfilling prophecy." If victim-sensitive individuals perceive and interpret situations against the background of their negative assumptions (others' untrustworthiness) and react accordingly (e.g., uncooperatively), others may react to this behavior in a similar way (e.g., uncooperatively), which, in turn, confirms the negative beliefs that victimsensitive individuals have about other people's untrustworthiness (see also How Does Victim Sensitivity Perpetuate Itself Across Social Situations?).

*Selective transactions* refer to the active selection of environments. Based on their individual preferences, attitudes, and competences, people actively seek out environments that "fit" their personality. For instance, adolescents prefer peers that are similar to themselves; this preference, in turn, stabilizes behavioral dispositions due to social reinforcement (Newcomb et al., 1993; Harris, 1995). Victim-sensitive individuals may thus select friends, partners, colleagues, etc., who are similarly suspicious about others' intentions as they are. This "confirms" the correctness of their (negative) assumptions and stabilizes them accordingly.

Finally, *manipulative transactions* involve active behaviors that establish environments which are consistent with one's own individual experiences and behaviors. Victim-sensitive individuals might influence their environment (their friends, colleagues, relatives, and children) to become just as suspicious as they are. By manipulating their environment in this way, victim-sensitive individuals therefore "create" social relationships that are in line with their own expectations, which, in turn, stabilizes their victim sensitivity even further.

According to Caspi (1998; see also Caspi et al., 1989), these four transactions can influence person-environment fit both in a direct and in a more indirect way. The indirect way describes a cumulative effect over a longer period of time. The latter one is also referred to as the principle of "cumulative continuity." It assumes that the possibility to establish a person-environment fit increases as one gets older. This implies that the stability of personality traits increases as a function of our capacity to select and control the environments we live in.

#### **Conclusion**

The arguments we discussed and the theories and studies we reviewed so far can be used to describe a model which describes the "ontogenetic" stabilization of victim sensitivity (see **Figure 1**). We started this discussion by referring to the "need to trust" as a core social motive that is likely to be innate and that requires attention and satisfaction already at very early age. As any other motive, the "need to trust" may differ between individuals, but a certain level of this need can most likely be found in all humans. Nevertheless, people with a strong need to trust may be particularly likely to develop a high sensitivity to victimization later in life.

Given its strong motivational component, people become sensitive to instances in which the need to trust is thwarted. We

<sup>1</sup>Genome-related stabilization effects will not be further discussed in this text since this would go beyond the scope of this article. It should be noted, however, that twin studies have demonstrated that a considerable amount of the variance in social trust is accounted for by an additive genetic component (e.g., Oskarsson et al., 2012).

referred to these instances as victimization experiences. Victimization can be directly experienced or observed from a third-party perspective. More importantly, victimization experiences can constitute "critical" life events if they are (a) self-relevant, (b) goalobstructing, (c) unpredictable, and (d) uncontrollable. Depending on characteristics of the person (i.e., vulnerabilities, sensitivities, opportunities for social support, etc.) and—especially—on habitual tendencies to perceive, interpret, and react to social situations (which, in turn, are rooted in social knowledge structures, the "data base"), victimization experiences shape future expectations regarding other people's trustworthiness. These expectations become increasingly stable via self- and environmental stabilization, and, especially, via person-environment "transactions." Stabilized and generalized untrustworthiness expectations in conjunction with a strong need to trust make a person dispositionally sensitive to victimization—the "dependent variable" in our model (see **Figure 1**). Victim sensitivity, in turn, feeds back into the "data base;" that is, victim sensitivity shapes how people perceive, interpret, and react to similar situations containing similar cues (in the SeMI model, this is referred to as the "suspicious mindset;" cf. Gollwitzer and Rothmund, 2009; Gollwitzer et al., 2013).

We have also argued that late childhood and early adolescence may be a particularly critical age for the formation and stabilization of victim sensitivity, because both (a) the need to trust others—especially peers, friends, and partners—and (b) the likelihood of being confronted with instances of victimization are particularly high during this phase. To date, there are no empirical studies in which the stabilization of victim sensitivity in adolescence is systematically investigated. The only study that may be informative in this regard has been published by Bondü and Krahé (2014). These authors have shown that victim sensitivity can be reliably assessed and distinguished from other constructs by the age of 9. In this study, the predictive effect of victim sensitivity over and above other factors (e.g., rejection sensitivity) on aggressive behavior was examined in a large sample with ages ranging between 9 and 19 years. Victim sensitivity turned out to be the strongest predictor of various forms and functions of aggressive behavior. Interestingly, victim sensitivity was the only variable that increased as children grew older. This is in line with Flanagan and Stout's (2010) finding that social trust declines during adolescence.

# **How Does Victim Sensitivity Perpetuate Itself Across Social Situations?**

After having discussed the "ontogenetic" development and stabilization of victim sensitivity across the life course, we will now turn to our second question: how do specific instances of victimization contribute to a stabilization of victim sensitivity across situations? This question addresses the "actual-genetic" stabilization of victim sensitivity. We will argue that this stabilization can be reasonably well explained by associative learning and avoidance learning processes. As outlined above, victim-sensitive individuals are not only characterized by a high need to trust but also by a stabilized and generalized negative expectation concerning others' trustworthiness—probably due to experiences of victimization. These experiences are relevant for associative learning processes. According to the SeMI model, victim-sensitive individuals are particularly sensitive toward "cues" in their social environments that are associated with untrustworthiness (Gollwitzer et al., 2013). Being confronted with these cues evokes a "suspicious mindset" and makes preventive reactions, such as pre-emptive selfishness, more likely. Associative learning can explain why and how a sensitivity to "untrustworthiness cues" generalizes and, thus, stabilizes across situations.

# **Associative Learning, Untrustworthiness Cues, and Trusting Behavior**

Associative learning refers to the process by which associations between stimuli (including behavior) are learned. Associative learning encompasses classical, operant, and evaluative conditioning. In *classical conditioning* (or Pavlovian conditioning), a neutral stimulus is paired with an unconditioned stimulus (i.e., a reflex-evoking stimulus) until the neutral stimulus acquires the unconditioned stimulus' capability to evoke the reflex; thus, a *stimulus-outcome association* is learned. In *operant conditioning*, a behavior is paired with a pleasant (reinforcement) or unpleasant (punishment) stimulus/action until the frequency of the behavior is changed; thus, a *behavior-outcome association* is learned. In *evaluative conditioning*, a neutral stimulus is paired with an affective stimulus until the neutral stimulus acquires the valence of the affective stimulus; thus, a *stimulus-stimulus association* is learned. In the following, we discuss three processes that are relevant for the generalization (and, thus, the stabilization) of victim sensitivity across situations.

# Conditioned Stimuli

A first relevant assumption is that previously unconditioned stimuli that are associated with victimization become "conditioned." After this association is learned, such stimuli function as "untrustworthiness cues" that activate a suspicious mindset among victim-sensitive individuals (Gollwitzer and Rothmund, 2009). Importantly, whereas some untrustworthiness cues are rather idiosyncratic (e.g., the first name of a perpetrator, the location at which a victimization took place), others are more universal: expressions of anger (e.g., angry facial expressions, aggression-related behavioral patterns, hostile verbal remarks) are arguably less idiosyncratic untrustworthiness cues as they are associated with perceived aggression and victimization in general. In one of our recent studies (Gollwitzer et al., 2012), participants rated the trustworthiness of targets whose emotional facial expression varied from happy to angry. Results showed that victim-sensitive persons were more distrustful of angry (and neutral) but not of happy faces. We suggest that associative learning can explain how certain stimuli, such as angry facial expressions or even behavioral cues (such as a colleague asking for a favor), become "untrustworthiness cues" for victim-sensitive individuals. However, it is important to keep in mind that victim sensitivity is assumed to entail a heightened responsiveness to *any* information that indicates untrustworthiness, irrespective of how that information was acquired. Thus, instances of *observed* non-cooperation (Gollwitzer et al., 2009) or the activation of culturally shared stereotypes (e.g., untrustworthy car salesman) might suffice to trigger a suspicious mindset in relevant contexts.

## Exploitation as Punishment

Two other processes that are relevant for explaining the stabilization of victim sensitivity are operant conditioning (via punishment) and avoidance learning. Punishment occurs when cooperative behavior (e.g., telling a friend a secret, agreeing to switch shifts with a colleague) is followed by victimization (e.g., being betrayed, learning that one's helpfulness was exploited). In line with operant conditioning, one might say that one's willingness to trust others was "punished" and therefore becomes less likely to occur. Furthermore, behavior that reduces the likelihood of victimization becomes more frequent (via avoidance learning; see Avoidance Learning and the Stabilization of Victim Sensitivity).

# Implicit Cognition

Finally, direct (and observed) experiences of victimization may not only influence people's *explicit* evaluations of others' trustworthiness (via propositional processes), but are also likely to affect people's *implicit* evaluations of others (via associative processes). More precisely, victim-sensitive individuals might implicitly associate other people with untrustworthiness. According to the affective-propositional evaluation (APE) model (Gawronski and Bodenhausen, 2006), such negative implicit evaluations of others are particularly likely to drive explicit evaluations and behaviors in situations in which no inconsistent propositional information is considered (e.g., failing to realize that a different colleague requesting a favor has demonstrated her trustworthiness in the past), or in situations in which self-regulation resources are low (e.g., after having suppressed one's bad mood at work for a while, see Hofmann et al., 2007). By default, victim-sensitive individuals' evaluation of a new interaction partner can thus be understood as an affirmation of more general negative implicit expectations of others (Gilbert, 1991) unless contradictory trustworthiness cues are present. Taking victim-sensitive individuals' implicit evaluations of others' trustworthiness into account might be particularly important when it comes to changing their expectations of others' (un)trustworthiness. Whereas the APE model describes a number of ways in which implicit associations can be influenced, research on evaluative conditioning suggests that affective reactions are highly resistant to extinction (De Houwer et al., 2001) and, thus, more difficult to alter than individuals' explicit beliefs.

To sum up, we assume that associative learning plays a key role in the explanation of (a) victim-sensitive individuals' heightened responsiveness toward certain untrustworthiness cues, (b) victim-sensitive individuals' reduced trusting behavior, and (c) their implicit evaluations of other people's trustworthiness (and accompanied affective reactions). Importantly, whereas probably all people have been victimized in their lives to some extent, we assume that victim-sensitive individuals not only have *more* of these aversive experiences, but also that they experience them more *intensely* due to their strong need to trust. More concretely, a high need to trust is likely associated with more attention and stronger negative emotions elicited by experiences of victimization (cf., Gollwitzer and Rothmund, 2011), thereby rendering these experiences psychologically more meaningful. Thus, a high need to trust exacerbates associative learning in victimization experiences because it increases the intensity of the unconditioned stimulus (Passey, 1948; Pearce and Hall, 1980)—especially if this stimulus occurs unpredictably and uncontrollably (see our discussion of critical life events in Section "When and How Does Victim Sensitivity Begin to Emerge and Stabilize?"; cf. Filipp and Aymanns, 2010).

# **Avoidance Learning and the Stabilization of Victim Sensitivity**

Avoidance learning is a basic learning principle that refers to a process of behavior modification by which an animal or human reduces exposure to an aversive stimulus through an avoidance response. In early studies on avoidance learning (e.g., Mowrer and Miller, 1942), animals learned that an aversive stimulus (e.g., electric shock) was preceded by a warning signal (e.g., a tone). The aversive stimulus could, however, be postponed with a certain response (e.g., change of location). Avoidant behavior demonstrated in these studies could not be explained by a purely behavioristic stimulus-response pattern because the avoidant response occurred without direct reinforcement (it was in fact driven by the *non-occurrence* of an aversive stimulus). Consequently, avoidance learning was explained as a combination of two factors: classical and operant conditioning (Mowrer, 1947). First, due to its pairing with the unconditioned stimulus (e.g., the electric shock), the former neutral stimulus (e.g., the tone) becomes a conditioned stimulus (i.e., classical conditioning). Importantly, the conditioned stimulus is assumed to elicit fear when it occurs. Second, when the organism then happens to perform the avoidance response in the presence of the conditioned stimulus and thus prevents the occurrence of the unconditioned stimulus, the fear elicited by the conditioned stimulus is reduced. This, in turn, reinforces the avoidance response (i.e., operant conditioning). Thus, avoidance learning is assumed to be driven and maintained by feelings of fear. However, cognitive aspects such as expectations are likely to be involved in human avoidance learning as well (see Rescorla and Wagner, 1972; Lovibond, 2006; Declercq et al., 2008).

In clinical psychology, avoidance learning is considered a crucial factor for the maintenance of anxiety disorders; it refers to the process by which individuals reduce their exposure to a phobic stimulus through avoidant behavior (Bouton et al., 2001; Mineka and Zinbarg, 2006). Trying to avoid an aversive stimulus deprives the individual of positive learning experiences in which the conditioned stimulus might *not* be followed by the aversive stimulus. Thus, avoidant behavior is strongly self-reinforcing.

Associative learning and avoidance learning are likely to play a central role for the stabilization of victim sensitivity across situations. As described earlier, operant conditioning can explain how behavior related to victimization—such as cooperation and trusting others (e.g., doing a colleague a favor)—becomes less frequent when followed by victimization (i.e., "punishment"). Consistent with this notion, highly victim-sensitive individuals have been found to withdraw their cooperation in a trust game after experiencing victimization in an entirely different context, such as a virtual world (e.g., Rothmund et al., 2011). Furthermore, classical conditioning can explain how stimuli that indicate victimization (such as angry facial expressions) are learned and become untrustworthiness cues. Drawing on avoidance learning, we assume that due to their connection with victimization, untrustworthiness cues may elicit fear as a conditioned response.

# **Uncooperative Behavior as a Defense Reaction**

Confrontations with "untrustworthiness cues" signal a threat to a particular need, the need to trust (see The Need to Trust). As described in the previous section, one way to cope with this threat would be to avoid the threat. Victim-sensitive individuals should tend to avoid situations in which they might fall prey to the egoistic intentions of others and instead prefer situations in which exploitation is unlikely. For instance, victim-sensitive individuals can be expected to prefer individual (i.e., independent) over cooperative (i.e., interdependent) work situations and situations in which free-riding is rigorously punished over situations in which free-riding is unlikely to be detected. Of course, these situational preferences also have an impact on the quantity and quality of their friendships and, especially, the extent to which a close relationship remains stable and satisfactory for both partners (cf. Gerlach et al., 2012).

However, research shows that victim sensitivity is not exclusively related to avoidance-oriented behaviors; victim-sensitive individuals show typical "approach-oriented" behaviors as well: whenever untrustworthiness cues are present, victim-sensitive individuals tend to behave uncooperatively in social dilemma situations (Fetchenhauer and Huang, 2004; Gollwitzer et al., 2009; Rothmund et al., 2011), even at the cost of their own benefit. Notably, victim-sensitive persons are not more egoistic *per se*; rather, they tend to be more hostile when faced with injustice. For instance, when given the opportunity to punish a defector or to compensate a victim in a third-party intervention game, victim-sensitive individuals prefer punishing the offender over compensating the victim, even if punishment is costly for them (Lotz et al., 2011).

The General Process Model of Threat and Defense (Jonas et al., 2014) provides a helpful and informative theoretical framework for explaining why and when avoidance-oriented behaviors turn into approach-oriented ones. This model posits that being confronted with threat (of any kind) first activates the behavioral inhibition system (including anxious arousal and attentional vigilance toward fear-eliciting cues) and facilitates avoidanceoriented defense reactions. Since a state of avoidance is perceived as inherently unpleasant, avoidance-oriented behaviors eventually turn into approach-oriented behaviors. These approach-oriented behaviors can be more or less concrete (e.g., seeking stimulation or social affiliation; attacking the source of the threat) vs. abstract (e.g., increased adherence to personal and moral values; endorsing punitive systems).

Regarding victim sensitivity, it is reasonable to assume that, when confronted with untrustworthiness cues, victim-sensitive individuals initially show avoidance-oriented reactions such as an increased attentional vigilance toward untrustworthiness. Prior research has shown that, even in the absence of an untrustworthiness prime, victim-sensitive persons show a greater attentional vigilance toward justice- and injustice-related semantic concepts (Baumert et al., 2012), and more recent research shows that, in the presence of an untrustworthiness prime (i.e., a victimization experience), victim-sensitive individuals are more likely to associate ambiguous social situations with injustice (Maltese et al., 2014). Especially the latter finding is in line with the notion that victim-sensitive individuals show avoidance-oriented reactions after being confronted with untrustworthiness cues. Avoidance, however, may eventually transform into approach, such as hostility, uncooperativeness, and recklessness. In other words, avoidance- and approach-related behaviors can be positively related to each other.

According to the General Process Model of Threat and Defense (Jonas et al., 2014), hostile, uncooperative, aggressive, and cynical behavioral reactions toward experienced or anticipated victimization can be regarded approach-related reactions that aim to defend or satisfy a certain need: for victim-sensitive persons, it is the need to trust. Such distal defense reactions tend to reinforce themselves, as we have discussed before. Uncooperativeness and selfishness as "pre-emptive" reactions to anticipated victimization therefore stabilize over time. Notably, such selfishness may backfire: Other people may take the "pre-emptive" selfishness displayed by victim-sensitive individuals as a cue for the fact that these individuals cannot be trusted, and behave uncooperatively in return. This, in turn, confirms what victim-sensitive individuals had expected. The pre-emptive selfishness that victim-sensitive persons are likely to display in social interdependence situations and the fear of exploitation that triggered this hostility both create a self-reinforcing system; a self-fulfilling prophecy.

Taken together, experiences of victimization increase avoidance-related (e.g., attentional vigilance toward untrustworthiness cues) *and* approach-related behaviors (e.g., pre-emptive selfishness). Whereas direct experiences of victimization are the starting ground for these processes to unfold, the nature of the behavioral reactions toward them contributes to the stabilization of victim sensitivity across situations. Because avoiding social exchange and social dilemma situations deprives individuals of contrary learning opportunities (e.g., changing shifts with a colleague who does you a favor in return) and because pre-emptive selfishness as an approach-oriented response will generate non-cooperation in response (e.g., loafing in a joint task), these behaviors eventually

reinforce negative expectations concerning others' trustworthiness.

#### **Conclusion**

In Section "How Does Victim Sensitivity Perpetuate Itself Across Social Situations?" of this article, we focused on the role of general learning mechanisms for the formation and stabilization of victim-sensitive individuals' biased responses to untrustworthiness cues as well as their non-cooperative behavior. The model that results from these arguments is displayed in **Figure 2**. Associative learning can explain how victimization experiences result in (a) a generalization of untrustworthiness cues (via associative learning), (b) decreasing levels of trusting behavior (via operant conditioning due to punishment), and (c) the stabilization of negative implicit trustworthiness expectations. In addition, avoidance learning and self-fulfilling prophecies create a selfreinforcing cycle which stabilizes generalized untrustworthiness expectations as well as low trusting behavior both via avoidant and pre-emptively selfish or hostile behavior.

Notably, some of the processes we discussed with regard to the "actual-genetic" stabilization of victim sensitivity in the present Section can be meaningfully related to the four personenvironment transactions that we discussed with regard to the "ontogenetic" stabilization of victim sensitivity in Section "When and How Does Victim Sensitivity Begin to Emerge and Stabilize?". For instance, by selectively seeking social environments that reinforce their untrustworthiness expectations ("selective transactions" according to Caspi and Roberts, 1999, 2001), victimsensitive individuals never challenge these expectations—which resembles an instance of avoidance learning. And self-fulfilling prophecies, as we defined them here, resembles what Caspi and Roberts (1999, 2001) referred to as evocative transactions: victimsensitive individuals behave in ways that indirectly validate their beliefs that others are untrustworthy.

# **Summary and Outlook**

In this article, we developed a theoretical framework (or, rather, two theoretical frameworks) that aim at explaining how and why victim sensitivity emerges and stabilizes. Notably, victim sensitivity is not only a risk factor for antisocial behaviors in various kinds of social encounters (e.g., Gerlach et al., 2012; Gollwitzer et al., 2013), but also for a number of behavioral problems during adolescence, such as aggressiveness (Bondü and Krahé, 2014), anxious and angry rejection sensitivity and conduct problems (Bondü and Elsner, 2015) as well as symptoms related to attention deficit/hyperactivity disorders (Schäfer and Kraneburg, 2012; Bondü and Esser, 2015).

In Section "When and How Does Victim Sensitivity Begin to Emerge and Stabilize?", we borrowed concepts from developmental psychology, research on coping with critical life events, and life-span personality psychology to derive a model that explains the "ontogenetic" stabilization of victim sensitivity during the life span. Victimization experiences and social information processes that describe how a person copes with these experiences are assumed to play a major role for the stabilization of victim sensitivity—more precisely, for the tendency to expect other people to be untrustworthy. From this model, which is depicted in **Figure 1**, testable hypotheses can be derived.

*First*, we assume that victimization experiences during late childhood and early adolescence increase a person's victim sensitivity especially when these experiences are (a) self-relevant, (b) imply an obstruction of relevant personal goals, (c) are unpredictable, and (d) uncontrollable—in other words, when these experiences fulfill the criteria of "critical" life events. Examples for such events could be experiences of being bullied, cybermobbed, or socially excluded by significant peers. *Second*, we hypothesize that victim-sensitive individuals actively contribute to a stabilization of this trait by reacting consistently to potential victimization situations ("reactive transactions"). More precisely, we assume that victim sensitivity provides people with a set of cognitive schemas (e.g., attributional styles regarding other people's untrustworthiness) and behavioral scripts (e.g., behaving uncooperatively) that bias their information processing in specific situations—situations that are marked by social interdependence and uncertainty regarding other people's intentions and behaviors (i.e., social dilemma situations). A *third* hypothesis that can be deduced from our framework is that victim-sensitive individuals actively select environments (e.g., peers, friends, partners, etc.) that fit their own attitudes and worldviews ("selective transactions"). Such a fit between personality and the social environment reinforces victim sensitivity and stabilizes it over time. All of these hypotheses can be tested in carefully designed cohort—or, even more preferably, longitudinal—studies in which the variables that are assumed to predict the formation and stabilization of victim sensitivity are either measured or experimentally manipulated.We believe that late childhood to mid-adolescence is a critical phase for the formation and stabilization of victim sensitivity. Thus, cohort studies should at least compare age groups ranging between 9 and 15 years (Bondü and Elsner, 2015).

In Section "How Does Victim Sensitivity Perpetuate Itself Across Social Situations?", we borrowed concepts from research on associative learning and social cognition to explain why and how victim sensitivity perpetuates across social situations. Associative learning can explain how neutral stimuli can become "untrustworthiness cues" for victim-sensitive persons, and avoidance learning can explain why social expectations regarding the untrustworthiness of other people reinforce themselves. In addition, approach-oriented behavior such as "pre-emptive" hostility and selfishness, which may be regarded a distal defense to threats to the "need to trust," create a vicious cycle or a selffulfilling prophecy: the degree of pre-emptive hostility displayed by victim-sensitive individuals in the face of untrustworthiness cues may lead their interaction partners to infer that cooperation is futile, which, in turn, reinforces the expectations held by victim-sensitive individuals. Again, a number of predictions follow from the framework we developed in Section "How Does Victim Sensitivity Perpetuate Itself Across Social Situations?" (see also **Figure 2**).

*First*, untrustworthiness cues are "stronger" unconditioned stimuli for people high (than for people low) in victim sensitivity. This hypothesis could be tested in an evaluative conditioning study featuring untrustworthiness and trustworthiness cues as well as neutral stimuli. In such a design, participants' victim sensitivity should predict the change of liking toward neutral stimuli that were paired with untrustworthiness cues (but not with trustworthiness-related or neutral cues). *Second*, victim-sensitive individuals should harbor negative implicit evaluations of others' trustworthiness due to associative learning. Using a singletarget Implicit Association Test, it could be investigated whether victim-sensitive individuals associate "others" more readily with untrustworthiness relative to trustworthiness. More importantly, the influence of participants' implicit untrustworthiness expectations on behavior (i.e., cooperation) should be examined *visà-vis* their explicit untrustworthiness expectations (i.e., victim sensitivity) in different situations (e.g., under ego depletion; in the presence vs. absence of trustworthiness information). *Third*, drawing on avoidance learning as well as the General Process Model of Threat and Defense (Jonas et al., 2014), we assume

# **References**


that in potentially exploitative situations, victim-sensitive individuals will first show avoidance-related reactions (e.g., a higher attentional vigilance to untrustworthiness cues), which eventually transform into approach-related reactions (e.g., "pre-emptive" selfishness). *Fourth*, victim-sensitive individuals contribute to the confirmation of their expectations and create cycles of noncooperation through their own behavior (self-fulfilling prophecy or "evocative transactions"). This hypothesis could be tested in a repeated public goods game in which players have to decide how much to contribute to a common good (and do so iteratively for a number of rounds). In such a paradigm, we would expect, for instance, that one highly victim-sensitive individual eventually reduces the other players' willingness to contribute, which confirms this individual's *a priori* expectation: that other people are untrustworthy and harbor mean intentions.

To sum up, research on victim sensitivity, and on justice sensitivity in general, has gained momentum in various areas during recent years, and most of what we know about this trait so far is that it is a double-edged sword: in a way, it represents a true concern for justice and trust, but this concern leads to maladaptive behavioral decisions when social situations become uncertain. Thus, understanding how such a trait that is correlated with so many problematic behavioral tendencies emerges and stabilizes is therefore of vital importance, not only from a theoretical, but also an applied psychological perspective.

# **Acknowledgments**

We would like to thank Anna Baumert for helpful comments on an earlier version of this article. Further, we would like to thank the guest editors of this Research Topic, Eva Jonas and Christina Steindl, as well as two reviewers, for providing us with very helpful, detailed and constructive feedback.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2015 Gollwitzer, Süssenbach and Hannuschke. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Facing danger: how do people behave in times of need?The case of adult attachment styles

#### *Tsachi Ein-Dor\**

Interdisciplinary Center Herzliya, School of Psychology, Herzliya, Israel

#### *Edited by:*

Christina Steindl, University of Salzburg, Austria

#### *Reviewed by:*

Johannes Klackl, University of Salzburg, Austria Ina Grau, University of Bonn, Germany

#### *\*Correspondence:*

Tsachi Ein-Dor, Interdisciplinary Center Herzliya, School of Psychology, Herzliya 46150, Israel e-mail: teindor@idc.ac.il

Bowlby's (1982) attachment theory has generated an enormous body of research and conceptual elaborations. Although attachment theory and research propose that attachment security provides a person with many adaptive advantages, during all phases of the life cycle, numerous studies indicate that almost half of the human species can be classified as insecurely attached or insecure with respect to attachment. It seems odd that evolution left humans in this vulnerable position, unless there are some advantages to individuals or groups, under at least some conditions, of anxious and avoidant attachment styles. I argue that a social group containing members with different attachment patterns may be more conducive to survival than a homogeneous group of securely attached individuals because each attachment disposition has specific adaptive advantages that promote the survival of the individual and people around him or her when facing threats and perils. In making this argument, I extend the scope of attachment theory and research by considering a broader range of adaptive functions of insecure attachment strategies, and present data to support my argument.

**Keywords: attachment, anxiety, avoidance, defensive behavior, social defense theory**

## **BACKGROUND**

As illuminated in cave painting, humans have faced threats and danger throughout the eons (Valladas et al., 2001). To survive, animals commonly employ fight-or-flight responses in times of need (Cannon, 1929; Jansen et al., 1995). Humans, however, are wanting in their physiological ability to effectively fight threats, and fall short in their ability to flight by climbing or running afoot. For example, humans are almost bare of body hair (Bergman, 2004) and its protective attributes against cuts, bruises, and bites (e.g., Blanchard, 2009), which make them vulnerable when fighting threats. Humans were probably perfected by evolution to solve the worriment of survival by cooperating with others and utilizing the strength of numbers (Axelrod and Hamilton, 1981; Cosmides, 1989; Brewer and Caporael, 1990; Axelrod, 2006). To date, however, research on human defensive reactions to threats has focused mainly on individual-level responses such as people's attentional bias toward signals of threats (e.g., Brandtstädter et al., 2004), people's responses to threat scenarios (Perkins and Corr, 2006) and people's actions and reactions in life-endangering events (see Mawson, 2012 for a comprehensive review). In the present paper, I present social defense theory (SDT; Ein-Dor et al., 2010), in an attempt to bridge the gap in the literature on human defensive behavior by suggesting one possible group-level process by which people promote the likelihood of surviving perilous events.

Social defense theory suggests that we ought to acknowledge the effects of other people around us on our responses to threats and on our ability to prevail dicey challenges. Specifically, SDT proposes that some people are more perceptive of threat-related cues and tend to detect threats quicker and more accurately than others. Other people are compulsively self-reliant and upon detection of threats tend to employ self-protective actions more rapidly and

effectively than others. Still other people are better at massing collective efforts and leading group actions because they are more relationship-oriented than others. Because each of these responses promotes survival in a unique way, I contend that groups comprising these three styles of people (i.e., more heterogeneous groups with respect to people's personality and related action tendencies) will be more effective when dealing with threats than more homogeneous groups because they combine the abilities for early detection of threats, rapid responses, and effective cooperation. According to SDT (Ein-Dor et al., 2010), these three personality dispositions are the manifestation of people's attachment orientations (see Mikulincer and Shaver, 2007 for an extensive review).

# **ATTACHMENT THEORY**

According to attachment theory (Bowlby, 1973, 1980, 1982), humans possess a mammalian innate psycho-biological system – the attachment system – that was perfected by evolution to promote the survival of infants, although it is remains active "from the cradle to the grave" (Bowlby, 1982, p. 208). It motivates the individual to seek proximity to significant others (attachment figures) when he or she feels a need for protection and care. When attachment figures respond sympathetically to a person's needs over a long series of interactions, they fosters a sense of attachment security, which in time formulates into a trait-like disposition of security about the self, others and the world. In adulthood, secure people respond to threats either by relying on internal resources to regulate stress and to maintain high self-esteem and self-efficacy or by seeking concrete support from others or collaborating with them to regain safety and to restore a sense of security (Shaver and Mikulincer, 2002).

Conversely, when attachment figures often fend off bids for support or respond unreliably to a person's needs, they unintentionally foment one out of two chronic states of insecurity – avoidance, marked by abysmal independence, lack of trust in others, and maintaining a defensive pretense of security while employing cognitive and emotional avoidance especially in times of need; and anxiety, marked by symbiotic dependence and immutable sense of strain while maintaining constant hypervigilance to threats and intensified negative affectivity (see Mikulincer and Shaver, 2007, for a review). Social and personality psychologists generally conceptualize adult attachment dispositions as regions in a continuous two-dimensional space and not as typologies (e.g., Brennan et al., 1998). One dimension relates to avoidance and the second to attachment-related anxiety. Attachment security is defined by low scores on both anxiety and avoidance.

The dominant perspective regarding attachment security is that secure people enjoy adaptive advantages in all fabric of life compared with people high on anxiety and/or avoidance (Mikulincer and Shaver, 2007). For example, secure people endorse fewer psychopathologies (Ein-Dor and Doron, in press), adopt more constructing coping strategies with relationship-related conflicts and stress (Mikulincer et al., 2002a), and tend to be viewed by potential partners as more attractive (e.g., Klohnen and Luo, 2003). These advantages of security and related benefits lead researchers to ponder why approximately half of the human population of earth are insecure with respect to attachment. Belsky et al. (2010) and Del Giudice and Belsky (2011) were the first to suggest that attachment anxiety and avoidance have adaptive benefits such as earlier menarche in females (also see Chisholm et al., 2005) that allows earlier reproduction in stressful and risky environments in which waiting for optimal conditions might result in failing to reproduce.

Research has indicated, however, that the probable selection pressure that caused the emergence of the attachment behavioral system in mammalian evolution was survival-related and not early reproduction (Bowlby, 1973, 1982; Mikulincer and Shaver, 2007; Ein-Dor et al., 2010; Ein-Dor, 2013). Various physical and psychological threats such as loud noises, darkness, fatigue, and illness activate the attachment system (Bowlby, 1973), and the behavioral and cognitive outcomes of the system such is proximity seeking increase the likelihood of protection and survival (Mikulincer et al., 2000, 2002b). In keeping with these findings, Ein-Dor et al. (2010) and his colleagues proposed SDT, which suggests that each of the three major attachment dispositions – security, anxiety, and avoidance – awards unique adaptive advantages for the individual and for people around her or him that increase the likelihood of surviving perilous events.

#### **SOCIAL DEFENSE THEORY**

A decade ago,Nettle's(2006, p. 625) has contended that personality variations can be understood in terms of tradeoffs among fitness costs and benefits: "Behavioral alternatives can be considered as tradeoffs, with a particular trait producing not unalloyed advantage but a mixture of costs and benefits such that the optimal value for fitness may depend on very specific local circumstances". SDT

(Ein-Dor et al., 2010) adopts Nettle's perspective and proposes that security and insecurity dispositions alike endow adaptive advantages that increase the likelihood of survival while also incurring distinct disadvantages that might hinder survival unless they are complemented by contributions from people with different attachment dispositions.

#### **ADVANTAGES AND DISADVANTAGES OF SECURE INDIVIDUALS' DEFENSIVE REACTIONS**

Attachment research has indicated that secure people tend to lead team efforts and promote the success of their social group by collaborating with others in times of need. For example, secure individuals endorse greater prosocial and task-oriented leadership motivations and lower self-enhancing and self-reliance motivations than their more insecure counterparts (Davidovitz et al., 2007; Hinojosa et al., 2014). Secure leaders are also apprised by their followers as demonstrating higher efficacy in emotionfocused situations and task-focused ones (Davidovitz et al., 2007). As teammates they work more effectively with other group members when solving problems and facing challenges (e.g., Rom and Mikulincer, 2003; Molero et al., 2013). These advantages are believed to be the manifestation of a sense of security that was developed in past supportive experiences with attachment figures (Mikulincer and Shaver, 2007) and which comprises core beliefs regarding the safeness of the world and people in it. These optimistic, solacing mental representations promote self-palliative reappraisals of threats, which help secure people to outperform insecure individuals in many daily and challenging situations alike (Mikulincer and Shaver, 2007).

A sense of security is not always linked with actual physical security, however. In times of danger, it can be maladaptive if it clogs the recognition of threats and slows down rapid, effective responses. For example, Mawson (2012, p. 233) has indicated that "while mass panic (and/or violence) and self-preservation are often assumed to be the natural response to physical danger and perceived entrapment... the typical response to a variety of threats and disasters is not to flee but to seek the proximity of familiar persons and places". Therefore, secure individuals tend to seek proximity to others (Waters and Waters, 2006; Mikulincer et al., 2009), even if this is not always the safest strategy. Adopting a schema of security about the self, others, and the world may, therefore, incur two prominent disadvantages: (a) delayed perception of danger and (b) slower employment of effective defensive behaviors in response to threats and danger.

Sime (1983, 1985), in keeping with Mawson's (1978) suggestions, examined these disadvantages in a police-reports-based study of reactions to a fire in a large coastal resort on the Isle of Man, Great Britain, 1973. Sime showed that people who reported being close to family members were less likely to react to early signs of danger such as noises and shouts. Rather, they tended to react only after witnessing unambiguous cues of danger, such as people running while holding fire extinguishers, smoke, and flames – which usually means a loss of precious time in successfully escaping the situation. Other studies of survivors' behavior during perilous events have also indicated that people who reported being close to family members perceived that they were in danger slower than people who were alone in the situation (Aguirre et al., 1998; Köster et al., 2011). According to SDT, the tendency to mainly react to clear signs of threats and not to take heed to earlier cues of danger characterizes secure people (Ein-Dor et al., 2010).

Regarding slower employment of effective defensive behaviors in times of need, research has indicated that security with respect to attachment may be linked with non-optimal reactions in times of danger. For example, Mawson (2012, p. 153) noted that in combat situations, "what may be important for the individual soldier is maintaining proximity with his fellows, even though this may involve moving into situations of greater physical danger". Studies on natural disasters have also indicated that"people tended to turn to and protect loved ones rather than flee from the threat" (Form and Nosow, 1958, p. 26) and that "traditional family ties often keep individual members in the danger zone until it is too late" (Hill and Hansen, 1962, p. 217).

In keeping with SDT, research has indicated that the advantages of secure people come into play in their better leadership qualities and their ability to coordinate group activities. Nevertheless, these advantages are partially countervail by their sluggish perceptiveness of actual threats and their somewhat imperfect reactions to danger because of their will to stay close to people around them. This suggests that secure people's inclination to focus on ongoing tasks and chores irrespective of mounting danger may impede their survival. Attentiveness to early signs of danger and hasty fight-or-flight responses may be necessary to evade disaster. SDT suggests that being high on either attachment avoidance or anxiety might confer such adaptive abilities and counterpoise the disadvantages of attachment security when facing perilous events.

#### **ADVANTAGES AND DISADVANTAGES OF PEOPLE HIGH ON ATTACHMENT ANXIETY**

People high on attachment anxiety often appraise their own functioning in groups as imperfect and are judged by others as falling short in their ability to effectively lead team efforts in completing various tasks (Davidovitz et al., 2007). They take work less seriously than their secure counterparts and make fewer contributions to a team and of poorer quality (Rom and Mikulincer, 2003). Despite these shortcomings, the hypervigilant strategies that anxious people adopt when dealing with threats might nevertheless promote their survival and benefit others in their social surrounding: They sensitively monitor the environment for threats and upon detection of danger they seek support by actively calling on others for help and by overacting their emotions (Cassidy and Kobak, 1988; Feeney and Noller, 1990). Ein-Dor et al. (2010) named these behavioral tendencies sentinel behavior.

According to SDT, the sentinel behavior is stemmingfrom a selfschema that guides anxious people's responses in times of need. It comprises default action tendencies that cause people high on attachment anxiety "(a) to remain vigilant with respect to possible threats, especially in unfamiliar or ambiguous situations; (b) to react quickly and strongly to early, perhaps unclear cues of danger (e.g., unusual noises, shuffling feet, shouts); (c) to alert others about the imminent danger; (d) if others are not immediately supportive, to heighten efforts to get them to provide support; and (e) to minimize distance from others when coping with a threat" (Ein-Dor et al., 2011a, p. 2).

The benefits of sentinel behavior is apparent in many species of animals. For example, African elephants (Soltis et al., 2014) and chimpanzees (e.g., Schel et al., 2013), among other species of animals such as birds (Evans et al., 1993) and fish (Smith, 1992), produce shrill alarm calls when they detect a potential threat as predators. Humans may also benefit from the hyperactivating strategies of people high on anxiety in similar ways.

The first evidence in favor of this notion linked attachment anxiety with heightened accessibility to core components of the sentinel schema – noticing danger quicker than others and warning them about the danger (Ein-Dor et al., 2011a). For example, when participants were asked to write a story about a TAT-like (Thematic Apperception Test; Murray, 1943) card with a scary scenario, those higher on attachment anxiety composed stories with more sentinel-related narratives. After reading a story about a person who behave in a sentinel way, participants who scored higher on attachment anxiety were more likely to generate more inferences about this person and his personality.

Attachment anxiety was later linked with actual sentinel-related behavior in times of need (Ein-Dor et al., 2011b). Specifically, the behavior of small groups of three people were observed in an experimentally manipulated threatening situation: a room progressively filling up with non-toxic smoke from what seems like a malfunctioning computer. In line with SDT, groups higher on attachment anxiety detected the presence of smoke quicker than less anxious groups. Specifically, 1-point increase in anxiety was linked with 11.5 s decrease in detection time. In addition, the person with the highest score on anxiety detected the presence of smoke in the room more often than predicted by chance alone (Ein-Dor et al., 2011a). In a complementary self-report-based research, participants were asked to report on the first action that they are likely to take on various threat scenarios (Ein-Dor and Perry-Paldi, 2014). Results indicated that attachment anxiety qualified the effects of situational features (e.g., degree of dangerousness and clarity of the threat) to increase the likelihood of sentinel (e.g., yelling) and fear-related behaviors (e.g., running away).

Aside from establishing a link between attachment anxiety and reaction to potential life-engendering threats, people high on attachment anxiety were also found to have a tendency to deliver a warning message without delay (Ein-Dor and Orgad, 2012). Using a designated software, participants were led to believe that they accidently activated a Trojan horse that completely erased the experimenter's hard drive and possibly the campus's server. Participants were then asked to alert the computer technicians about the hazard. On their way, the researchers created four behavioral settings in which they tried to delay the participants from delivering the warning message (e.g., a confederate who asked them to help her completing a short questionnaire). In line with SDT, results indicated that high attachment anxiety was linked with fewer delays.

Research has also shown that attachment anxiety is associated with the ability to accurately detect social-based threats. For example, people high on attachment anxiety are better apt in foretelling their partners' true thoughts and feelings in situations that pose a threat to the relationship such as when partners rate an attractive opposite-sex person (Simpson et al., 1999, 2011). People high on attachment anxiety were also better at detecting cues of interpersonal deceit (Ein-Dor and Perry, 2014). Specifically, participants watched a series of seven video clips in which people retold the events they experienced the day before. In some of the clips the speaker was honest and in some – dishonest. Participants were asked to appraise whether the person in the clip lied or told the truth. In an additional study, semi-professional poker players completed a self-report questionnaire measuring attachment dispositions and then they participated in a poker tournament that was held outside campus. Results indicated that people higher on attachment anxiety were more accurate in detecting deceitful statements, and that players higher on anxiety won greater amount of money during a game of poker, which relates to an ability to call opponents' bluffs. Taken together, research has supported SDT's premise regarding the possibility that people high on attachment anxiety adopt sentinel-related cognitions and behaviors that promote survival.

#### **ADVANTAGES AND DISADVANTAGES OF PEOPLE HIGH ON ATTACHMENT-RELATED AVOIDANCE**

People high on attachment avoidance relegate appraisals of threats and downgrade sensations of pain and vulnerability (e.g., Fraley and Shaver, 1997). Therefore, they are usually less vigilant to signs of danger and tend to recognize the extent of threat later than others (Ein-Dor et al., 2010). They tend to appraise team cohesion as more fractured than others and are often appraised by others as less apt to lead because of withered emotional abilities (Davidovitz et al., 2007). They do not tend to collaborate with others and, hence, they do not perform well as teammates (Rom and Mikulincer, 2003). In times of need, they are compulsively selfreliant (Bowlby, 1973) and tend to take self-protective actions that promote their own interests (Feeney and Collins, 2001), a reaction tendency that Ein-Dor et al. (2010) named rapid fight-or-flight behavior. As a result, while anxious and secure individuals focus their attention on the whereabouts of significant others around them, without focusing quickly enough on how to evade the progressive threat, avoidant people are able to discover a way to effectively deal with the threat.

The asocial tendencies of people high on avoidance might actually help people around them eluding danger. Suppose that an avoidant person is in a shopping mall engrossed by flames. To save her or himself, he or she will take quick protective actions to espy the best route to escape or to quickly extinguish the fire. These behaviors increase the avoidant person's survival chances but might also save other people's lives. For example, the sight of people running from danger can motivate the escape of others around them and unintentionally save lives (e.g., Mawson, 2012). As Marshall (1947, pp. 145–146) noted regarding the military behavior duringWorldWar II:"It can be laid down as a general rule that nothing is more likely to collapse a line of infantry than the sight of a few of its number in full and unexplained flight to the rear... One or two or more men made a sudden run to the rear which others in the vicinity did not understand... In every case the testimony of all witnesses clearly [indicated] that those who started the run... had a legitimate or at least a reasonable excuse

for the action". Aside from promoting the motivation for escape, people who flee before others do must clear an escape route of possible obstacles and, thus, others can enjoy an open route to follow. Taken together, people high on avoidance may increase their own and their group members' chances of survival in times of need.

According to SDT, the asocial behavior of avoidant individuals stems from a rapid fight-or-flight schema that comprises the following action tendencies: "(a) minimize the importance of threatening stimuli; (b) when danger is clearly imminent, take quick self-protective action, either by escaping the situation or by taking action against the danger; and (c) at such times, do not worry about coordinating one's efforts with those of other people" (Ein-Dor et al., 2011a, p. 3).

The first evidence in favor of this notion linked attachmentrelated avoidance with the following core narratives of the rapid fight-or-flight schema when writing a story about a scary scenario: (a) escaping a perilous event without helping others, (b) acting without collaborating or deliberating with others, and (c) reacting quickly. After reading a story about a person who behave in a rapid fight-or-flight way, participants high on attachment avoidance generated more inferences about the person's behaviors and thoughts than people low on avoidance.

Attachment avoidance was later linked with actual rapid fightor-flight behavior in times of need (Ein-Dor et al., 2011b). Specifically, research has indicated that the typical response to a room progressively filling up with smoke was fleeing to the adjunct corridor. In line with SDT, groups higher on attachment avoidance was quicker to escape the room than more secure groups, and were appraised by judges as more effective in dealing with the situation. In a complementary self-report-based research, attachment avoidance was found to qualify the effects of situational features (e.g., degree of dangerousness and clarity of the threat) to increase the likelihood of rapid-responder (e.g., attacking; which relates to fight responses), fear-related (e.g., running away; which relates to flight reactions), and anxiety-related (e.g., risk assessment) reactions (Ein-Dor and Perry-Paldi, 2014). Taken together, research has supported SDT's notion that attachmentrelated avoidance is associated with rapid fight-or-flight cognitions and behaviors.

#### **GROUP COMPOSITION AND ITS ASSOCIATION WITH EFFECTIVENESS WHEN DEALING WITH THREAT**

In the course of evolution, humans lived in relatively small groups or tribes of kin and often faced threats and perils. As individuals, we are lacking in our ability to survive: we have a fragile body, which hinders our ability to effectively fight threats, and we were evolved to walk on two legs, which limits our ability to effectively escape threats. We survived by utilizing the strength of numbers and by facing perils as a group. SDT contends that to survive we needed several abilities that one person can never hope to have: heightened vigilance to threats and danger, quick responses to threats once they are detected, and calm and calculated collective efforts to overcome the threats. An effective response to threats could only be achieved by the combining efforts of people with different attachment dispositions. According to SDT, each of the three major styles of attachment dispositions – security, anxiety, and avoidance – have both unique adaptive advantages that promote survival and disadvantages that might hinder survival. Heterogeneous groups with respect to attachment dispositions should be more sensitive to early signs of threat by utilizing the sentinel abilities of anxious members; act quickly in response to threats without much deliberation by utilizing the rapid fight-or-flight abilities of avoidant members; and manage complex group-level tasks by utilizing the leadership and social-oriented abilities of secure members. Accordingly, a group comprising all three styles of attachment patterns may benefit from the combined abilities of each disposition and offset their shortcomings. Therefore, such groups might be superior to other groups in dealing with threats and survival problems.

In support of this proposition, heterogeneous groups with respect to attachment patterns were appraised by external judges as dealing more effectively with a room gradually filling up with smoke than more homogenous groups (Ein-Dor et al., 2011b). Heterogeneity in attachment patterns was also found to promote the success of work teams. Specifically, teams' heterogeneity in attachment anxiety and avoidance scores was related to better academic grades (Lavy et al., 2014). This latter finding was moderated by teams' cohesion, however. Heterogeneity was linked with better performance only among teams that were able to construct high sense of cohesion. In other words, heterogeneity could be a double-ended sword.

Individuals with either anxiety or avoidance dispositions could present a social challenge to groups' dynamics: People high on attachment anxiety because of their hyperactivation tendencies are clingy, needy, vexed, and fearful and are constantly seeking approval of others, sometimes by being intrusive (Smith et al., 1999). People high on attachment avoidance might neglect the needs of others and keep their distance of others, which may hinder effective communication within the group (Smith et al., 1999; Rom and Mikulincer, 2003). These tendencies may cause conflicts between team members and reduce teams' socio-emotional functioning (Pelled et al., 1999; although teams' objective performance might still be high). Nevertheless, when insecure team members are in a reassuring environment that accept them and let them feel safe and trusted, their challenging relationship-related perceptions and behaviors might be turned into advantages.

#### **CONCLUDING COMMENTS**

Social defense theory was devised to suggest one possible grouplevel process by which people promote the likelihood of surviving perilous events. It is based on the proposition that different attachment dispositions bring different abilities to a group – sentinel, rapid fight-or-flight, and leadership abilities – rendering the group superior to other groups in dealing with threats and survival problems. Pending on receiving additional empirical support, SDT may have important implications for theory and research concerning human defensive behaviors, group dynamics, threat detection, and adaptive benefits of personality diversity.

#### **ACKNOWLEDGMENT**

This research was supported by the ISRAEL SCIENCE FOUNDA-TION (grant No. 942414).

#### **REFERENCES**


Hill, R., and Hansen, D. A. (1962). Families in disaster. *Man Soc. Disaster* 185, 221.


Marshall, S. (1947). *Men Against Fire*. Norman, OK: University of Oklahoma Press.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 02 July 2014; accepted: 27 November 2014; published online: 10 December 2014.*

*Citation: Ein-Dor T (2014) Facing danger: how do people behave in times of need? The case of adult attachment styles. Front. Psychol. 5:1452. doi: 10.3389/fpsyg.2014.01452 This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology.*

*Copyright © 2014 Ein-Dor. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Inhibition Underlies the Effect of High Need for Closure on Cultural Closed-Mindedness under Mortality Salience

#### Dmitrij Agroskin, Eva Jonas, Johannes Klackl and Mike Prentice\*

*Department of Psychology, University of Salzburg, Salzburg, Austria*

The hypothesis that people respond to reminders of mortality with closed-minded, ethnocentric attitudes has received extensive empirical support, largely from research in the Terror Management Theory (TMT) tradition. However, the basic motivational and neural processes that underlie this effect remain largely hypothetical. According to recent neuropsychological theorizing, mortality salience (MS) effects on cultural closed-mindedness may be mediated by activity in the behavioral inhibition system (BIS), which leads to passive avoidance and decreased approach motivation. This should be especially true for people motivated to avoid unfamiliar and potentially threatening stimuli as reflected in a high need for closure (NFC). In two studies involving moderated mediation analyses, people high on trait NFC responded to MS with increased BIS activity (as indicated by EEG and the line bisection task), which is characteristic of inhibited approach motivation. BIS activity, in turn, predicted a reluctance to explore foreign cultures (Study 1) and generalized ethnocentric attitudes (Study 2). In a third study, inhibition was induced directly and caused an increase in ethnocentrism for people high on NFC. Moreover, the effect of the inhibition manipulation × NFC interaction on ethnocentrism was explained by increases in BIS-related affect (i.e., anxious inhibition) at high NFC. To our knowledge, this research is the first to establish an empirical link between very basic, neurally-instantiated inhibitory processes and rather complex, higher-order manifestations of intergroup negativity in response to MS. Our findings contribute to a fuller understanding of the cultural worldview defense phenomenon by illuminating the motivational underpinnings of cultural closed-mindedness in the wake of existential threat.

Keywords: mortality salience, behavioral inhibition system, inhibition, worldview defense, approach-avoidance motivation, ethnocentrism

We encapsulate ourselves to avoid death. And life escapes us while we huddle within the defended fortress (Becker, 1997, p. XIII).

# INTRODUCTION

According to cultural anthropologist Ernest Becker, humans are uniquely preoccupied as "terrified, death-avoiding animal[s]" (Becker, 1997, p. 99). Inspired by Becker's writings, terror management theory (TMT) proposes that humans avoid the threat of death on a symbolic level by identifying with something that outlasts their individual life, such as their nation or culture

#### Edited by:

*Snehlata Jaswal, Indian Institute of Technology Jodhpur, India*

#### Reviewed by:

*Aaron L. Wichman, Western Kentucky University, USA Simon Schindler, University of Kassel, Germany*

> \*Correspondence: *Mike Prentice mptg2@mail.missouri.edu*

#### Specialty section:

*This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology*

Received: *13 July 2016* Accepted: *29 September 2016* Published: *25 October 2016*

#### Citation:

*Agroskin D, Jonas E, Klackl J and Prentice M (2016) Inhibition Underlies the Effect of High Need for Closure on Cultural Closed-Mindedness under Mortality Salience. Front. Psychol. 7:1583. doi: 10.3389/fpsyg.2016.01583* (Pyszczynski et al., 1997). In support of this proposition, numerous TMT studies have demonstrated that people respond to mortality salience (MS) with increased adherence to their cultural worldviews, which often manifests in avoidance and derogation of culturally different people, customs, and ideologies (Burke et al., 2010). Hence, this research reveals a fascinating irony inherent in human efforts at self-preservation: in order to manage death-related anxiety, people develop an aversion to the cultural "other," thereby curtailing their opportunities for psychologically enriching experiences. In other words, an unfortunate consequence of "huddling in the defended fortress" is that people avoid sampling opportunities provided by getting to know other cultures and their members.

Although it is well-documented that MS can drive cultural closed-mindedness, little is known about the basic neural and motivational processes that underlie this effect. According to the process model of threat and defense (Jonas et al., 2014), MSinduced aversion to cultural otherness may be mediated by a very basic motivation to avoid punishment. In the present research, we test this assumption using neurobehavioral measures of the activity of the behavioral inhibition system (BIS) as measured by the line bisection task (LBT; Study 1) and frontal alpha asymmetry as measured by EEG (Study 2). We also directly manipulate the proposed inhibitory motivational mediator. Specifically, we investigate whether (a) MS activates inhibitory motivational processes, (b) inhibition mediates MS effects on cultural closed-mindedness, and (c) these effects are particularly strong for persons with a dispositional inclination to avoid unfamiliar and potentially threatening stimuli.

# Terror Management and Avoidance Motivation

According to TMT, the awareness of one's inevitable demise entails the potential for paralyzing terror, which can be managed by defensive processes that are oriented toward symbolic self-preservation (Pyszczynski et al., 1997). Symbolic selfpreservation is attained by identifying with entities that transcend one's individual death, such as cultural ingroups. Accordingly, there is ample evidence that MS increases cultural closedmindedness in the form of more negative (positive) evaluations of the culturally unfamiliar (familiar; Burke et al., 2010).

Terror management theorists have generally assumed that these defenses are products of a basic motivation to avoid threatening cognitions (Pyszczynski et al., 1997, 2000, 2012). According to TMT, mortality threat is initially avoided by cognitive means, such as death-thought suppression or vulnerability denial (i.e., proximal defense; Pyszczynski et al., 1999). Second, after this active cognitive avoidance has subsided, subtle mechanisms of threat-avoidance are utilized to maintain defenses on a symbolic level (i.e., distal defense; Pyszczynski et al., 1999). These distal defenses often manifest in various forms of ethnocentric intergroup bias (e.g., avoidance and derogation of foreign people and worldviews; Burke et al., 2010). In other words, defensive responses to MS are assumed to be mediated by the motivation to avoid experiencing terror in the face of inescapable mortality (Greenberg et al., 2003). TMT argues that "when confronted with reminders of their mortality, people avoid the subjective experience of distress by increasing their commitment to the cultural worldview" (Greenberg et al., 1995b, p. 431). In sum, TMT proposes that threat leads people to shy away from cultural outgroups and cling to cultural ingroups in order to avoid the distress inherent in the existential insecurities aroused by reminders of the finitude of life.

# The General Process Model of Threat and Defense

Recent neuropsychological theorizing suggests that MS-induced discomfort with cultural otherness is rooted in a basic motivation to avoid negative/punishing outcomes (Jonas et al., 2014). Specifically, the process model of threat and defense (Jonas et al., 2014) states that MS activates the BIS, an avoidance motivational system that responds to distant or anticipated threat (such as one's eventual demise), which inhibits ongoing goal approach (McNaughton and Corr, 2004). The inhibition of goal approach functions to reorient the individual's attention so that threat can be resolved and viable goal pursuit resumed (cf. McGregor et al., 2009). This dynamic is accompanied by several attentional, affective, and motivational processes, including hypervigilance, anxious arousal, and the motivation to avoid threat by inhibiting behaviors that might put the individual at risk (passive avoidance; Gray and McNaughton, 2000; McNaughton and Corr, 2004; Corr et al., 2013a) 1 .

Importantly, upon detecting stimuli signaling the potential for conflict or punishment, the BIS further biases negativity, which entails a tendency to avoid everything that is unfamiliar and potentially threatening (McNaughton and Corr, 2004, 2014). Accordingly, passive threat-avoidance strategies include the inhibition of exploratory approach behavior in order to maintain distance from potentially threatening stimuli (e.g., here, potentially hostile outgroup members), especially in risky or unfamiliar contexts. From an evolutionary perspective, members of cultural outgroups represent unfamiliar and potentially threatening stimuli, since hostile strangers may have been among the most serious ancestral threats to survival (McEachron and Baer, 1982; see also Kenrick et al., 2010). Accordingly, we propose that reminders of mortality activate a basic threat-avoidance system—the BIS—that subsequently inhibits the motivation to engage with cultural outgroups. This proposed process is similar to the TMT account of MS sequelae, but it differs in one key way: people's defensive responses to MS are not merely anticipatory of a negative motivational-emotional state (in TMT, the potential for anxiety; Pyszczynski et al., 1999), but instead these defenses are driven by BIS-mediated processes, such as

<sup>1</sup>Corr et al. (2013a) propose a distinction between a system for passive avoidance, generated by the BIS (e.g., inhibition of approach motivational behavior toward unfamiliar and potentially threatening people or places), and a system for active avoidance, produced by the fight-flight-freeze system (FFFS; e.g., running away from a predator). Contrary to the BIS that responds to more distant, abstract, and ambiguous threats, the FFFS responds to more proximate, concrete, and unequivocal threats, such as immediate attackers. The present line of research pertains to existential threats, which are more distant, abstract, and ambiguous (e.g., the time and way of one's death is usually unknown; see also Huang and Wyer, 2015). Therefore, we focus on BIS-mediated passive avoidance motivational responses to threat in this paper.

motivational inhibition and felt anxious uncertainty, that are aroused immediately following the contemplation of one's death.

Although increased ethnocentric thinking has been frequently found following MS (Burke et al., 2010), the mediating processes proposed by the general process model (Jonas et al., 2014) are to date largely hypothetical. Advancing on this empirical gap from the general process model is all the more important given what is arguably equivocal evidence for TMT's main cognitive mediator candidate—death-thought accessibility (Das et al., 2009; Golec de Zavala et al., 2012; Trafimow and Hughes, 2012; Agroskin and Jonas, 2013; Hart, 2014). This failure has spawned a lively debate in the existential threat literature about whether MS-induced ethnocentrism might be driven by specific needs to restore meaning (Proulx and Heine, 2010), certainty (van den Bos et al., 2005), security (Hart et al., 2005), a sense of coalition (Navarrete et al., 2004), or control (Fritsche et al., 2008). However, it has also been proposed that MS-induced defensiveness may be driven by rather basic processes such as approach motivational (McGregor et al., 2013) or unconsciously vigilant (Holbrook et al., 2011) responses to any kind of threat. By contrast, Jonas et al. (2014) have suggested that MS effects on ethnocentric cognitions and behaviors are attributable to BIS-generated passive avoidance of cultural novelty. The present research tests this novel notion.

# Individual Differences in MS Effects on Cultural Closed-Mindedness

TMT research has demonstrated that individual differences in traits related to motivated avoidance of the unfamiliar are useful predictors of terror management processes. For instance, persons with high need for closure (NFC; Webster and Kruglanski, 1994), who strive for simple, unambiguous, and predictable environments, are accordingly reluctant to explore novel or ambiguous stimuli under threat, such as unfamiliar places (Vess et al., 2009) 2 . They also tend to adopt ethnocentric or xenophobic attitudes, suggesting that their discomfort with ambiguity and unfamiliarity biases their evaluations of cultural outgroups (Shah et al., 1998; Agroskin and Jonas, 2010, 2013). Accordingly, high NFC has been found to amplify MS effects on derogation of culturally different beliefs, such as differing religious worldviews (Juhl and Routledge, 2010). On a more general level, the unease with unfamiliar, unpredictable, and ambiguous stimuli following MS exhibited by people with high trait NFC can also manifest in heightened stereotypic thinking, as well as antipathy toward behaviorally inconsistent persons and even modern art (Schimel et al., 1999; Landau et al., 2004, 2006). Low NFC, conversely, is related to openness to novelty, uncertainty, and ambiguity, thus promoting increased interest in exploring unfamiliar stimuli following MS (Vess et al., 2009). In sum, high NFC appears to represent a disposition to avoidance-related cognition and behaviors, especially under threat.

This avoidance-motivational interpretation of NFC is bolstered by the fact that chronic BIS sensitivity and NFC are positively related (Corr et al., 2013b). This again suggests that high NFC may be indicative of the BIS-mediated negativity bias toward exaggerated threat perceptions and subsequent avoidance behaviors (McNaughton and Corr, 2004). The reluctance to engage with culturally-different stimuli following MS exhibited by high NFC people may reflect BIS-mediated inhibition of approach motivation under threat. On a more general level, Jonas et al. (2014) have proposed that BIS-related traits intensify and prolong BIS activity after threat. This is in line with neuroscientifically-informed theorizing regarding the hierarchical interplay between trait and state avoidance motivation (Elliot, 2006; Spielberg et al., 2013). In sum, people with high trait NFC are likely to show increased levels of BISmediated passive avoidance following MS, which may explain why NFC has been linked to many avoidance-related outcomes in previous TMT research.

# The Role of BIS Activity in Mortality Salience Effects

Although the hypothesis that MS effects on ethnocentric thinking are mediated by BIS activity has not yet been explicitly tested yet, there is some evidence suggesting an association between MS and BIS-related processes. For instance, people prefer familiar over unfamiliar products after experiencing death anxiety, suggesting a basic tendency toward novelty-avoidance (Huang and Wyer, 2015). MS also causes people to avoid stimuli that remind them of their embodied, temporal existence (Goldenberg et al., 2001; Cox et al., 2007), which suggests that MS inclines an avoidance of information that would further highlight a threat. Finally, people are even reluctant to look into mirrors and engage in self-focused writing under MS, possibly because heightened selfawareness exacerbates distressing vulnerability concerns after contemplating one's own death (Arndt et al., 1998).

In line with the process model of threat and defense (Jonas et al., 2014), these avoidance reactions are particularly likely to occur for individuals with anxious personalities. For example, MS has been found to aggravate anxious and avoidant behaviors among people especially prone to phobic or compulsive behaviors (Strachan et al., 2007), such as people high on BIS and trait anxiety (Smits and Boeck, 2006; Bijttebier et al., 2009; Erdle and Rushton, 2010). Furthermore, stimuli related to physicality and creatureliness are particularly problematic for persons high in neuroticism after MS (Goldenberg et al., 2000, 2006a,b, 2008). Finally, research from diverse labs demonstrates noveltyavoidant, risk-averse behaviors under conditions of threat and anxiety (e.g., Maner and Schmidt, 2006; Maner et al., 2007; Mortensen et al., 2010; Litt et al., 2011; Clark et al., 2012; see Kenrick et al., 2010 for review), and these reactions appear to be especially likely for people high on BIS-related traits (Landau and Greenberg, 2006; Cavallo et al., 2009; Routledge et al., 2010). Taken together, these results are consistent with the notion that BIS-sensitive people are particularly prone to passive-avoidant behaviors in the wake of threat (Jonas et al., 2014).

# Measuring BIS Activity

In neural terms, BIS activity has been linked to relative righthemispheric brain activity, as reflected in prefrontal EEG alpha asymmetry. Specifically, dispositional BIS sensitivity is

<sup>2</sup> In some studies discussed in this paper, NFC was measured with the personal need for structure (PNS) scale (Neuberg and Newsom, 1993), which has been used as an alternative operationalization of NFC in prior TMT research (e.g., Dechesne et al., 2000).

associated with stronger relative right-frontal brain activation, whereas BAS sensitivity is robustly related to greater left-frontal activation (Sutton and Davidson, 1997; Shackman et al., 2009). There is also evidence linking frontal asymmetric activity to prevention/promotion regulatory focus, which is conceptually related to BIS/BAS activity, respectively (Amodio et al., 2004). Importantly, relative frontal EEG activity is computed using difference scores between right- and left-frontal electrodes, implying that increased BIS-related right-frontal asymmetry is functionally equivalent to reduced BAS-related left-frontal asymmetry and vice-versa (e.g., Sutton and Davidson, 1997). This is also consistent with an inverse interrelationship between the activity of both hemispheres (e.g., Amodio et al., 2004) and the joint subsystems hypothesis of BIS/BAS activity in reinforcement sensitivity theory (Corr, 2004), which states that the two systems mutually inhibit each other. Thus, BIS-mediated inhibition of approach motivation is equivalently reflected in reduced leftfrontal asymmetry as well as increased right-frontal asymmetry in the context of EEG asymmetry scores (see also Coan and Allen, 2004).

Moreover, right-frontal EEG asymmetry has been found to contribute to numerous neural and behavioral indicators of anxious or inhibitory responding. For example, relative right-frontal activity is positively associated with state anxiety (Davidson et al., 2000; Harmon-Jones et al., 2009) and hypervigilance to threat (Pérez-Edgar et al., 2013; Grimshaw et al., 2014). It also predicts the amplitude of the error-related negativity (Nash et al., 2012), which is linked to BIS sensitivity (Boksem et al., 2006, 2008). Importantly, it has also been demonstrated that right-frontal asymmetry is specifically related to the anxious inhibitory state of passive avoidance, whereas left-frontal asymmetry has been linked to the fearful state of active avoidance (Wacker et al., 2008; see also Perkins et al., 2007 for behavioral evidence on the fear/anxiety distinction). Overall, there is a substantial body of evidence to support the notion that right-frontal EEG asymmetry is associated with numerous neural and behavioral indicators of anxious or inhibitory responding (for more evidence, see Thibodeau et al., 2006; Wacker et al., 2010; McNaughton et al., 2013). Convergent evidence from functional magnetic resonance imaging (fMRI) has also demonstrated a link between right-frontal brain activity and anxiety (e.g., Dalton et al., 2005; Engels et al., 2010) and inhibition (see Aron et al., 2004 for review).

Nash et al. (2010) demonstrated that BIS/BAS activity can also be investigated using the line bisection task (LBT) as a perceptual measure of relative prefrontal hemisphericity. The LBT gauges BIS/BAS activity (or left-frontal asymmetry) by determining the extent to which people's perceptions of the midpoints of horizontal lines are biased to the right visual field, mirroring neural activity in the contralateral hemisphere (Nash et al., 2010). Accordingly, reduced rightward line bisection bias (i.e., reduced left or increased right hemisphericity) has been associated with BIS-related phenomena, such as anxious-avoidant arousal (Friedman and Förster, 2005), feelings of passivity/inhibition, and powerlessness (Drake and Myers, 2006; Wilkinson et al., 2010), as well as reduced risk taking, optimism, and goal pursuit (Drake and Ulrich, 1992; Nash et al., 2011). Notably, line bisection bias is specifically related to frontal (F7/F8) but not central, parietal, temporal, or occipital asymmetry, suggesting that the LBT represents a neurobehavioral marker specific to frontal alpha asymmetry (Nash et al., 2010). In sum, the LBT has often been used to measure motivational tendencies (see Vallortigara and Rogers, 2005 for a review of motivationallyrelevant perceptual asymmetries), and has been validated as a neurobehavioral measure of asymmetric frontal brain activation (Nash et al., 2010).

# The Present Research

In the present research, we examined the hypotheses that (a) MS causes BIS activity, evidenced by increased by greater righthemispheric frontal asymmetry, (b) BIS activity mediates MS effects on cultural closed-mindedness, and (c) these effects are pronounced among persons with high NFC. Stated differently, we tested the moderated mediational hypothesis that MS effects on cultural closed-mindedness are mediated by BIS activity and are particularly strong for persons with high NFC. These hypotheses were examined by inducing MS (vs. a control topic), measuring NFC, measuring frontal brain asymmetry via the LBT (Study 1) and EEG (Study 2), and measuring cultural closed-mindedness in the form of reluctance to engage with cultural novelty (Study 1) and general ethnocentrism (Study 2). Moreover, in order to substantiate our claim that inhibition is the mediator, we directly manipulated the proposed mediator—inhibition—in Study 3 to test whether inhibition evokes ethnocentric thinking for high NFC persons (cf. Spencer et al., 2005).

# STUDY 1

Study 1 was designed to test whether (a) persons with high NFC respond with increased BIS activity (as assessed by LBTassessed right-frontal activity) to MS, and (b) their increased BIS activity mediates their increased unwillingness to explore foreign cultures.

# Method

# Participants and Design

Seventy-seven students at the University of Salzburg participated in this paper-pencil study, which took place in class, prior to a lecture<sup>3</sup> . Three participants failed to follow instructions and were dropped prior to analyses, leaving a final sample of 74 participants (54 females and 20 males). Their mean age was 22.0 years (SD = 1.8; range: 19–31). Participant nationalities were 54 German, 18 Austrian, and 2 other. Participants were randomly assigned to one of two conditions, MS (N = 35) or TV salience (N = 39), in a between-subjects design. The study was described as an investigation of personality and visual perception. The study was approved by the ethics committee of the University of Salzburg. All participants signed informed consent, and could withdraw participation at any point, although no participant made use of this option.

<sup>3</sup>Regarding sample size considerations, MS has an expected effect size of MS at d = 0.75 on ethnocentrism and related outcomes (Burke et al., 2010). Thus, power for a main effect is at 80% with a sample of n = 60. Given that a personality moderator should add prediction to a model, power was expected to be well over 80% for the current studies. For example in the current research domain, Vess et al. (2009) observed NFC × MS interaction effects on meaning in life in three studies (1a–1c) with an average d = 0.90.

# Procedure and Materials Need for Closure

Following instructions and the baseline LBT measure (described in more detail below), the NFC scale was presented. We measured NFC with three items that were taken from the German version of the NFC scale (Webster and Kruglanski, 1994; Schlink and Walther, 2007). The items were selected a priori based on two criteria: First, representativeness of the whole NFC scale, that is, corrected item-total correlations presented by Schlink and Walther (2007). Second, the items that best capture the willingness to approach/avoid novel and unpredictable stimuli should be of particular relevance (Green and Campbell, 2000), contrary to decisiveness, which is also encompassed by the NFC scale (see also Neuberg et al., 1997 for a critique of the full NFC scale's conceptual imprecision). One item was dropped prior to analyses because its corrected item-total correlation suggested it was unrelated with the other two items in the current sample (rit = 0.00). The two remaining items were: "I prefer familiar things over unfamiliar and unpredictable ones" and "I don't like unpredictable situations" (interrelation: r = 0.53, p < 0.001). A pilot study revealed this short NFC scale to correlate strongly with the conceptually related personal need for structure scale by Neuberg and Newsom (1993), r = 0.72, p < 0.001, N = 217. Responses were given on a 1 (almost never) to 6 (almost always) scale, and averaged to create a single composite score (M = 3.93, SD = 0.85).

# Manipulation

After a few further personality items, included to bolster the cover story, we presented the classic MS manipulation, asking participants two open-ended questions about what will happen to them physically when they die and after they are dead, and which emotions are caused by thinking about their own death. Participants in the control condition were asked parallel questions about watching TV.

# Line Bisection Task (Inhibition of Approach Motivation)

Following a free-thought delay of a few minutes, the LBT was presented for a second time (the first LBT was presented right at the beginning of the study to obtain a baseline LBT measure). In each measure, participants were instructed to mark the perceived center point of eight horizontal lines that ranged from 4 to 6 (102 to 152 mm) in length with their centers offset from each other (see also Drake and Myers, 2006, who used the same LBT). Estimation errors to the left reflect an overnoticing of the left visual field characteristic of relative right cerebral hemisphericity (Jewell and McCourt, 2000). Thus, we computed the index of relative right cerebral hemisphericity by subtracting each participant's number of right-of-center ticks from left-of-center ticks (see McGregor et al., 2010, Study 1). In line with prior research, both LBT measures were correlated, r = 0.67, p < 0.001 (McGregor et al., 2010), and participants exhibited a leftward bisection bias on average (LBT 1: M = 1.92, SD = 4.12; LBT 2: M = 1.10, SD = 3.80; Jewell and McCourt, 2000).

#### TABLE 1 | Regression statistics for the prediction of cultural novelty avoidance, Study 1.


*N* = *74. Model R<sup>2</sup>* = *0.16.*

#### Avoidance of Cultural Novelty

Passive avoidance of cultural novelty was measured with eight researcher-generated items designed to measure interest in exploring culturally foreign stimuli (e.g., people, information). Based on the conceptual view that exploration constitutes approach behavior toward novel stimuli (Green and Campbell, 2000), and evidence implicating BIS activity in reduced exploratory approach behaviors (Green and Campbell, 2000; Elliot and Reis, 2003; Kashdan et al., 2004), we recoded our cultural exploration items to obtain a measure for passive avoidance of cultural novelty. Example items include "I would like to get to know people from foreign cultures" and "I am curious to learn how people live in foreign cultures<sup>4</sup> ." Participants were asked to indicate the extent to which the items describe them at the present moment. Responses were given on a 1 (very little) to 5 (very much) scale, and averaged to create a single composite score (8 items; M = 2.40, SD = 0.80, α = 0.90).

After that, we included some newly created personality items that were unrelated to an interest in exploring culturally foreign stimuli, a newly created cross-cultural conflict resolution task that included a manipulation, and a criterion variable for explorative reasons. Finally, we collected some demographic data and measured participants' handedness (Oldfield, 1971).

# Results

# Threat × NFC on Cultural Novelty Avoidance and Avoidance Motivation

As an initial step we sought to confirm that MS heightens avoidance of cultural otherness especially at high NFC. For this, effect-coded threat (with zero as the non-threat and one as the threat condition), NFC, and the threat × NFC term were entered as predictors of cultural avoidance. Results are displayed in **Table 1**. Both the main effects of threat and NFC were significant. These main effects were qualified by the predicted interaction (1R <sup>2</sup> = 0.06) such that avoidance of cultural novelty was especially high under MS vs. control for people high (+1 SD) on NFC [b = 0.59, SE = 0.26, t(70) = 2.28, p = 0.025, see **Figure 1**]. There was no effect of threat at low NFC (−1 SD), b = −0.26, SE = 0.26, t(70) = 1.09, p = 0.277.

Next, we tested our key hypothesis that MS causes BIS activity—as indicated by a shift in leftward bisection bias—for persons with high NFC (the mediator model). In line with prior research using the LBT and to control for hemispheric differences arising from handedness (e.g., Drake and Myers, 2006), we

<sup>4</sup> In contrast to Vess et al. (2009) who used Green and Campbell's (2000) exploration scale, we developed new items for cultural novelty-avoidance because their exploration scale largely lacks explicit reference to culturally foreign stimuli.

TABLE 2 | Regression statistics for the prediction of post-threat LBT bisection error.


*N* = *58. Model R<sup>2</sup>* = *0.52.*

excluded 16 left-handed individuals from all analyses involving the LBT (Oldfield, 1971). Furthermore, the LBT baseline measure was entered as a predictor in all analyses including the LBT; thus, effects on LBT can be considered demonstrations of residual change. Baseline LBT, effect-coded threat, NFC, and the threat × NFC interaction were entered as predictors of post-threat LBT. As displayed in **Table 2**, there were no main effects, and the threat × NFC interaction was significant (1R <sup>2</sup> = 0.060). Avoidance of cultural novelty was especially high under MS vs. control for people high on NFC, though this effect did not emerge until very high values of NFC (∼+2 SD or the 95% of our observed distribution according to the Johnson-Neyman technique, 1936). There was an unpredicted effect of threat at low NFC (−1 SD), b = −2.42, SE = 1.11, t(53) = 2.17, p = 0.034. **Figure 2** shows that the effect of NFC on bisection bias under MS reversed from control such that NFC predicted more avoidance under MS.

#### Test of Moderated Mediation

Next, we tested whether BIS activity transmits or explains the effect of the interaction of MS and NFC to cultural avoidance. Because NFC is linked to both sensitivity to threat and motivated closed-mindedness under threat, it was allowed to moderate both the path from the threat to the mediator, as well as from the mediator to the dependent variable (i.e., model 58; Hayes, 2013).

Results of this analysis are displayed in **Figure 3**. In sum, the effect of post-threat BIS activity explained the effect of threat × NFC on cultural novelty avoidance, especially at high levels of BIS and NFC.

# Brief Discussion

These results provide the first evidence for the notion that (a) MS leads to increases in BIS activity, (b) BIS activity mediates MS effects on aversion to cultural novelty, and (c) these effects are particular to people high on NFC. Thus, the findings demonstrate that a basic, passive-avoidance motivational process underlies higher-order manifestations of behavioral inhibition, such as the reluctance to explore unfamiliar cultures. This dynamic was only present for high NFC individuals, suggesting that they are particularly inclined to avoid the terror of death through motivated closed-mindedness toward cultural otherness.

One limitation of the current study was that our cultural novelty-avoidance measure had not been used prior to the study, so its construct validity was questionable. We thus conducted a follow-up study to explore the relationship between this measure and a frequent operationalization of cultural worldview defense. We found that the cultural novelty-avoidance measure used in Study 1 correlated with a measure of general ethnocentrism, r(30) = 0.37, p = 0.037. This ethnocentrism measure has been used as a worldview defense measure in prior TMT research (Agroskin and Jonas, 2013), and found to be related to NFC (Agroskin and Jonas, 2010). Thus, our cultural novelty-avoidance items may at least partly tap into those destructive aspects of ethnocentric intergroup bias that are characteristic of cultural worldview defense.

Study 1 provided strong initial support for the proposition that MS activates the BIS particularly in high NFC people, which explains their post-threat closed-mindedness. However, we used a fairly indirect measure of frontal asymmetry measured via the LBT. In Study 2, we turned to a more direct measure of frontal

asymmetry available via EEG. Furthermore, Study 1 employed a short NFC scale instead of the full measure. Thus, Study 2 provides an opportunity for replication and extension of the findings of Study 1.

# STUDY 2

Building on Study 1, Study 2 employed a poem-based MS manipulation (Agroskin and Jonas, 2013), the full NFC scale (Schlink and Walther, 2007), and frontal EEG alpha asymmetry instead of the LBT as a more direct measure of BIS-related neural activity. We also used the general ethnocentrism measure from the follow-up study instead of cultural novelty-avoidance to tap into the destructive aspects of cultural closed-mindedness that often characterize worldview-defensive negativity toward cultural outgroups. Summing up, Study 2 was designed to provide multimethodological and conceptual replication for the findings of Study 1.

# Method

#### Participants and Design

The sample consisted of 33 participants (23 female) from the University of Salzburg without any reported history of neurological disorders or prior head trauma<sup>5</sup> . Their mean age was 22.1 years (SD = 3.0; range: 18–32)<sup>6</sup> . Twenty-two participants held German and eleven Austrian nationalities. Participants were randomly assigned to one of two conditions, MS (N = 16) or TV salience (N = 17), in a between-subjects design<sup>7</sup> . All participants gave written informed consent to participate in the study, which ostensibly investigated topics associated with literature and personality. They were rewarded with course credits. The study was approved by the ethics committee of the University of Salzburg. All participants signed informed consent, and could withdraw participation at any point, although no participant made use of this option.

# Procedure and Materials

#### Need for Closure

Prior to coming to the lab for the EEG measurement, participants were emailed and asked to fill out a few online questionnaires, including the whole NFC scale (Schlink and Walther, 2007) whose short form was used in Study 1 8 . Responses were given on a 1 (totally disagree) to 6 (totally agree) scale, and averaged to create a single composite score (14 items; M = 3.23, SD = 0.70, α = 0.82)<sup>9</sup> .

## Manipulation

Upon arriving at the lab, participants were asked to provide some demographic data. After that, they underwent a 90-s baseline EEG recording (see below for more detail). Then, they were presented a short poem related to death or a neutral control topic (weather). The death-related poem has been used as a MS manipulation before (Agroskin and Jonas, 2013, Study 2). Literally translated from German, it reads "The hardest and cruelest power, which man has to accept, is death." The weatherrelated poem reads "Look to the sky, clouds are passing by. Some are raining, others are not. Some are flashing, others are not." Participants were asked to read the poem and describe how they interpret it, what they associate with it (e.g., images, sounds, smells, and moods), whether they have experienced anything in their own lives that showed them that the poem was true, and

<sup>5</sup>Due to the intensive nature of EEG data collection and analysis, collection was stopped after reaching an n of 15 per cell plus 10% for anticipated data loss.

<sup>6</sup>Age was unequally distributed across experimental conditions. Participants were significantly younger in the MS condition (M = 20.4, SD = 1.93) than in the control condition (M = 23.7, SD = 2.91), t(31) = 3.78, p < 0.001. We included age as a covariate in all analyses reported below. Running the analysis without this covariate did not impact the status of the significance tests or the substance of the conclusions.

<sup>7</sup>This study was part of a larger project that included other experimental conditions for exploratory reasons. Specifically, conditions not reported here were presented with other poems about existential topics that did not refer to the subject of death

and that were not designed as neutral control conditions either. Given that this study was primarily designed as a conceptual replication of Study 1, we present only the analyses that concern the MS and neutral control conditions.

<sup>8</sup> In addition to the 33 participants reported above, nine more persons participated in the EEG study but failed to complete the personality questionnaires. They were dropped from all analyses.

<sup>9</sup>Two items were dropped prior to analyses because of very low corrected itemtotal correlations (rits < 0.20).

Agroskin et al. NFC and BIS in Ethnocentrism

whether their own existence had certain aspects reflecting the veracity of the poem. These questions were used to help ensure a deep and thorough processing of the poems and their meaning.

#### EEG (Inhibition of Approach Motivation)

Right after the manipulation, participants underwent another 90-s EEG recording in order to assess post-threat avoidance motivation as a function of MS. During all EEG recordings participants were asked to fixate a small black cross in the middle of the screen. After this EEG measurement, participants filled out the 20-item Positive and Negative Affect Schedule (PANAS; Watson et al., 1988). Then, participants underwent one more 90-s EEG recording.

#### Ethnocentrism

Following a delay of a few minutes (involving questions about which TV programs participants would like to watch and their reactions to two press reports about crimes10), which is necessary to obtain MS effects following explicit MS inductions (Pyszczynski et al., 1999), general ethnocentrism was measured using a short form of the ethnocentrism scale by Bizumic et al. (2009). This short form has been previously used by Agroskin and Jonas (2013) as a measure of cultural worldview defense, and was found to be associated with NFC, right-wing authoritarianism, avoidance of empathy, and anti-immigration attitudes (Agroskin and Jonas, 2010). Sample items are "In most circumstances it is right and natural to favor members from one's own cultural or ethnic group over strangers or foreigners" and "We need to do what's best for our own people, and stop worrying so much about what the effect might be on other peoples." Responses were given on a 1 (totally disagree) to 6 (totally agree) scale, and averaged to create a single composite score (4 items; M = 2.10, SD = 0.91, α = 0.71)<sup>11</sup> .

Then, for exploratory reasons, participants underwent another 90 s EEG recording trial and completed some questions about their attitudes toward the European currency. Finally, they indicated their nationality and were probed for suspicion. Our focal BIS-related hypotheses were not correctly guessed by any participant<sup>12</sup> .

## EEG Apparatus and Data Reduction

We used a 14-channel Emotiv EEG neuroheadset (Emotiv Systems Inc., San Francisco, CA, USA), recording data with Emotiv TestBench software at a sampling rate of 128 Hz. EEG data was recorded at sites AF3, AF4, F3, F4, F7, F8, FC5, FC6, P7, T7, T8, P8, O1, and O2 in accordance with the 10–20 International System. Two mastoid electrodes served as online reference. The Emotiv EEG neuroheadset employs goldplated contact-grade hardened copper electrodes with salinemoistened felt pads to record EEG. Importantly, the Emotiv EEG device has been found to record similar suppression of alpha power (8–13 Hz) in the Left Precentral AAL (Automated Anatomical Labeling) region in response to imagined rightfinger tapping trials as compared to a standard Biosemi Active-II 64-channel device (see also Stopczynski et al., 2014a,b). This suggests that the Emotiv device is suitable for investigating neural processes associated with regional changes in alpha power. Further, validation studies determined that when used no longer than about 60 min per recording session, the ERP signal quality produced by this device is comparable to (or only slightly lower than) the signal quality provided by traditional EEG systems (Badcock et al., 2013; Mayaud et al., 2013). Consequently, Emotiv EEG technology has recently been increasingly used for cognitive neuroscience and braincomputer interface (BCI) applications (Bobrov et al., 2011; Debener et al., 2012; Louwerse and Hutchinson, 2012; Choi and Jo, 2013; Khushaba et al., 2013; O'Regan et al., 2013; O'Regan and Marnane, 2013; De Vos et al., 2014; Vourvopoulos and Liarokapis, 2014; Aspinall et al., 2015; Steinhubl et al., 2015).

Using BrainVision Analyzer 2.0 (Brain Products, Gilching, Germany), EEG data were filtered with an IIR filter (highpass cutoff: 0.1 Hz, Slope: 24 db/Oct; low-pass cutoff: 30 Hz, Slope: 24 db/Oct). Next, the three 90-s EEG recording epochs were extracted from the continuous data and were further segmented into 2s segments (1.5s overlap). Epochs were included in condition averages when a difference between two values in a moving 200 ms interval did not exceed 100 microvolt, or when the signal was below −100 or above +100 microvolt. The 2s segments were then Fourier transformed (Fast Fourier Transform, 10% Hamming Window, frequency resolution 0.5 Hz) and the resulting power spectra were averaged. Individual frontal asymmetry scores (µV 2 ) for each epoch were calculated (log F7–log F8) in the alpha band (8–13 Hz) and exported for statistical analysis. Positive values index right-frontal hemisphericity indicative of avoidance motivation. Analyses focused on lateral frontal asymmetry (F7–F8) because these electrodes have been specifically linked to bisection bias in the LBT (Nash et al., 2010).

# Results

### Preliminary Analyses

Frontal asymmetry scores (L – R alpha) for the proximal and distal recordings were baseline adjusted for each participant so that they reflected changes in asymmetry, following past research on state frontal alpha asymmetry (e.g., Allen et al., 2001; Ravaja et al., 2013). The resultant asymmetry scores for the proximal and distal recordings were highly correlated, r = 0.63, p < 0.001. Given that Jonas et al. (2014) assume anxious persons to show prolonged BIS activity following threat, and NFC is known to affect proximal and distal MS effects in a similar way (Juhl and

<sup>10</sup>The press reports described crimes committed in Austria (or by Austrians). Then, participants' intentions to punish the perpetrators were measured. Given that 66.7% of our participants were Germans, we could not analyze their punishment intentions in response to Austrian crimes as indicators of their ethnocentric, worldview-defensive attitudes (cf. Rosenblatt et al., 1989). We also included four items measuring worldview-defensive prejudice toward immigrants (Agroskin and Jonas, 2010) right after the focal ethnocentrism measure (see below), which were not analyzed for the same reason (i.e., most of our participants were immigrants themselves).

<sup>11</sup>Two items were dropped prior to analyses because of very low corrected item-total correlations (rits < 0.20).

<sup>12</sup>Two participants mentioned death-related thoughts and ethnocentric attitudes, though. Dropping these two participants did not substantially change the results reported below, which is why we kept them in the final sample.

Routledge, 2010), we merged the proximal and distal asymmetry scores.

# Primary Analyses

#### Analytical Strategy

As in Study 1, the hypotheses were tested in several steps using moderated mediational analyses (Hayes, 2013) performed using Mplus 7 (Muthén and Muthén, 2012). The zero-order correlations are not explicitly reported because they were virtually identical (i.e., same directions and similar magnitudes and significance levels) with the respective main effects reported below<sup>13</sup> .

# MS × NFC Effect on EEG-Inhibition

First, we tested whether individuals with high NFC showed higher right-frontal (or reduced left-frontal) asymmetry following MS (coded as MS = 1, weather = 0) by regressing the post-threat asymmetry score onto pre-threat asymmetry, threat, NFC, and the threat × NFC interaction. There was no effect of NFC in the control condition, b = −0.05, SE = 0.06, t(28) = 0.78, p = 0.443 and a significant main effect of threat such that avoidance was higher under MS than control, b = 0.69, SE = 0.31, t(28) = 2.23, p = 0.025. This effect was qualified by a significant interaction between the manipulation and NFC, b = 0.24, SE = 0.09, β = 0.43, t(28) = 2.56, p = 0.01, 1R <sup>2</sup> = 0.16 (see **Figure 4**). Simple effects analyses revealed that high NFC individuals (SD = 1) showed the predicted increase in right-frontal asymmetry following MS, b = 0.27, SE = 0.10, β = 0.69, t(28) = 2.64, p < 0.01, whereas this was not true for low NFC persons (SD = −1), b = −0.06, SE = 0.10, β = −0.16, t(28) = −0.65, p = 0.52. Thus, our hypotheses were corroborated: MS increased BIS-based inhibition of approach motivation but only for high NFC persons.

#### EEG-Inhibition Effect on Ethnocentrism

Right-frontal asymmetry was entered as a predictor in addition to the manipulation and NFC. The interaction between the manipulation and NFC (i.e., direct interactive effect) was fixed to zero because we assumed a complete mediation model (Shrout and Bolger, 2002). As expected, right-frontal asymmetry predicted ethnocentrism, b = 2.49, SE = 0.69, β = 0.54, t(29) = 3.59, p < 0.001. No other relationships attained significance (ps > 0.34). In contrast to Study 1, adding the interaction between right-frontal asymmetry and NFC did not reveal moderation by NFC, t(27) = −0.62, p = 0.53. Thus, BIS activity affected ethnocentrism regardless of NFC levels. We thus fixed the effect of NFC to zero in accordance with Hayes (2013, Model 7) for the moderated mediational analyses. Summing up, the more participants had right-frontal asymmetry, the more they adopted ethnocentric attitudes. Our hypotheses were therefore supported.

#### Bootstrap Estimation of the Indirect Effects

As per bias-corrected bootstrap procedure using 75000 bootstrap samples (Hayes, 2013, Model 7, see **Figure 5**), MS significantly increased ethnocentrism through right-frontal asymmetry for high NFC persons, point estimate (b) = 0.67, 95% CI [0.02, 1.43]. In contrast, there was no conditional indirect effect for low NFC persons, b = −0.15, 95% CI [−0.84, 0.33].

Finally, the overall model test was consistent with our hypothesis of complete mediation (Hayes, 2013, Model 7), χ 2 (2) = 0.65, p = 0.72, CFI = 1.00. Thus, our key mediational hypothesis for high NFC individuals was supported: MS amplified ethnocentric thinking via the BIS-based inhibition of approach motivation<sup>14</sup> .

# Discussion

These results replicate and expand the findings from Study 1 in several important ways. First, they provide a more direct measure of right-frontal asymmetric activity in the prefrontal cortex via EEG. Second, they involve another operationalization of motivated closed-mindedness in the cultural domain generalized ethnocentric attitudes—which suggests that a similar process may underlie various forms of worldview defense. All in all, the findings are clearly supportive of the view that high NFC persons' approach motivation is strongly inhibited following mortality reminders, which underlies their inclination to engage in ethnocentric thinking.

In Study 3, we aimed at conceptually replicating the effect of BIS-related inhibition on ethnocentrism by directly manipulating inhibition. Further, this test provides another means of testing

<sup>13</sup>Four participants had missing data in either the proximal or the distal EEG measurement. All reported analyses have been run with the whole sample using Full-Information-Maximum Likelihood (FIML) parameters in Mplus 7 (Muthén and Muthén, 2012). However, excluding the participants with missing data using listwise deletion did not change the results substantially (see also footnote <sup>15</sup>).

<sup>14</sup>All critical effects stayed significant using non-bootstrapped coefficients, including the conditional indirect effect for high NFC, b = 0.69, SE = 0.33, t = 2.10, p < 0.05. However, although the point estimate for the indirect effect was even somewhat increased, the bootstrap estimates became less robust, b = 0.69, 90% CI [0.05, 1.54], 95% CI [−0.11, 1.73]. This may be due to reduced degrees of freedom. Note, that the same pattern occurred when computing the model without the PANAS using listwise deletion instead of FIML, which likewise reduced degrees of freedom, b = 0.62, SE = 0.31, t = 2.04, p < 0.05 (non-bootstrap test), b = 0.62, 90% CI [0.04, 1.23], 95% CI [−0.11, 1.34] (bootstrap test; see also footnote 16). The results were virtually unchanged after controlling for the PANAS scales.

for mediation by manipulating the proposed mediator (Spencer et al., 2005).

# STUDY 3

In Study 3, we employed an experimental approach to testing mediation by manipulating the assumed mediator—inhibition. Building on research on disinhibition by van den Bos et al. (2009), we developed a novel manipulation of inhibition to test whether bringing people into a state of inhibition produces the same ethnocentric bias as threat and BIS activity. Thus, we measured NFC, induced inhibition, and measured ethnocentrism with the same instrument as in Study 2. In addition, we included a BIS affect measure as a manipulation check. We expected inhibition to promote ethnocentric thinking for persons with high NFC.

# Method

#### Participants and Design

Seventy-six students at the University of Salzburg participated in this paper-pencil study, which took place in class, after a lecture15. Participants were 58 females, 15 males, and 3 undisclosed. Their mean age was 21.6 years (SD = 2.0; range: 19–30). Fifty-five participants held German, 17 Austrian, and 1 another nationality (3 undisclosed). Participants were randomly assigned to one of two conditions, inhibition (N = 36), or control (N = 40), in a between-subjects design. The study was described as an investigation of participants' personalities. The study was approved by the ethics committee of the University of Salzburg. All participants signed informed consent, and could withdraw participation at any point, although no participant made use of this option.

# Procedure and Materials Need for Closure

Because the two-item NFC measure used in Study 1 was shown to be sensitive to MS, we measured NFC with same two items (M = 3.77, SD = 0.85; r = 0.44, p < 0.001).

#### Manipulation

After a few further personality items, included to bolster the cover story, we presented a novel manipulation of inhibition. Building on the disinhibition manipulation from van den Bos et al. (2009), we asked participants three open-ended questions about behaving with inhibitions: "Please briefly describe a situation out of your own life in which you felt and behaved in an inhibited way like you typically do." In the next two questions, we asked participants how they behaved and felt in this situation. In the control condition, we posed the same three questions referring to their behaving in a normal way (cf. van den Bos et al., 2009): "Please briefly describe a situation out of your own life in which you felt and behaved in a normal way like you typically do." The other two questions referred to participants' behavior and feelings again.

#### Ethnocentrism

Ethnocentrism was measured in the same way as in Study 2 (4 items; M = 1.94, SD = 0.77, α = 0.60)<sup>16</sup> .

## BIS Affect

Following Schumann et al. (2014), we included a retrospective manipulation check asking about how participants felt when responding to the inhibition/control condition-related questions. We presented five items measuring BIS-related affect, including four adjectives from the German translation of the Carver and White (1994) BIS sensitivity scale (Strobel et al., 2001), namely nervous (German: nervös), anxious (ängstlich), worked up (unruhig), and worried (besorgt), along with the face-valid

<sup>15</sup>With respect to the stopping rule for data collection in this study, we aimed at recruiting the same amount of participants as in Study 1, distributing questionnaires to all students who attended a lecture that is usually attended by about 80 people.

<sup>16</sup>After the ethnocentrism items, we included four items measuring prejudice toward immigrants (Agroskin and Jonas, 2010). These items were included as another measure of an ethnocentric, worldview-defensive attitude but were not analyzed because 75% of participants of this study were technically immigrants themselves (i.e., Germans studying in Austria).

adjective inhibited (gehemmt; five items; M = 2.32, SD = 1.45, α = 0.94). Participants responded on a scale ranging from 1 (not at all) to 5 (extremely). Finally, participants completed the PANAS (Watson et al., 1988) and some demographic questions<sup>17</sup> .

# Results

# Inhibition × NFC on Ethnocentrism

To examine whether high NFC persons respond with increased ethnocentrism to the inhibition recall-task, we used moderated regression analyses18. Whereas, there were no significant main effects (ps > 0.14), the interaction between the manipulation (coded as inhibition = 1, control = 0) and NFC was significant, b = 0.49, SE = 0.21, β = 0.27, t(69) = 2.38, p < 0.02, 1R <sup>2</sup> = 0.07 (see **Figure 6**). As predicted, high NFC individuals (SD = 1) showed significantly increased ethnocentrism following MS, b = 0.59, SE = 0.24, β = 0.39, t(69) = 2.41, p < 0.05, whereas this was not true for low NFC persons (SD = −1), b = −0.24, SE = 0.25, β = −0.16, t(69) = −0.98, p = 0.33. Thus, our hypotheses were corroborated: inhibition caused ethnocentrism in high NFC persons, whereas low NFC persons were not affected by inhibition in this way.

#### BIS Affect (Manipulation Check)

As expected, BIS affect was significantly higher in the inhibition condition (M = 3.63, SD = 1.24) than in the control condition (M = 1.32, SD = 0.47), t(35.40) = 9.65, p < 0.001, η <sup>2</sup> = 0.63<sup>19</sup> .

#### PANAS

Mood was also affected by inhibition. Negative affect was significantly increased and positive affect was significantly decreased by inhibition, ts > 2.91, ps < 0.01. Therefore, we reanalyzed the inhibition effect on BIS affect including the PANAS scales as covariates. Inhibition still increased BIS affect

notwithstanding the PANAS, F(1, 65) = 49.64, p < 0.001, η <sup>2</sup> = 0.43, although BIS affect was highly correlated with negative mood, r = 0.79, p < 0.001. There was no correlation between BIS affect and positive mood, r = −0.16, p = 20.

Likewise, the interactive effect of inhibition and NFC on ethnocentrism was virtually unaltered after controlling for the PANAS subscales, b = 0.44, SE = 0.22, t(63) = 1.95, p = 0.051. Persons with high NFC still showed increased ethnocentrism after inhibition, b = 0.67, SE = 0.31, t(63) = 2.14, p < 0.05, contrary to low NFC persons, b = −0.07, SE = 0.29, t(63) = −0.26, p = 0.80. The PANAS scales did not affect ethnocentrism (ps > 0.21).

# Exploratory Analyses: BIS Affect × NFC on Ethnocentrism

For the sake of convergent validity, we explored whether BIS affect elicited ethnocentrism for high NFC persons similar to the inhibition manipulation. We also controlled for the PANAS scales to rule out the possibility that the effect of BIS affect could be explained by mood more generally. There was a marginally significant main effect of BIS affect, b = 0.17, SE = 0.10, β = 0.32, t(65) = 1.69, p < 0.10, all other ps > 0.22. Moreover, a significant interaction between BIS affect and NFC occurred, b = 0.24, SE = 0.09, β = 0.39, t(65) = 2.73, p < 0.01, 1R <sup>2</sup> = 0.09 (see **Figure 7**). As predicted, high NFC individuals showed significantly increased ethnocentrism following MS, b = 0.36, SE = 0.12, β = 0.66, t(65) = 3.04, p < 0.01, whereas this was not true for low NFC persons, b = −40.05, SE = 0.13, β = −0.11, t(65) = −0.43, p = 0.67. Again, the PANAS scales did not affect ethnocentrism (ps > 0.21). Thus, BIS affect evoked ethnocentrism for high NFC persons in the same way as inhibition, and this effect was independent of changes in mood.

# Discussion

These results constitute an important conceptual replication of the relationship between BIS activity and ethnocentric thinking. Manipulating inhibition, the proposed mediator of MS effects on ethnocentrism, provided further support for the proposition that inhibition plays a causal role in generating ethnocentrism following MS. Persons with high NFC responded

<sup>17</sup>The negative mood scale included only eight instead of ten items because the items anxious and nervous were included in the BIS affect measure.

<sup>18</sup>Three participants failed to complete the ethnocentrism measure.

<sup>19</sup>Seven participants failed to complete the BIS affect and PANAS measures. We also used the adjusted degrees of freedom because variances were not equal as indicated by a significant Levene's test (p < 0.001).

with increased ethnocentrism to not only induced inhibition but also experiences of BIS-related affect. These results provide important multimethod evidence for the notion that not only perceptual and neural (Studies 1–2) but also cognitive-affective indicators of BIS-mediated inhibition can provoke ethnocentric thinking, and these effects are particularly strong for persons with high NFC.

# GENERAL DISCUSSION

In this paper, we have presented evidence from three studies that shed light on several previously uninvestigated questions that have recently gathered much attention in the existential threat literature (Tritt et al., 2012; Jonas et al., 2014). We found that (a) MS causes BIS activity, as indexed by right-hemispheric asymmetric activation in the frontal cortex (measured by the LBT in Study 1 and EEG in Study 2), (b) BIS activity mediates MS effects on cultural closed-mindedness (operationalized as reluctance to explore foreign cultures in Study 1 and ethnocentrism in Studies 2 and 3), and (c) these effects are particularly strong for persons inclined to experience discomfort when faced with unfamiliar and potentially threatening stimuli (measured by NFC in all studies). In addition, Study 3 conceptually replicated the first two studies by showing that directly manipulating inhibition promotes ethnocentric thinking for high NFC persons in a manner similar to right-frontal asymmetry. To our knowledge, these results are the first to demonstrate that MS-based aversion to cultural otherness is driven by a very basic, neurally-instantiated inhibitory process.

# Theoretical Implications

The findings have important implications for understanding the neural and motivational underpinnings of ethnocentric attitudes in response to existential threat. They provide the first direct support for the hypothesis that inhibitory processes, as indicated by relative right-hemispheric activity, underlie MS effects on ethnocentric thinking, which is a key aspect of worldview defense. This mediational finding is consistent with the process model of threat and defense, which states that defensive responses to threat are driven by passive avoidance motivation, for which the BIS is primarily responsible (Jonas et al., 2014). This is also in line with the assumption of TMT that ethnocentric responses to MS constitute "avoidant defenses," which are aimed at avoiding experiencing existential anxiety (Greenberg et al., 2003, p. 519). Moreover, our results are exactly in line with Greenberg et al.'s (2003) speculation that these defenses might be mediated by "a relative increase in right-hemisphere frontal lobe activity" (Greenberg et al., 2003, p. 519). On a general level, our findings may advance and refine TMT research in that they show that MS can not only promote various BIS-related behaviors (Landau and Greenberg, 2006; Routledge et al., 2010; Huang and Wyer, 2015), but also that cultural closed-mindedness can be explained through very basic inhibitory processes.

To be clear, we are not suggesting that all manifestations of worldview defense in response to MS are driven by inhibitory processes based on the current observations. Although, this would follow from the general process model (Jonas et al., 2014), our results can only speak to cultural closed-mindedness. Thus, it largely remains an empirical question to what extent inhibition processes drive other forms of worldview defense given that inhibition has yet to be thoroughly tested as a mediator in this regard. Future research should sample the conceptual space of worldview defense more broadly to examine the generality of the process we present On a related note, it also remains to be seen whether approach-related cognition can alleviate the passive avoidance observed here and prevent negative effects of MS on desirable outcomes, like creativity (Sligte et al., 2013). The findings may also advance research on threat and defense in that they suggest that various responses to threat that are not obviously or intuitively related to a lack of approach motivation may be attributed to the approach-inhibitory activity of the BIS.

These advances also have important implications for inspiring research on the defensive function of BIS anxiety, which originates from ethological research on animals' responses to threat (Gray and McNaughton, 2000). It has been argued that "there are long-standing concerns that rodent models of psychological processes are too simple to apply in humans (Matthews, 2008): For example, there is no evidence that rodents experience anxiety of an abstract type related to existential issues, whereas historical and literary accounts abundantly point to the existence of such angst in humans. Concerns of this type have raised a need for studies of human defense that can test the validity of the defensive explanation for anxiety in humans" (Perkins and Corr, 2014, pp. 42–43). Our findings show that rodent models of BIS anxiety can indeed be applied to defensive behaviors in the wake of threat in humans and point toward the possibility of a more consilient science of motivation.

On the whole, our findings provide valuable insight into contextual, interindividual, motivational, and neural determinants of cultural closed-mindedness. To our knowledge, this research is the first to establish an empirical link between very basic, neurally instantiated inhibitory processes and rather complex, higher-order manifestations of intergroup negativity. By investigating the neural and motivational underpinnings of ethnocentric attitudes in the wake of existential threat, this research might contribute to a fuller understanding of cultural worldview defensive phenomena, among the most investigated outcomes in social psychology over the last 25 years (Burke et al., 2010).

# LIMITATIONS AND FUTURE DIRECTIONS

We used an indirect measure of BIS activity in Studies 1 and 2. Following theoretical and empirical research relating reduced left-frontal asymmetry (or increased right-frontal asymmetry) to BIS-mediated inhibition of approach motivation, we used perceptual and neural indicators of frontal asymmetry to gauge BIS activity (e.g., Davidson et al., 2000; Coan and Allen, 2004; Corr, 2004; Harmon-Jones et al., 2009; Nash et al., 2010). To provide more direct evidence for the role of inhibition in the emergence of ethnocentric thinking, we directly manipulated inhibition in Study 3, finding increased ethnocentrism for high NFC persons following experiences of inhibition and BIS-related affect. In combination, these studies yield support for BIS-driven ethnocentrism in the wake of existential threat. Further, we have other lines of converging evidence for the notion that perceptual and neural indicators of right-frontal asymmetry reflect BIS activity (Agroskin and Jonas, 2015; Agroskin et al., 2015). For instance, we have found in four EEG and LBT studies that directly boosting/blocking BIS anxiety leads to increased/reduced right-frontal asymmetry following MS (and also without MS). Mirroring increased BIS affect following induced inhibition (Study 3), we have also found amplified BIS affect following MS. In summary, accumulating, multi-method evidence supports the notion that threat-induced anxious inhibition (BIS activity) is a cause of ethnocentric thinking.

Future research might consider an intriguing existential interpretation of the avoidance—approach motivational systems. According to TMT, avoidance—approach conflicts may reflect the duality between the core motives for self-protective defense (avoidance motivation) and self-expansive growth (approach motivation; Greenberg et al., 1995a; Pyszczynski et al., 2000, 2012). From this perspective, closed-minded persons engage in novelty-avoidant defense of the "familiar" in response to mortality awareness, whereas open-minded individuals approach the "unfamiliar," expanding the self by gaining new experiences. Future research may investigate to what extent these motivational tendencies mediate MS effects on various behavioral manifestations of the defense and growth motives, such as support of extreme military action against cultural outgroups (defense motive), and commitment to important life goals that convey a sense of existential meaning (growth motive; Greenberg et al., 1995a; Pyszczynski et al., 2012; see Vail et al., 2012 for a review of growth-oriented MS effects).

Another avenue for future research might be to investigate further neural underpinnings of ethnocentric attitudes following MS. Along with the right dorsolateral prefrontal cortex, the BIS may be subserved by the anterior cingulate cortex (ACC; McNaughton and Corr, 2004). In addition to generating neural signals of error-related distress (ERN) and anxiety (see Shackman et al., 2011 for a review), the ACC contributes to risk aversion in decision making (Brown and Braver, 2007). This suggests that the ACC may also support risk-avoidant behaviors after MS, such as the ethnocentric avoidance of cultural outgroups. Future research might shed some more light on previously unexamined neural mediators of terror management mechanisms.

# REFERENCES


# CONCLUSION

Becker (1997) saw man as a "terrified, death-avoiding animal" (p. 99) that, by seeking to avoid death, kills off "so large a spectrum of his action-world that he is actually isolating and diminishing himself and becomes as though dead" (Becker, 1997, p. 181). Psychiatrist Joseph Rheingold arrived at the same conclusion: "The common denominator of all negative ways of dealing with anxiety is a shrinking of the area of awareness and of activity. (...) We are afraid to die, and therefore we are afraid to live, (...) we avoid non-being by avoiding being. The avoidance of anxiety then means a kind of death in life" (Rheingold, 1967, pp. 204–205). These insightful analyses shed light on a fascinating irony inherent in human striving for self-preservation, or in terms of terror management, a life free from consciously experienced death anxiety (Greenberg et al., 2003). In trying to avoid death, people actually avoid life, given that exploring (cultural) novelty contributes to various indicators of well-being, including subjective vitality and satisfaction with life (Kashdan et al., 2004). Our findings represent the first neural evidence, to our knowledge, that this existential ironical twist is driven by an ancient motivational system, oriented toward passively avoiding punishment from the environment. By symbolically avoiding the punishment of death, people in reality obtain the punishment of an unlived life.

# AUTHOR CONTRIBUTIONS

DA and EJ conceptualized the project and collected the data. JK advised the EEG data collection and analysis and assisted manuscript preparation. DA analyzed the data and wrote the manuscript. MP assisted with data analysis and manuscript preparation.

# FUNDING

The research reported in this article has been financed by the Austrian Science Fund (FWF-P27457). The first author of this article was also financially supported by the Doctoral College "Imaging the Mind" of the Austrian Science Fund (FWF-W1233).

responses and facial EMG. Psychophysiology 38, 685–693. doi: 10.1111/1469- 8986.3840685


Badcock, N. A., Mousikou, P., Mahajan, Y., de Lissa, P., Thie, J., and McArthur, G. (2013). Validation of the Emotiv EPOC <sup>R</sup> EEG gaming system for measuring research quality auditory ERPs. PeerJ 1:e38. doi: 10.7717/peerj.38


Becker, E. (1997). The Denial of Death. New York, NY: Free Press Paperbacks.


phobic and compulsive behaviors. Pers. Soc. Psychol. Bull. 33, 1137–1151. doi: 10.1177/0146167207303018


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Agroskin, Jonas, Klackl and Prentice. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Striving for group agency: threat to personal control increases the attractiveness of agentic groups

Janine Stollberg\*, Immo Fritsche and Anna Bäcker

Department of Social Psychology, Institute of Psychology, University of Leipzig, Leipzig, Germany

When their sense of personal control is threatened people try to restore perceived control through the social self. We propose that it is the perceived agency of ingroups that provides the self with a sense of control. In three experiments, we for the first time tested the hypothesis that threat to personal control increases the attractiveness of being part or joining those groups that are perceived as coherent entities engaging in coordinated group goal pursuit (agentic groups) but not of those groups whose agency is perceived to be low. Consistent with this hypothesis we found in Study 1 (N = 93) that threat to personal control increased ingroup identification only with task groups, but not with less agentic types of ingroups that were made salient simultaneously. Furthermore, personal control threat increased a sense of collective control and support within the task group, mediated through task-group identification (indirect effects). Turning to groups people are not (yet) part of, Study 2 (N = 47) showed that personal control threat increased relative attractiveness ratings of small groups as possible future ingroups only when the relative agency of small groups was perceived to be high. Perceived group homogeneity or social power did not moderate the effect. Study 3 (N = 78) replicated the moderating role of perceived group agency for attractiveness ratings of entitative groups, whereas perceived group status did not moderate the effect. These findings extend previous research on group-based control, showing that perceived agency accounts for group-based responses to threatened control.

Keywords: social identity, control motivation, responses to threat, agency, group processes

# Introduction

Group membership is important to people. It helps them to define who they are and to whom they belong, and it serves the satisfaction of basic human needs. The link between group membership and need satisfaction has been delineated for different motives, like self-esteem (Tajfel and Turner, 1979; Rubin and Hewstone, 1998), belongingness (Baumeister and Leary, 1995; Hornsey and Jetten, 2004), self-concept certainty (Jetten et al., 2004; Hogg, 2007), symbolic immortality (Greenberg et al., 1986; Castano et al., 2002), and, more recently, for personal control (Fritsche et al., 2013). Although, the need for control has been counted among the most basic human motives (Fiske, 2003; Pittman and Zeigler, 2007) only recently empirical research has begun to systematically test how group membership may satisfy the need for control (Fritsche et al., 2008, 2011, 2013). People need a general sense of control (White, 1959), they want to experience themselves as autonomous agents who are capable to exert influence on important aspects of their environment. We propose

#### Edited by:

Christina Steindl, University of Salzburg, Austria

#### Reviewed by:

Andreas Mojzisch, University of Hildesheim, Germany Ian McGregor, York University, Canada

#### \*Correspondence:

Janine Stollberg, Department of Social Psychology, Institute of Psychology, University of Leipzig, Neumarkt 9-19, D-04109 Leipzig, Germany janine.stollberg@uni-leipzig.de

#### Specialty section:

This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology

Received: 24 November 2014 Accepted: 04 May 2015 Published: *27 May 2015*

#### Citation:

Stollberg J, Fritsche I and Bäcker A (2015) Striving for group agency: threat to personal control increases the attractiveness of agentic groups. Front. Psychol. 6:649. doi: 10.3389/fpsyg.2015.00649 that a feeling of "having an impact on the world" is not limited to individuals ("I have an impact on"), but can be extended to social ingroups ("We have an impact on"), and that both contribute to the perception of the self as an agent. Accordingly, people can rely on their group membership to perceive themselves as agents when their sense of personal control is threatened (Fritsche et al., 2013). Building on research showing increased identification and group-based cognition following threat to personal control (Fritsche et al., 2008, 2013; Agroskin and Jonas, 2010, 2013), in the present article we aim at investigating what kind of groups work best for restoring a sense of control. Specifically, we tested the novel hypothesis that when a sense of personal control is threatened, people identify more strongly with agentic ingroups and find it more attractive to join agentic than less agentic groups.

People usually belong to many different groups that represent their social self (Tajfel and Turner, 1979). They identify with ingroups that are both accessible and salient in a particular context (Turner et al., 1987), and which may satisfy important self-related motives (Correll and Park, 2005). The need for control has been referred to as one of the most central human motives (White, 1959; Pittman and Zeigler, 2007). Deprivation of control can impair people's performance, well-being and health (Szpiler and Epstein, 1976; Rodin and Langer, 1977; Abramson et al., 1978), which is why people strive to uphold or restore a sense of control when control seems threatened (Rothbaum et al., 1982; Thompson et al., 1993). The model of group-based control (Fritsche et al., 2011, 2013) assumes that defining the self in terms of group membership (social identity; Tajfel and Turner, 1979) and joining in group-based action bolsters or restores people's sense of control through the (social) self when their sense of personal control is threatened. Initial empirical evidence has supported the group-based control approach: Salient lack of personal control increased ingroup bias (Fritsche et al., 2008, 2013), the expression of prejudice (Agroskin and Jonas, 2013; Fritsche et al., 2013; Greenaway et al., 2014a) and conformity to salient ingroup (but not to outgroup) norms (Stollberg et al., Unpublished manuscript). The reported control threat effects were obtained for different ingroups, ranging from classic social categories like nationality, and students, to work groups. However, it is an open question, whether people deprived of personal control support and identify with any ingroup. The model of group-based control suggests that people prefer groups that appear as (collective) agents and that could equip the self with a sense of (collective) control. This novel hypothesis is tested in the present research for both, identification with present ingroups and perceived attractiveness of joining new groups.

What are the properties of groups that give people a sense of collective control? According to Skinner (1996), personal control comprises the idea that the self as an agent can affect end states through instrumental actions (means). In a related vein, Preston and Wegner (2005) proposed that people are motivated to think of themselves as agents ("ideal agency"). According to them, agency is experienced for voluntary actions and can be traced back to three features: an intention to act, the free will to exert the action, and performing the action itself. The concept of agency closely resembles and helps to specify the foundations of control. Thus, drawing on both Skinner, and Preston and Wegner, we define a sense of control as the perceived potential to affect important aspects of the environment through the autonomous self. That is, perceptions of agency are essential for perceiving control, and striving for perceptions of the self as agentic should be a means to (re-) establish a perception of control.

Building on social identity principles (Tajfel and Turner, 1979), we propose that perceptions of agency are not limited to the personal level of the self, but can also describe the social level of the self. That is, in situations where the personal self is salient, people infer control from their sense of personal agency, whereas when a social self is salient, people's sense of control should be determined by perceptions of collective agency. In cases when personal agency and control seem thwarted, people may thus be motivated to define the self in terms of an agentic ingroup or to become a member of an agentic group. Applying the components of personal agency specified by Preston and Wegner (2005) to a collective agent (i.e., a social group) means that, for instance, a group of colleagues voluntarily cleaning up a messy meeting room would be perceived as agentic, because they are actually cleaning (perform action), they have planned to do so (collective intention), because they were annoyed by all the mess (free will). If people are motivated to reestablish a sense of control and to perceive themselves as autonomous collective agents, they should prefer groups that best represent all features of an ideal agent. First evidence that groups differ with regard to how well they satisfy agency needs was reported in research on lay people's group typologies (Lickel et al., 2000), showing that only task groups were associated with agency (Johnson et al., 2006). In a similar vein, Deaux et al. (1995) found that vocational identities were seen as more agentic than other identities.

Entitativity—the perception of a group as a real entity (Campbell, 1958; Yzerbyt et al., 2000; Brewer et al., 2004) should be a necessary precondition to ascribe perceptions of agency to a group. Entitativity was linked to various group properties, such as boundedness, common fate, similarity of group behavior, homogeneity of group members, and sharedness of goals (Yzerbyt et al., 2000; Brewer et al., 2004). However, in most empirical studies, entitativity was understood only in terms of group homogeneity (Yzerbyt et al., 2000; Castano et al., 2003; Hogg et al., 2007) or within-group similarities (Lickel et al., 2000; Johnson et al., 2006; Crawford and Salaman, 2012). It was shown that identification with homogenous and bounded groups increased following self-uncertainty (Hogg et al., 2007), supporting the notion that groups with clearly defined prototypes satisfy a need for self-certainty. However, entitativity (i.e., the perception of a group as coherent entity through a certain level of homogeneity and boundedness among group members) should not be sufficient for satisfying the need for control, although it might be a necessary condition for group-based control to emerge. First, individuals require a certain level of similarity and boundedness to be perceived as a group at all. Second, basic agreement on collective goals and representations of the environment seems necessary to engage in collective behavior (which is not true for unspecific homogeneity that is not related to group goals). Nevertheless, whether or not the group is pursuing goals and has formed them autonomously is independent of perceived consensus and thus entitativity, but it is basically a question of collective agency. Although, some authors included agency in the concept of entitativity (Brewer et al., 2004), for the sake of clarity, here we treat both concepts separately, proposing that entitativity is necessary but not sufficient for group-based control to emerge. We hypothesize that threat to control increases identification with ingroups and increases the attractiveness of new groups, when these groups are highly entitative and at the same time highly agentic. Further, we control for other group features that may be related to collective agency, such as group size, group power, or status to flesh out the unique effect of collective agency for increasing group identification and group selection under threat.

To test our hypothesis that threat to control increases the attractiveness of agentic groups we conducted three studies. In Study 1, we investigated the effect of a control salience manipulation on identification with different types of ingroups, intimacy groups, task groups, social categories and loose associations. We hypothesized that control threat increases identification only for those types of groups that are characterized by both entitativity and agency (i.e., task groups; Lickel et al., 2000). Moreover, we investigated whether increased group identification following threat indirectly strengthened perceptions of intragroup support and collective control. According to the model of group-based control (Fritsche et al., 2011, 2013), when people respond to threatened control through identification with agentic ingroups they should perceive an increase in perceived control through the (social) self, that is, on the group level. Perceptions of heightened collective efficacy and perceived support within the ingroup should thus be indicative of control experienced on a group level. In line with this assumption Drury and Reicher (2009) found that identification with a social movement increased group members' perceptions of empowerment and collective efficacy, and therefore maintained the belief that the ingroup can attain its goal collectively.

In Studies 2 and 3, we studied the attractiveness of groups that were not (yet) a part of participants' identity. In particular, in Study 2 we were interested in the factors that determine the attractiveness of small vs. large groups. Small groups have been associated with better group performance due to less procedural losses on coordination (Kravitz and Martin, 1986), and motivation, like social loafing (Latané, 1981), which might affect the efficacy of group goal pursuit and increases the perception of small groups as agentic and efficacious, as it has been found in social dilemma studies (Kerr, 1989). However, there is contrary evidence as well: participants, who were asked to interact with others as a representative of a majority group reported a higher sense of personal control than those who thought to represent a minority group (Guinote et al., 2006). Finally, the relation between size and group agency perceptions was questioned in a multilevel-analysis by Watson et al. (2001), who investigated the impact of several individual and group factors on collective efficacy beliefs and who did not find a relation between group size and collective efficacy beliefs of group members. The ambiguity of these results suggests that whether small or large groups are perceived as more agentic may vary from situation to situation. We hypothesized that following threat to personal control participants should be more strongly attracted to groups of that size they perceive as relatively more agentic. Other perceived group characteristics that do not represent full-blown agency, such as social power or unspecific homogeneity, are not expected to moderate the effects of personal control threat on relative group attractiveness.

In Study 3, we measured the three components of perceived ideal group agency for entitative and non-entitative groups, hypothesizing that control threat affects group attractiveness only when the group is entitative and highly agentic. Following a control salience manipulation, we assessed the attractiveness of entitative and non-entitative realistic groups that were displayed on pictures and measured perceived agency and perceived group status as possible moderators.

# Study 1

We conducted Study 1 to examine the effect of threatened control on identification with ingroups of different types. As we argued above, not all groups should restore a sense of control to the same degree. Those high in both, entitativity and agency should be preferred when people have the opportunity to choose between ingroups of different types. A taxonomy of group types published by Lickel et al. (2001) reflects the intuitive typing of groups by lay people. They differentiate between task groups (e.g., work teams), intimacy groups (e.g., families), social categories (e.g., nations) and loose associations (e.g., people waiting together at the bus stop), showing that these four group types provide different benefits for people: participants primed with adjectives linked to a need for affiliation (e.g., connectedness, belonging) listed significantly more intimacy groups, whereas for achievement related adjectives (e.g., success, competence) more task groups were selected, and for self-esteem related adjectives (e.g., identity, distinctiveness) more social categories were chosen (Johnson et al., 2006; Crawford and Salaman, 2012). None of the studies investigating need fulfillment by these four different group types has explicitly investigated a need for control as a driver of ingroup identification. However, it seems obvious that task groups are those that are most intimately associated with the notion of agency and therefore collective control as, unlike the other groups, it is their most primary purpose to act. Therefore, we expect an increase in identification following threat to control only for task groups, because these groups are primarily perceived as agentic groups.

In line with the notion of group-based control, increased identification with a task ingroup following threat to control should further result in increased perceptions of collective efficacy and within-group support. Collective efficacy directly expresses a sense of collective control and agency whereas within-group support facilitates effective group coordination and goal pursuit and may thus indicate collective control in an indirect fashion. Recent findings initially support this assumption, by showing that group identification increased people's perceptions of control, which in turn enhanced personal well-being (Greenaway et al., 2015). Thus, we tested for indirect effects of control threat on collective efficacy and within-group support, mediated via identification with task groups.

# Method Participants and Design

One hundred and one university students participated in the study. We excluded three participants who had guessed the aim of the study, and five who had participated in a similar experiment previously. Thus, the final sample consisted of 93 participants, 66 were female and 26 were male, with a mean age of 21.41 years (SD = 2.22), one person did not indicate age or sex. The experiment had a 2 Control Salience (high/low) × 4 Group Identification (task group/intimacy group/social category/loose association) design with repeated measurement on the last factor.

# Procedure

Participants were recruited at the campus of a German university. After they had agreed upon participation, they received a questionnaire, which introduced the study as a survey on personality traits. Then, participants were exposed to a control salience manipulation, similar to a manipulation that has been used previously in control threat research (Whitson and Galinsky, 2008). In the low control salience condition, participants read the following instruction (instructions for the high control salience condition in parentheses): Please think about an important situation in your life, in which you had no (full) control over the things going on. Please, try to remember exactly and imagine the event vividly! How did you feel right now? Now, describe the situation and your thoughts about it in the following lines. Please, take as much time as necessary!

The control manipulation was followed by a German version of the PANAS (Watson et al., 1988) and a questionnaire on sleep- and awakening patterns, which served as a delay task. We included a delay task, because different kinds of threat have been shown to produce effects only after a short delay (Wichman et al., 2008; Burke et al., 2010; Fritsche et al., 2012). Afterwards, participants received descriptions of the four group types (task group, intimacy group, social category, loose association) according to Lickel et al. (2000). For each group type they were asked to identify an example group to which they belonged. Then, identification with the group, collective efficacy, and within-group support were assessed. The order in which the group types were presented was counterbalanced, resulting in four different versions of the questionnaire.

# **Identification**

Participants rated their identification with each group on five items on a 4-point-scale (1 = do not agree to 4 = agree). Four were adopted from Henry et al. (1999): "I think of this group as part of who I am.", "I see myself as quite similar to other members of the group.", "I enjoy interacting with the members of this group.", "Members of this group like one another.", another item "I am happy to be a member of this group." was added, α(task group) = 0.83, α(intimacy group) = 0.53, α(social category) = 0.71, and α(loose association) = 0.86<sup>1</sup> .

# **Collective efficacy**

Then, participants completed four items (1 = do not agree to 4 = agree), assessing collective efficacy beliefs for each group: "Together we are strong.", "We can achieve things collectively, one cannot achieve individually.", "Nobody should think you cannot count on us.", "Together we even come through hard times.", α(task group) = 0.82, α(intimacy group) = 0.61, α(social category) = 0.79, and α(loose association) = 0.87.

# **Within-group support**

Perceived support among group members was assessed with three items (1 = do not agree to 4 = agree) for each group, adapted from Zimet et al. (1988): "There is always a member of the group around when I am in need.", "I get the emotional support and help I need from my group.", "I can count on my group when things go wrong."α(task group) = 0.80, α(intimacy group) = 0.77, α(social category) = 0.87, and α(loose association) = 0.84. After completing the experiment, participants were debriefed and thanked for participation and received a chocolate bar.

# Results

We expected low control salience to increase identification with the task group, but not with other group types. Therefore, we conducted a 2 Control Salience (low/high) × Order of Group Types × 4 Group Identification (task group/intimacy group/social category/loose association) analysis of variance, with repeated measurement on the last factor (for cell values see **Table 1**). The general level of group identification differed marginally for the order of group type presentation, F(3, 78) = 2.62, p = 0.06, η <sup>2</sup> = 0.09. However, as it did not interact with control salience, we report the descriptive values of the analysis regardless of presentation order. The results showed different levels of identification, depending on group type, F(3, 78) = 334.29, p < 0.001, η <sup>2</sup> = 0.93. Participants reported the highest level of identification with intimacy groups, followed by social categories and task groups, with the lowest levels of identification with loose associations. More importantly, a significant interaction of control salience and group type suggests that control salience affected participants' identification with ingroups differently, depending on group type, F(3, 78) = 3.18, p = 0.03, η <sup>2</sup> = 0.11. In line with predictions, people with low perceptions of control reported higher levels of identification with the task group than people with high perceptions of control, F(1, 80) = 6.12, p = 0.02, η <sup>2</sup> = 0.07. There were no simple effects of control salience on identification with intimacy groups, social categories or loose associations, all p's > 0.12.

#### Mediational Analyses

To test whether increased identification with a task group mediates the effect of control salience on collective efficacy beliefs and perceived within-group support from the task group, we conducted two separate simple mediation analyses (see **Figure 1**), using the macro process for SPSS (Hayes, 2013). Control salience was contrast coded for both analyses (low control = −1 vs. high control = 1). Participants in the low control salient condition showed more identification with their task group as participants

<sup>1</sup>Participants rated each group on warmth (warm, good-natured) and competence (competent, intelligent). Ratings were not affected by control salience manipulation, all F 's < 1. As we did not expect any difference on these dimensions, we did not consider it further.



in the high control salient condition (a = −0.15, p = 0.02), as identification increased in participants, collective efficacy beliefs increased too (b = 0.56, p < 0.001). A bias-corrected 95% confidence interval based on 2000 bootstrap samples for the indirect effect (ab = −0.09) was entirely below zero, CI [−0.18, −0.02], indicating an indirect effect. No evidence was found for a direct effect of control salience on collective efficacy beliefs (c ′ = 0.05, p = 0.44).

Simple mediation analysis for within-group support revealed that low control salience increased participants identification with their task group as compared to high control salience (a = −0.15, p = 0.02), as identification increased, perceived group support increased too (b = 0.79, p < 0.001). Again, a bias-corrected 95% confidence interval (2000 bootstrap samples) for the indirect effect (ab = −0.12) was entirely below zero, CI [−0.23, −0.01]. No direct effect was found, (c' = 0.09, p = 0.15). Both reported mediations support our prediction that control threat indirectly increased collective efficacy beliefs, as well as perceived within-group support, via increased task group identification. As the control salience manipulation did not affect identification with another group type, we tested, whether there was a direct effect of control salience on collective efficacy beliefs and within-group support for the other group types. No direct effect could be observed for any of the two dependent measures, all p's > 0.21.

#### Discussion

The results show that threat to personal control increased people's reports of identification with self-representative task groups, but not with other types of ingroups. This means that people under threat either identified more with task groups or more often choose task groups (but not other types of groups) they strongly identified with. Both possible processes express their motivation to uniquely associate the self with an ingroup, whose primary purpose is the active pursuit of a shared goal, as well as the perception to achieve the goal through joint effort, and therefore agency (Spencer-Rodgers et al., 2007). Identification with other groups that were also high on entitativity (Lickel et al., 2001), but rather low on agency, such as intimacy groups, was not affected by control threat. This supports our contention that only membership in entitative groups that at the same time is characterized by agency can restore feelings of perceived control through the collective self. Entitativity per se is not sufficient to increase the self-importance of a group following threat to control when more than one group is salient to people. While people identified highest with intimacy groups, which is in line with previous findings (Castano et al., 2003), selfimportance of intimacy groups did not increase for people low in control, suggesting a specific mechanism of group-based control restoration. Nevertheless, previous studies also found increased support of other ingroups, such as nations, following threat to control (Fritsche et al., 2013). This could be explained by the fact that in all of these studies only one ingroup was salient to participants and thus only one option for experiencing collective control, whereas in the current study participants could choose that ingroup that appeared to be most appropriate for reestablishing a sense of control. Although in general, groups should heuristically be perceived as homogeneous agents (Brewer et al., 2004), the present study made salient the differences between the group types. Obviously, as the current results show, some ingroups are better suited to restore control perceptions than others, probably because they can provide a greater sense of agency.

The control restorative function of task ingroups is further supported by mediational evidence. The results show, as their sense of personal control decreased, people increased identification with a task group they belonged to, which in turn enhanced perceptions of collective efficacy for the task group. The same was true for perceived support among group members. Although, support by others may indicate less control through the self in interpersonal contexts, salient common group membership is likely to transform interpersonal support into an expression of the agentic "We," and thus represents an indirect indicator of collective control. Hence, we conclude that people increase perceptions of collective control through identification with agentic ingroups, when control perceptions for the personal self are depleted. These findings illustrate that ingroup identification can in fact restore perceptions of control and may thus add to research examining beneficial effects of social identity for human well-being (social cure effects; Jetten et al., 2014).

However, agency is not the only characteristic of task groups and thus the effects of salient threat to personal control may be attributed to processes other than group-based control. To solve this issue, we conducted Studies 2 and 3, where we directly tested whether perceived group agency is necessary for increased group attractiveness following threatened personal control<sup>2</sup> . In Studies 2 and 3, we also intended to contrast the proposed moderating effect of perceived collective agency to the effects of other potential moderators, thereby turning from ingroup identification to the attractiveness of groups people do not (yet) belong. Here, we assume that potential new members become more likely to join the group when they perceive features of the group satisfying their personal needs (Moreland and Levine, 1982). Therefore, threat to personal control should increase the attractiveness of groups that are perceived as being agentic.

# Study 2

The results of Study 1 are a first indication that agentic ingroups are especially attractive for people who lack personal control. In Study 2 we extend this hypothesis to potential ingroups of which people are not part of yet. Furthermore, to test whether the effect of perceived collective agency is unique, possible effects of other factors, such as group size, perceived group power or perceived unspecific group homogeneity should be teased apart from agency perceptions.

In Study 2, we investigated the impact of control threat on attractiveness of groups differing in size. Small groups are better able to act in a coordinated manner, whereas in cases where large groups solve their coordination problems, as majority groups, they will have more social power to attain resources and could thus elicit effects on the environment. As thus, small and large groups' agency may be in the eye of the beholder, we decided to use relative measures, assessing participants' idiosyncratic perceptions of how agentic large groups are in comparison to small groups. In addition, we measured participants' relative perceptions of large vs. small groups on the possible group attribute moderators power and homogeneity, as well as group attractiveness as dependent variable. We expected participants to respond to threatened personal control by showing a relative preference for groups of that size they perceive as relatively more agentic. Although both group power and homogeneity could be related to perceptions of collective agency, they are each insufficient to cover all aspects of agency. Power can refer to personal influence, but also to the possession of resources and high status. Only the former perception shows a conceptual association with the outcome aspect of agency but less so with its process (shared intention and coordinated collective action). In a similar vein, homogeneity may prepare the ground for building shared intentions but is not implying that coordinated action or visible outcomes do occur. That is why we only expected full-blown agency to moderate control threat effects on group attractiveness.

# Method

#### Participants and Design

Fifty university students participated in the study. We excluded one participant who had guessed the aim of the study, and two participants who participated in a similar experiment previously, resulting in a final sample of 47 participants. Twenty participants were female, and 27 were male, with a mean age of M = 27.81 years (SD = 7.95). A manipulation of control salience (high/low) served as independent variable. For both, dependent variable and moderators, we used measures that assessed participants' relative perceptions of large vs. small groups. We tested relative group agency, power, and group member homogeneity of large vs. small groups as a function of control threat on relative group attractiveness ratings.

# Procedure

Participants were asked at the campus of a German university to take part in a study on attitudes, biorhythm and personality. After they had agreed upon participation, they received a questionnaire, which opened with a control salience manipulation similar to those used in Study 1. In the low control salience condition (high control salience in parentheses) they read: Take a moment to think about situations or incidents, in which you realized that you have very little (very much) control and impact on important things in your life. Please describe briefly in your own words one event or situation that made you feel helpless (influential). How did you feel in that situation? After answering the two questions, participants were asked to indicate on a 7-point-scale (1 = not at all to 7 = very much), how much they felt in control over important aspects in their life in the situation they had just described, which served as a manipulation check. This was followed by a questionnaire on sleep- and awakening patterns, which served as a delay task, as in Study 1.

### **Group attractiveness**

Then, participants made attractiveness rating for six pairs of groups, representing environmental organizations, political parties, companies, aid agencies, and cliques, that were each briefly described. Within one pair, the group descriptions differed only in group size (one small group, one large group). Participants were instructed to imagine that both groups would equally correspond to their attitudes and beliefs and that they should make a decision, which group they would rather like to join. To avoid a forced choice task, participants made ratings for each group of the pair with regard to the likelihood that they would join the group on a 7-point-scales (1 = very unlikely to 7 = very likely). To preserve the comparative nature of the judgments, we computed a difference score for the dependent variable of

<sup>2</sup> In the present study, however, we did not expect a moderation of the control threat effect by the measure of task group efficacy on identification. This is because we explicitly contrasted task groups with group types of lower agency that should have led to decreased variance on people's perceptions of task groups' agency (contrasted to the other groups, task groups are clearly agentic). Variance on perceived task group efficacy should have been primarily caused by the indirect effect of threat through task group identification instead, which appeared to represent a process of group-based control restoration.

Stollberg et al. Control threat and group attractiveness

group attractiveness ratings, subtracting ratings of small groups from ratings of large groups, resulting in a variable that reflected the relative attractiveness of large compared to small groups.

#### **Group attributes**

Group attributes were assessed as relative measures, too. After group attractiveness ratings, participants were asked to indicate, which attributes large groups possess as compared to small groups: "Large groups are rather. . . " (1 = powerless to 7 = powerful); "Members of large groups are rather. . . " (1 = dissimilar to 7 = similar); "Large groups are rather. . . " (1 = non-agentic to 7 = agentic). After finishing the experiment, participants were thanked, fully debriefed and received a chocolate bar for their participation.

### Results

Participants in the low control salience condition indicated having perceived less control over important aspects in their life (M = 2.19, SD = 1.33) than participants in the high control salience condition (M = 5.40, SD = 1.24), t(34) = −7.34, p < 0.001. Thus, the manipulation of control salience was successful.

We tested whether control threat would increase relative preference for the group size that participants saw as relatively more agentic. For participants who saw large groups as more agentic than small groups, we predicted that the control threat would increase their relative preference for large groups. For participants who saw small groups as more agentic than large groups, we predicted the reverse, that control threat would increase their relative preference for small groups. Thus, we predicted a crossover interaction. We tested for moderation by perceived relative agency of large groups, using the process macro for SPSS by Hayes (2013). We regressed relative attractiveness of large vs. small groups on control salience (contrast coded with: high = 1, low = −1), perceived relative agency, and the Control Salience × Agency interaction. Agency and the interaction term were mean centered. The results showed that perceived agency moderated the effect of control salience on perceived relative attractiveness, indicated by a significant Control Salience × Agency interaction, b = −0.20, t(44) = −1.81, p = 0.08 (see **Figure 2**). As simple slope analysis showed, when large groups were perceived as less agentic than small groups (−1 SD), the relative attractiveness of small groups increased under low control salience compared to high control salience, b = 0.45, t(44) = 1.89, p = 0.07. No effect of control salience on relative attractiveness could be observed, when large groups were perceived as more agentic than small groups (+1 SD), b = −0.16, t(44) = −0.68, p = 0.50. Looked at differently, participants in the low control salience condition perceived large groups as relatively more attractive than small groups, when large groups were perceived as more agentic (+1 SD), than when large compared to small groups were perceived as rather non-agentic (−1 SD), b = 0.55, t(44) = 3.60, p < 0.001. In the high control salience condition, no effect occurred, b = 0.15, t(44) = 0.92, p = 0.36.

We conducted another regression analysis, including all possible moderators (agency, power, and homogeneity), as well as the interaction terms of Control Salience × Agency, Control Salience × Homogeneity, and Control Salience × Power, to test

for all moderators simultaneously. No moderation was found for power or homogeneity, p's > 0.38.The initial interaction effect of Control Salience × Agency and the pattern of results remained the same.

#### Discussion

The results of Study 2 confirm our hypothesis that agency, and not perceived power or group member homogeneity, moderates the effect of control salience on perceived attractiveness of potential ingroups. Threat to control increased the relative attractiveness of small groups when these were perceived as more agentic than large groups. For people with low perceptions of control, attractiveness of small compared to large groups was higher, when they perceived small groups to be relatively more agentic than large groups. However, control threat did not increase preference for large groups, when large groups were perceived as relatively more agentic than small groups. Perceived group power and member homogeneity did not moderate the control salience effect on group attractiveness. These findings emphasize the importance of agency as the crucial group feature that defines whether groups can serve as a resource for personal control.

Moreover, the results of the study provide first evidence that control threat can affect the choice of group membership. This could offer new insights, why some groups become attractive for people with low control perception and others do not. Although, perceived group power did not elevate control threat effects on group attractiveness, it is possible that groups high in power offer more opportunities to their members to perceive control than their low power counterparts. Power has been defined in terms of dependence, as an asymmetrical control over resources (e.g., Dépret and Fiske, 1999), but it has been also defined in terms of agency, as exerting control over the environment (van Dijke and Poppe, 2006). In intergroup research, collective action, the coordinated, voluntary action on behalf of the group, has been referred to as the basis for group power (Turner, 2005). This distinction between resource control and agency definitions of power may help to understand why perceived power is not related to control perceptions, as it seemed to be in the present study, and sometimes it is related to perceptions of control (Guinote et al., 2006). When groups are perceived as powerful due to agency reasons, they should have the potential to provide members with a sense of control. However, power due to disposability of resources does not necessarily imply that the group has acquired the resources through collective effort (high status groups often obtain resources because they are entitled but not because they have invested high effort) or that they apply these resources for common goal pursuit. Therefore, controlling resources does not necessarily express collective agency but may sometimes even indicate passivity ("being served"). This would support the finding from interpersonal power research that people often strive for personal power, which reflects the desire to increase one's sense of agency, but less so for social power, which reflects the desire to have control over others (van Dijke and Poppe, 2006).

# Study 3

In Study 2 perceived agency determined the effect of control threat on attractiveness ratings, however, as we used a global measure of agency, we were not able to assess whether all components of agency were present. It is possible that people indeed infer agency, but inference should be stronger when more components are present (Preston and Wegner, 2005). In Study 3 we thus used a more elaborated measure of collective agency perceptions, covering collective intention (cf. shared goals) and voluntary and active collective goal pursuit that represent important indicators of agency (Preston and Wegner, 2005). By employing realistic pictures of entitative and nonentitative groups we thought to use a more subtle procedure to present group features to participants. Groups were preselected as entitative or non-entitative, that means, the degree to which they were perceived to form a coherent entity ("groupness"). This allowed us to test directly our assumption that entitativity is necessary for perceptions of collective control, but that it is not sufficient to explain control threat effects on group attractiveness ratings. We expected personal control threat to increase attractiveness ratings of entitative and agentic groups, whereas attractiveness of groups that lack either entitativity or agency should not be affected by control threat.

Moreover, we tested whether the proposed effect of personal control threat on the evaluation of agentic groups is independent of the social status participants ascribe to the groups. This enabled us to test whether the effect is specific to processes of group-based control, which would imply a moderation by perceived agency but no parallel moderation by status. Alternatively, a moderation by status that cancels out the moderation by perceived agency would suggest a self-esteem account of the personal control salience effect, assuming that people strive for collective status in order to satisfy self-esteem needs (Tajfel and Turner, 1979; Rubin and Hewstone, 1998).

#### Method

#### Participants and Design

Eighty university students participated in the study, two participants were excluded from the sample, one guessed the aim of the study, and one participated in a similar experiment previously. Thus, the final sample consisted of 78 participants, 43 were female and 35 were male, with a mean age of 23.44 years (SD = 4.50). A manipulation of personal control salience (high/low) served as independent variable.

#### Procedure

Participants recruited at the university campus, were asked to take part in a study on personality traits and group perceptions. After participants had agreed upon participation, they received the questionnaire first containing a control salience manipulation, that has been previously used in group-based control research (Fritsche et al., 2013). Participants in the low control salience condition (high control salience in parentheses) read: Take some moments to think of those aspects of your life, that give you a sense of helplessness and lacking impact (power and impact) on the important things of your life. Please, briefly jot down in your own words those three aspects of your life that make you feel most helpless (powerful)<sup>3</sup> . As in Studies 1 and 2, this was followed by a delay task, the German version of the PANAS. Then, participants were presented with 12 pictures of an aggregate of people, six pictures depicted people forming an entitative group, whereas six other pictures depicted people in a similar context who did not form an entitative group (see Supplementary Material). The pictures were selected from a Pre-Study (N = 40), where we asked participants to estimate the degree to which the people displayed in the pictures constituted a real entity. The people in the entitative group pictures that we used in this study were perceived significantly more as a coherent entity than the people in the non-entitative pictures, F(1, 39) = 179.30, p < 0.001. We told participants to imagine that they would like to start a similar group as it was depicted in the picture with people they like. We did so to ensure that participants perceived all groups as potential future ingroups. The pictures differed with regard to content, representing typical groups of the daily life. They displayed people in a seminar room, people in the streets of a city, people painting an artwork, people in an office room, people in a park, and people fishing. Then participants should rate each of the depicted groups on attractiveness of the group for the self, perceived agency, likeability of depicted group members, and perceived status.

#### **Group attractiveness**

Participants rated the attractiveness of the group for the self on four items: "I find the group attractive," "I can nicely imagine myself being a part of a similar group," "I think, the group members feel comfortable with their group," "I would found a similar group myself." Ratings were made on a 7-point-scale (1 = not at all to 7 = absolutely). Internal consistency of attractiveness ratings for each group ranged from α = 0.75 to α = 0.92. Group ratings were averaged over six pictures each, to build two composite score of general group attractiveness, one for entitative

<sup>3</sup>Please note, that the term "power" is used here as a synonym for "effectance" or "control" as it explicitly refers to objects of control that are not necessarily social. In German language the use of the word "power" [Macht] is not limited to social subjects, but can be used for non-social objects either. We used it in the present manipulation as in German everyday language the term "control" is not clearly associated with effectance.

and one for non-entitative groups, which served as dependent variables.

#### **Group agency**

We created a five item measure of perceived agency to assess the three components of agency as we initially defined it: Sharedness of a common goal, voluntary group coordination to achieve the goal, and active pursuit of the goal. Items were: "The people in the group have a common goal that they are able to attain," "I think, it is likely that the group will reach their common goal," "The group is pursuing its goal collectively," "The people in the group are actively working together," "The group acts rather passively."(reverse coded). Ratings were made on a 7-point-scale (1 = not at all to 7 = absolutely), internal consistency was good for all groups, ranging from α = 0.80 to α = 0.90, except for one group depicting people in a seminar room, with an internal consistency of α = 0.65. Again, ratings were averaged over six groups each, resulting in perceived agency of entitative groups and perceived agency of non-entitative groups.

# **Group status**

Group status was measured with one item, on a 7-point-scale (1 = not at all to 7 = absolutely): "The group gives the impression of being held in high esteem.", and averaged for entitative groups, α = 0.73, and non-entitative groups, α = 0.65.

## **Personal likeability**

Further, we assessed perceived personal likeability of the individuals depicted in the pictures with one item ("The people in the group look sympathetic"). We intended to make sure that the depicted individuals did not a priori differ on personal likeability. As likeability ratings for people in entitative and nonentitative groups pictures did not differ for participants with low as compared to high perceptions of control, p = 0.32, we did not consider likeability in any further analysis. After finishing the experiment, participants were thanked, fully debriefed and received a chocolate bar for their participation.

## Results

We expected low control salience to increase the attractiveness of entitative groups that at the same time are perceived as highly agentic. Thus, we conducted a moderation analysis, using the process macro for SPSS (Hayes, 2013). As control threat may affect attractiveness of entitative groups, depending on perceived group status, we tested for both moderations simultaneously (Model 2). Attractiveness of entitative groups was regressed on control salience (low = −1, high = +1), perceived agency of entitative groups, perceived status of entitative groups, Control Salience × Agency, and Control Salience × Status. Agency and Status, as well as both interaction terms were mean centered prior to the analysis. The results showed the predicted moderation for perceived agency, b = −0.23, t(76) = −2.47, p = 0.02, and as expected no moderation for perceived status, b = 0.07, t(76) = 1.08, p = 0.29. Simple slope analysis revealed that participants with low perceptions of control find entitative groups more attractive than participants with high perceptions of control, when they perceived the groups as very agentic (+1 SD), b = −0.28, t(76) = −3.13, p = 0.003. When entitative groups were perceived as less agentic (−1 SD), attractiveness ratings were not affected by control perceptions, b = 0.05, t(76) = 0.51, p = 0.62 (see **Figure 3**). Looked at differently, when low control was salient, very agentic groups were more attractive than less agentic groups, b = 0.59, t(76) = 4.67, p < 0.001, whereas agency perceptions were not related to attractiveness ratings when high control was salient, b = 0.19, t(76) = 1.45, p = 0.15. Thus, the results fully supported our hypothesis.

We conducted the same moderation analysis for nonentitative groups. Attractiveness of non-entitative groups was neither affected directly by control salience, nor did perceived agency or perceived status moderate the effect, all p's > 0.41.

#### Discussion

The results of Study 3 further support our hypothesis that threat to control increases the attractiveness of groups that are perceived as both coherent entities and as highly agentic. Low control salience increased attractiveness ratings of entitative groups, when these groups were perceived as highly agentic. Of interest, perceived status did not moderate control threat effects on attractiveness ratings. Although, it should be acknowledged that, for economical reasons, we assessed group status with a single item measure, the results support our contention that perceived collective agency but not collective status accounts for the effects of personal control threat on group-based cognition and action. This enables us to distinguish processes of group-based control from processes of self-esteem maintenance, which should both represent central functions fulfilled by group membership.

# General Discussion

Across three studies, we found converging evidence that threat to control increases the attractiveness of groups that are perceived as agentic, indicating that people try to restore and maintain their sense of control on the social level of the self when control is threatened for the personal self. Study 1 shows that threat to personal control increased identification only with task groups, but not with intimacy groups, social categories

(Study 3).

or loose associations. This supports our assumption that some groups are better suited to fulfill a need for control than others: when multiple ingroups are salient in a situation, people respond to control threat by increasing identification only with those groups that are both highly entitative and agentic (i.e., task groups). Entitativity in terms of "groupness" seems to be a necessary but not sufficient condition for group-based control as intimacy groups that are usually perceived as high in entitativity (Lickel et al., 2000) were not affected by control threat. Study 2 yielded more specific evidence that agency is the crucial group feature making groups attractive for group-based control restoration. Salient threat to personal control increased the relative attractiveness of small vs. large groups only when small groups were perceived as relatively more agentic than large groups. Other group features, such as perceived group power and perceived ingroup homogeneity did not moderate the effect of threatened control on group attractiveness ratings. This provides specific evidence for the moderating role of agency perceptions. Although, homogeneity and power may relate to specific components of agency perceptions they cannot be equated with agency. While similarity among group members may facilitate the generation of a shared group goal that increases agency perceptions, it does not imply joint goal pursuit and action. In a similar vein, group power usually allows for more opportunities and access to resources that could be used for effective goal pursuit, but powerful groups do not necessarily engage in goal achievement, because they might be satisfied with their current situation and behave rather passively. In Study 3, we conceptually replicated that perceived group agency moderates control threat effects on group attractiveness. Threat to control increased attractiveness ratings of entitative (but not of nonentitative) groups, when these were perceived as highly agentic, that is, when the groups were perceived as collectively and actively pursuing a common goal. Moreover, perceived group status did not moderate control threat effects on attractiveness ratings. The mere perception of high status or majority status (which is often used as a proxy for a high status group), is not sufficient to attract people with a deprived sense of control.

The present findings support the novel hypothesis derived from the model of group-based control (Fritsche et al., 2011, 2013), that people seek out for membership in and identification with agentic groups to restore a sense of control. Group membership satisfies different kinds of needs, but it is perceived collective agency that accounts for group-based control. Other group characteristics like homogeneity, power or group status did not affect group attractiveness ratings for control deprived people. Therefore, our findings extend previous research showing correlational evidence for differential need satisfaction by different groups (Johnson et al., 2006; Crawford and Salaman, 2012). They are also in line with research that considers the ingroup as a social resource (Correll and Park, 2005) for satisfying different needs, such as needs for certainty (Hogg, 2007), selfesteem (Rubin and Hewstone, 1998), and control (Fritsche et al., 2008). While, for instance, ingroup homogeneity can serve best a need for self-certainty (Hogg, 2007) or distinctiveness (Brewer, 1991; Jetten et al., 2004), perceived agency is crucial for control restoration. This adds to research showing that group features like agency or homogeneity are empirically related but functionally different constructs (Spencer-Rodgers et al., 2007; Crump et al., 2010). Whereas, entitativity understood in terms of groupness and homogeneity moderates the effect of a self-uncertainty threat on group identification (Hogg et al., 2007), our findings show that although entitativity seems necessary for group-based control, control deprived people only increase identification with those groups that are additionally perceived as agentic. Further research could investigate this more directly by manipulating group entitativity and group agency as independent between-subjects factors.

Although, the present findings support the notion that different threats elicit different reactions, they could also be understood in terms of a general threat and defense model (Jonas et al., 2014). This model assumes that threat causes a motivational discrepancy that could be resolved by approach-oriented reactions on the personal or social level. Increased attraction to ingroups following control threat represents such a distal defense mechanism on the social level. The present results showing identification with and attraction to agentic groups following threat support the notion that switching from behavioral inhibition to behavioral activation describes the threat defense process (Greenaway et al., 2014b). Future research should clarify the conditions under which people get to behavioral activation and regain a sense of agency and control either by engaging in personal or in social responses. The present research on groupbased control indicates that threat influences social interactions on the group level as it determines people's sense of whom they belong and which groups they seek to join or to found.

# People Increase Perceptions of Collective Control Following Personal Threat Through Identification with Agentic Ingroups

Further evidence for the mechanism of group-based control comes from mediational analyses of Study 1. These findings are the first to show indirect effects of personal control threat on collective control through ingroup identification. This supports the assumption of group-based control that people can restore feelings of control through group membership. Although, previous research has demonstrated increased collective reactions to control threat (Fritsche et al., 2013), evidence that collective reactions in turn increase a sense of collective control has not been shown yet. Further, mediational evidence points to the specific benefits group membership has for people deprived of personal control: low personal control increased perceived social support by group members, mediated via ingroup identification. This complements research investigating the curative potential of social groups and the effects of ingroup identification on health and well-being (Haslam et al., 2009; Jetten et al., 2014). Previous findings showed the beneficial effects of identification on well-being following rejection and therefore threat to self-esteem (Branscombe et al., 1999) and the stress reducing effect of ingroup identification through perceived ingroup support (Haslam et al., 2009). More recently, Greenaway et al. (2015) found that perceived personal control mediated the beneficial effects of ingroup identification on personal well-being. The present findings support a control path on which the curative function of ingroup identification can unfold. They show that control threat enhances identification with agentic groups that in turn alleviate control loss: membership in agentic groups can help to restore and defend a threatened sense of control by providing group members with sense of collective control.

### Limitations

In the present research, we compared a control threat salience condition with a condition in which high personal control was salient, without employing a neutral condition. Hence, it is possible that attractiveness of agentic groups does not increase following control threat, but decrease following control salience. However, we assumed the control threat condition to drive the effects, because in previous research control threat effects on ingroup defense emerged as compared to both, a high control condition and a neutral (i.e., dental pain) condition (Fritsche et al., 2008). No differences were observed between high control and the neutral topic. Nevertheless, future research would benefit from including a neutral condition to distinguish the control threat effects from possible effects of boosted control.

# Implications for Collective Action and Attractiveness of Social Movements

The finding that collective agency is the crucial feature that increases group attractiveness for people with low perceptions of personal control could help to explain why some groups gain members and support in society in times of threat and crisis while other groups do not (Fritsche et al., 2011). As perceived collective agency seems to restore a sense of control in people that are personally affected by societal crises, those groups that allow for perceptions of collective agency should gain most. Social movement organizations (Stürmer and Simon, 2004), such as gay-rights organizations or pro-environmental action groups are among those groups that are intimately associated with collective agency as they are set up for mastering collective tasks (i.e., they are true task groups). However, societal crises may also give rise to destructive forces and radicalized groups that become attractive for control deprived people when these groups unite against a common enemy thereby demonstrating collective agency through zealous protests or even violent actions. Summed up, there is reason to believe that, beyond perceptions of collective threat or disadvantage (van Zomeren and Iyer, 2009), collective action participation might be fertilized by threat to people's personal sense of control.

The possibility of increased collective action under conditions of threatened control shows that mechanisms of group-based control might be adaptive for resolving personal helplessness. This perspective adds to other recent research on social responses to control threat (Kay et al., 2008, 2009). Kay et al. (2008) have proposed that instead of trying to regain control through the (social) self, people may respond to threatened personal control by supporting external agents of control, such as God or the national government, and attributing control over bad outcomes to powerful enemies (Sullivan et al., 2010). This is thought to prevent a sense of randomness and lacking structure as external agents impose their order on the world. However, the present findings indicate that although personally helpless, people may first check out possibilities of extending the self to an agentic social ingroup that restores a sense of control through the self (primary control; Rothbaum et al., 1982) before they resort to external agents to preserve perceptions of order (secondary control; Rothbaum et al., 1982).

Our findings further imply which group people prefer when different groups are available. For control-deprived individuals groups focusing on similarity and similar appearance and less on coordinative effort in goal attainment, seem not to be first choice, if other groups exist that appear highly agentic. A group with a certain lifestyle, such as bohemianism, might be attractive to people because of the shared idea to live for art and love in an unorthodox, impoverished and unconventional manner, but that group will probably not be known for its agency, and might be therefore not primary for experiencing collective control. Instead, a successful political action group that actively fights for its goals may provide collective agency to potential members. This implies that the attractiveness of groups for new members could be emphasized by indications of agency. Groups that engage in collective doing, project planning or other forms indicative of concerted action should foster the perception of the group as agentic and allure people especially in times of personal threat.

In view of political extremism, deprivation of personal control may be added as another motivational determinant, in addition to self-uncertainty (Hogg, 2014), for engagement in political extremist groups that particularly consist of ideological or religious zealots that pursue group goals in a vigorous and tough-minded manner. Radicalized groups not only provide their members with a sense of a clear and distinct self that helps to reduce their self-uncertainty (Hogg, 2014) but in addition, radicalized groups can provide a sense of collective agency that helps to restore their sense of control. Violent extremist groups may stress their agency when they violate all norms of human co-existence as this highlights absolute commitment to a shared superordinate goal (it is even worth to violate all rules of conduct), free decision (against "social desirability" concerns), and the strength of active goal pursuit (which obviously cannot be prevented by others although they should be extremely motivated to stop the violations). These days, the rapid surge of extremist groups that attract foreigners from different countries who want to fight zealously for an "Islamic State" illustrates nicely the agentic potential these extremist and inhuman groups could offer. Personal feelings of lacking control may be one motivational factor that helps to explain why such vigorous campaigns gain support and new followers.

# Acknowledgments

The present research was supported by the German Research Association DFG grant FR 2067 3-1/-2/-3. We thank Judith Berles, Angelika Bühler, Nils Kupfer, Juliane Seiler, and Fleur Zeunert for conducting Study 1.

# Supplementary Material

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpsyg. 2015.00649/abstract

# References


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2015 Stollberg, Fritsche and Bäcker. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Motivation for aggressive religious radicalization: goal regulation theory and a personality **×** threat affordance hypothesis **×**

#### *Ian McGregor1\*, Joseph Hayes1 and Mike Prentice2*

*<sup>1</sup> Department of Psychology, University of Waterloo, Waterloo, ON, Canada, <sup>2</sup> Department of Psychology, University of Salzburg, Salzburg, Austria*

A new set of hypotheses is presented regarding the cause of aggressive religious radicalization (ARR). It is grounded in classic and contemporary theory of human motivation and goal regulation, together with recent empirical advances in personality, social, and neurophysiological psychology. We specify personality traits, threats, and group affordances that combine to divert normal motivational processes toward ARR. Conducive personality traits are oppositional, anxiety-prone, and identity-weak (i.e., morally bewildered). Conducive threats are those that arise from seemingly insurmountable external forces and frustrate effective goal regulation. Conducive affordances include opportunity for immediate and concrete engagement in active groups that are powered by conspiracy narratives, infused with cosmic significance, encouraging of moral violence, and sealed with religious unfalsifiability. We propose that ARR is rewarding because it can spur approach motivated states that mask vulnerability for people whose dispositions and circumstances would otherwise leave them mired in anxious distress.

Keywords: religion, radicalization, aggression, approach motivation, avoidance motivation, anxiety, groups

# Introduction

*"We have killed all of the children in the auditorium...what do we do now?"*

*—Taliban gunman, December 16, 2014*

After methodically shooting all 132 children and 12 teachers at a school in Peshawar, Pakistan, the Taliban militant, Abuzar, called his handler for further instructions. The answer came back, "wait for the army to arrive, kill them, and then blow yourself up" (Khan, 2014). Abuzar and the six other gunmen complied and detonated their vests on cue. This kind of self-immolating, aggressive religious radicalization (ARR) has recently animated high profile Islamist atrocities by the Islamic State (IS), al-Qaeda, Taliban, Boko Harem, al-Shabaab, and lone-wolf extremists all over the world. Although Islam is currently in the spotlight, ARR is not a Muslim phenomenon. Throughout history its callous extremes have blighted all major religions traditions (Armstrong, 2000). This paper provides a theoretical framework with data-driven hypotheses about how basic human motivations interact with situational affordances to make ARR alluring.

#### *Edited by:*

*Eddy J. Davelaar, Birkbeck, University of London, UK*

#### *Reviewed by:*

*Kai Sassenberg, Knowledge Media Research Center, Germany Aaron L. Wichman, Western Kentucky University, USA*

#### *\*Correspondence:*

*Ian McGregor, Department of Psychology, University of Waterloo, 200 University Avenue West, Waterloo, ON N2L 3G1, Canada ian.mcgregor@uwaterloo.ca*

#### *Specialty section:*

*This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology*

*Received: 30 October 2014 Accepted: 18 August 2015 Published: 15 September 2015*

#### *Citation:*

*McGregor I, Hayes J and Prentice M (2015) Motivation for aggressive religious radicalization: goal regulation theory and a personality* × *threat* × *affordance hypothesis. Front. Psychol. 6:1325. doi: 10.3389/fpsyg.2015.01325*

In the recent IS instantiation, over 30,000 young men and women from over 100 Asian, Middle Eastern, and Western countries have abandoned their normal lives to go fight with IS in foreign territory and abject living conditions. Indeed, more British Muslims have joined IS than the British military (Barrett, 2014a,b; BBC, 2015; The Soufan Group, 2015; Weaver, 2015). When they get there they sometimes burn their passports as a show of commitment before submitting to the harsh regimen of discipline for the fascist cause. They know they may die soon after arriving, and many do, either as suicide bombers or as casualties in battles picked with a more powerful enemy—a coalition that includes many of the world's most powerful countries. Despite the shocking atrocities perpetrated by IS that include beheadings, crucifixions, rapes, pedophilia, and genocidal slaughter (BBC, 2014), and that have been well publicized on YouTube, IS continues to attract foreign recruits who are often among the most zealous (Barrett, 2014b). What is the appeal of enthusiastically perpetrating atrocities in the name of a religion that preaches mercy? How can personality and demographic profiles of recruits so often be normal and well-educated (Post, 1990, 2005; Barrett, 2014a).

We present a basic motivational framework that contests popular views claiming ARR is primarily a pragmatic revolutionary strategy, or driven by self-serving superstitions, or fueled by the particularly aggressive nature of any particular religion. We draw on classic and contemporary psychological theory and over 30 years of experimental research showing how and why psychological threats cause belligerent defenses and reactive approach motivation (RAM; McGregor et al., 2010a; Jonas et al., 2014). We hypothesize that ARR arises from personal predispositions, anxiogenic threats, and group affordances that combine to divert normal motivational processes toward approach motivation for ARR. Each factor alone is not enough, but conducive personality, threats, and affordances, together, are potent. Our view is compatible with insights about how motivation for worldview defenses, significance, meaningful engagement, identity-fusion, group-based control, belonging in action groups, and compensatory conviction and consensus might propel ARR (McGregor et al., 2001, 2005; Rothschild et al., 2009; Atran, 2010; Sageman, 2011; Fritsche et al., 2013; Hogg, 2014; Kruglanski et al., 2014; Swann and Buhrmester, 2015). Our view augments this previous work by grounding its hypotheses in primitive motivational substrates that can provide additional depth to the emerging understanding of motivation for ARR.

Our main premise is that ARR is rewarding because it spurs approach motivated states that mute anxiety for people whose personalities and social circumstances would otherwise leave them anxious and depressed (McGregor et al., 2010b; Jonas et al., 2014; Hayes et al., 2015). We begin by reviewing classic theories foundational to understanding ARR. We then link them to contemporary goal-regulation premises, theories, and recent advances in RAM theory. We next define the components of ARR, and use empirical research findings to justify our personality × threat × affordance hypotheses. We conclude by suggesting strategies for testing our hypotheses in the lab and real world.

# Classic Theory

A premise in classic and contemporary theory is that conflicts, frustrations, and uncertainties can propel belligerent reactions far removed from the eliciting difficulties. Sigmund and Anna Freud popularized the view that conflicting motivations in competing directions can arouse bizarre and extreme defenses against the ensuing anxiety. "Excessively intense," "supervalent" thoughts form "mental dams" that effectively repress the offending conflicts Gay (1989, pp. 200, 261–262). From this perspective ARR-relevant defense mechanisms such as turning against the self, rationalization, fantasy, regression to childish tendencies, and projection of one's own hostilities onto others would be considered mechanisms of repression that help people escape from other motivational conflicts in their lives (Freud, 1946).

Lewin (1933) brought Freudian ideas into a more general theory of motivation and goal regulation by showing that conflicts and uncertainties cause a kind of anxious tension that persists as long as goals remain impeded. If people have no clear way to relieve the tension arising from their (often social) conflicts then they escape from the field of tension by resorting to fantasy, submission, or belligerence (Lewin, 1933, 1935).

Neo-analytic theorists similarly proposed that aggression is a reflexive response to frustration for people and animals. The frustration-aggression hypothesis (Dollard et al., 1939; Berkowitz, 1989) is implicit in neoanalytic views that failure to forge identities grounded in prosociality tilts people toward fanaticism and oppositional power over others (Horney, 1950; Ansbacher and Ansbacher, 1956; Erikson, 1959). Horney (1950, pp. 86–109, 184) and Adler (Ansbacher and Ansbacher, 1956, pp. 259–261) respectively referred to this tendency as "arrogant righteousness" and "neurotic pride" or a "superiority complex," wherein grandiose and oppositional facades mask insecure selves. Fromm (1941) viewed such defensive tendencies as escapes from existential freedom and saw conformity, authoritarianism, fascism, and destructiveness as symptoms of human's inability to cope with the frustrating uncertainty about how to make choices in life. People cling aggressively to externally referenced "frames of orientation and devotion" to avoid becoming overwhelmed by existential uncertainty (Fromm, 1947, p. 48). Antisocial extremes replace gnawing uncertainty with decisive commitment that relieves angst. Durkheim (1897/1951) viewed uncertainty about what to do as noxious enough to cause suicide. If viable family relationships or cultural norms are not available to provide clear direction and purpose, people use suicide to escape the unbearable burden of choice (Durkheim, 1897/1951; see also Baumeister, 1990).

Sartre's (1956) existential philosophical perspective similarly held that uncertainty arising from radical freedom spurs attempts to escape from the nausea of uncertain self-awareness by conforming to group norms (Barnes, 1973). Under this view, masochistic extremes of submission to authority and sadistic extremes of domination and hate are attempts to suppress uncertainty arising from absence of objective truths that could guide one's choices in life. Masochism and sadism reduce the uncertainty associated with awareness of multiple perspectives by eliminating one's own and/or others' subjective perspectives.

Problematic subjectivity, with its conflicting perspectives and uncertainties, is replaced by the solid ground of supposed objective authority.

These neoanalytic and existential ideas were the foundation for the Adorno et al. (1950) treatise on the authoritarian personality and causes of fascist disdain for deviants. Authentic identity development requires a vulnerable process of trial and error exploration (Rogers, 1951; Erikson, 1959; Marcia, 1980; Deci and Ryan, 1991). If this process is blocked by unsupportive people or chaotic social structures, then confident personal values and priorities for resolving uncertainty are not developed, and people turn to the dictates of powerful others and groups. Fusing or identifying with a fascist ingroup can thereby become psychologically vital, and critics become the enemy.

These classic theoretical views predict the link between conflict, uncertainty, and frustration, and aggressive radicalization, but they do not imply a specifically religious form of aggressive radicalization. James (2006) made that connection around the time that Freud was penning his first ideas about repression via commitment to excessively intense thoughts (Gay, 1989, pp. 200, 261–262). James (2006, p. 240, 399) observed that "religious rapture and moral enthusiasm are unifying states of mind that incline the sand and grit of selfhood to disappear," and that can unify a "discordant self." Consistent with our approach-motivation hypothesis he further proposed: "We shall see how infinitely passionate a thing religion at its highest flights can be. Like love, like wrath, like hope, ambition, jealousy, like every other instinctive eagerness and impulse, it adds... a new sphere of power" (James, 2006, p. 58).

These repression-related views were first put into goalregulation language by Lewin's (1933, 1935) understanding of how strong goals and commitments can clear away other conflicts, leaving people feeling sanguine and single-minded (reviewed in McGregor, 2003). Lewin (1933, p. 609) paved the way for a goal-regulation view of zealous religious devotion as an idealistic commitment that can function like a motivational "field of force" to push other uncertainties and frustrations out of awareness (see McGregor et al., 2010b, 2012b, for elaboration on the goal and emotion-regulation function of ideals). Contemporary social psychological and social neuroscience research now provides a clearer, less metaphorical understanding of the basic motivational mechanics beneath Lewin's seminal goal-regulation ideas.

# Contemporary Theories of Threat and Defense

In the second half of the 20th century, personality and social psychological research began empirically testing and refining classic theories about causes of defensive social phenomena related to ARR. Hundreds of experiments on cognitive dissonance theory (begun by Lewin's student, Festinger, 1957) demonstrated that experimentally manipulated cognitive conflicts could cause extremes of opinion rationalization that defied logic, including dubious claims by religious cult members (Festinger et al., 1956). The first generation of dissonance theory and research had its roots in classic psychodynamic theory, from Freud, through Lewin, to Festinger's experimental demonstrations. Its growth became even more psychodynamic with neo-analytically inspired demonstrations of compensation. Research began to support Allport's (1943, p. 466) assertion that various ego defenses provide "fluid compensation" for psychological discomfort arising from threats, conflicts, and uncertainties. Self-serving affirmations of worth, conviction, morality, meaning, or adaptive adequacy of any kind were found to mute threat-induced distress, even if they did not directly address the content of the original threat (e.g., see Steele, 1988; Tesser, 2000; McGregor et al., 2001; Heine et al., 2006). As Lewin would have predicted, the affirmations make the anxious uncertainties and conflicts less motivationally salient (McGregor, 2006a), which makes them less aversive (McGregor et al., 1999).

A proliferation of threat and defense theories sprouted around the intuitive idea that fluid compensation occurred because threats to a psychological resource (esteem, security, integrity, immortality, belongingness, meaning, or control) aroused compensatory reactions that served to replenish the threatened psychological resource, often in disguised ways. For example, if an anxiety-inducing failure threat caused a worldview defense reaction involving hostile derogation of an outgroup or moral offender, various compensatory theories would interpret this as disguised compensation for threatened esteem, certainty, conviction, integrity, immortality, security, belongingness, meaning, or control (i.e., depending on the authors' theoretical allegiance; see Jonas et al., 2014 for review). Compensation perspectives would accordingly view ARR as an indirect strategy for restoring whatever basic need had been indirectly undermined by psychological threat. As powerful and generative as resource compensation theories became, their proliferation and explanatory competition ushered in an integrative theoretical approach to understanding threat and defense processes at a more basic motivational level.

# Goal Regulation Theory: A Lens for Understanding ARR

Primitive motivational structures in humans, mice, and other vertebrate brains are organized around goal dynamics and anxiety. Behavioral, lesion, and pharmacological studies reveal that goal frustration and uncertainty are the prime causes of anxious distress, mediated by the septo-hippocampal behavioral inhibition system (BIS; Gray and McNaughton, 2000). The neurophysiology of anxiety is different from that arising from other aversive states like sadness or panic. As with Lewin's (1933, 1935) idea of tension, anxiety arises from the approach-avoidance conflicts inherent in goal blockage, uncertainty, novelty, and frustration. Recent revisions to Gray's theory (by his student, Corr, 2008), emphasize that BIS activity is inversely related to activity of the other main motivational system, the behavioral activation system (BAS). The BAS promotes single-minded approach motivation—the "impulse to go toward" (Harmon-Jones et al., 2013, p. 291). It is an eager, goal commitment system that, when active, mutes the BIS (Corr, 2008; Nash et al., 2012). Effective goal regulation is maintained by the interplay between these two systems. When goals are going well the BAS predominates and makes people feel energized, eager, and single-mindedly committed to approach of the focal goal. BAS activation automatically inhibits possible distractions and conflicts, and allows people to focus on eagerly approaching identified incentives, free from anxious preoccupation with their worries. If a serious conflict or uncertainty erupts along the way and succeeds in activating the BIS, however, then the BAS is muted and three primary BIS outputs result. All ongoing goals are inhibited to stop the animal from doing whatever is not working; anxious distress further discourages persistence at ongoing goals; and vigilance dilates to help the animal notice a wider range of possible threats or opportunities that could cue single-minded withdrawal or approach to get the animal out of the conflicted and anxious state, and onto a more viable track.

This account of BIS and BAS follows Lewin's (1933, 1935) view of how tensions aroused by goal conflict can be blocked out by focused immersion in other goals. In more contemporary language, activating the BAS through goal commitment eliminates the hesitant, anxious vigilance associated with the BIS. Throwing oneself into a commitment like ARR could accordingly be a way to effectively repress other anxious conflicts (Harmon-Jones et al., 2009; Nash et al., 2011; McGregor et al., 2013b; see Jonas et al., 2014 for overview of the underlying basic processes and links to threats and defensive reactions).

This goal and emotion regulation function of the BAS has been empirically demonstrated in humans by personality and social psychology experiments focused on goal commitment, shielding, and implementation. Tenacious goal commitments and related eager approach motivation processes activate the BAS and narrow motivational attention to the focal goal (Harmon-Jones et al., 2011, 2012), shielding it from interference by other conflicting or competing goals (Shah et al., 2002). The single-mindedness facilitates vigorous goal completion (Elliot and McGregor, 1999; Harmon-Jones et al., 2009) and is also affectively rewarding—people feel more optimistic when actively engaged in implementing a commitment than when deliberating about alternative possibilities (Taylor and Gollwitzer, 1995). The same insulation from conflicts and uncertainties can come from eager immersion in more abstract goals, values, and group identifications (McGregor et al., 2001; McGregor, 2006a, Study 1; McGregor and Marigold, 2003, Study 4; McGregor et al., 2005, Study 4) because values and groups also activate the BAS (Agroskin, unpublished doctoral thesis, McGregor, unpublished data), especially if they are extreme (Sleegers et al., 2015). Cybernetic theories of goal regulation posit that ideals and values are abstract goals that function as self-guides or system concepts to organize and coordinate the array of subordinate, more concrete goals (Higgins, 1996; Carver and Scheier, 1998; Hirsh and Kang, 2015). Extremes are resistant to ambivalence, so they should be particularly effective as clear self-guides (Newby-Clark et al., 2002).

These basic goal regulation processes furnish a parsimonious, mechanistic account for the operation of classic theoretical ideas about why people in anxious circumstances turn to excessively intense thoughts, moral enthusiasms, exclusive fields of force associated with goals, or the other manifestations of rigid and extreme opinion, devotion, and authoritarian hostility. All might effectively function as levers for BASactivated approach-motivated states that are rewarding because they mute BIS activity. Considering that anger is also a powerfully BAS-activating phenomenon (Carver and Harmon-Jones, 2009), it seems plausible that extreme commitment to ARR might be an appealing response to anxiety-inducing threats, uncertainties, frustrations, and injustices because ARR contains several elements capable of transitioning people from "anxiety to approach" (Jonas et al., 2014).

# From Anxiety to Reactive Approach Motivation for ARR

The basic-process ideas outlined above were first organized in papers identifying exaggerated conviction, pride, consensus, and intergroup animosity as levers for approach-motivationrelated states that people use to downregulate threat-activated BIS (McGregor et al., 2005; McGregor, 2006b). Over the last 10 years this speculation has been empirically supported by rigorous experimental research. Anxiety-related threats (dissonance, uncertainty, failure, control loss, mortality salience, relationship distress, insecurity, goal-frustration) that have caused extreme "compensatory" reactions in past research have also now been found to cause neural indicators of BIS activation at first, and then RAM (reviewed in Proulx et al., 2012; Jonas et al., 2014). The measures of RAM include basic neural, perceptual, and affective evidence, along with eager and idealistic commitment to personal goals and commitments in everyday life (McGregor et al., 2007, 2009a, 2010a, 2013b; Nash et al., 2011; Greenaway et al., 2015). Importantly, these same threats also cause self-reported endorsement of religious extremes and increased willingness to kill and die for religious beliefs (Pyszczynski et al., 2006; McGregor et al., 2008, 2010b, 2013b; Rothschild et al., 2009; Wichman, 2010). Further, laboratory experiments now indicate that elements of ARR, and religious devotion itself, can cause neural, perceptual, and self-report evidence of approach motivation (reviewed in Jonas et al., 2014; Agroskin, unpublished doctoral thesis, McGregor, unpublished data). Anxiety-to-approach dynamics are thus well positioned to help explain the enigma of ARR.

# Pragmatic and Palliative RAM

It is important to acknowledge, however, that levers for activating RAM are not necessarily defensive and irrational. Constructive responses to anxiogenic circumstances can also provide RAM relief from anxiety. Indeed, this may be the most usual and adaptive function of anxiety-to-approach processes, as in tenacious striving for financial security after deprivation, or for success or love after failure or rejection. It is when direct resolution opportunities seem blocked and hopeless, however, that people turn to merely palliative defenses like ARR to activate RAM for relief.

It has been argued that ARR might be a direct, constructive form of freedom fighting aimed at restoring social justice for oppressed people (Armstrong, 2014). The suffering inflicted on Muslims caused by the long history of Western hegemony, hypocrisy, political interference, exploitation for oil, economic sanctions that mostly harm poor civilians, and military invasions have been identified as catalysts for Islamist extremes. Although realistic grievance is surely part of the story (Armstrong, 2014; Barrett, 2014a,b), and ARR may be partly motivated by pragmatic motivation to make constructive improvements, characteristics of ARR cast doubt on the adequacy of such straightforward explanations.

First, ARR is often rash and counter-productive. In one of the first examples of ARR that gave rise to the word "zeal," an oppressed Zealot sect of Judaism assassinated anyone who disagreed with their extreme agenda, even those in their own group who did not seem devout enough. Their extremism brought annihilation from the governing Romans. A similar plight befell the first Christian Crusaders. After Pope Urban's rousing 1086 CE speech about restoring the glory of Charlemagne and saving the Holy Land from Evil, a band of overly enthusiastic and unprepared Crusaders broke from France for Jerusalem before the designated date. In their zeal they began slaughtering anyone along the way who seemed a different race or religion. They were soon annihilated by the first wave of organized resistance that they faced (Durant, 1950). The violence of IS and other ARR groups seems similarly rash and counter-productive (Post, 1990; Barrett, 2014b). Their self-publicized atrocities may have some strategic value insofar as they discourage resistance in the towns they occupy, but they have also turned most of the world against them. Even predominantly Muslim countries like Iran, Iraq, Jordan, Saudi Arabia, Bahrain, United Arab Emirates, and Turkey, that might otherwise have been sympathetic to legitimate social justice grievances, have joined a coalition with Western powers against IS. The Muslimon-Muslim atrocities of IS have even alienated the radical Islamist group, al-Qaeda, from which IS evolved (Barrett, 2014b).

A second reason for doubting the pragmatics of ARR is the retrograde nature of its religious claims. Mature religion is usually associated with humility, recognition of mystery, compassionate concern for others, and aversion to violence (Armstrong, 2006, 2009; Schumann et al., 2014; Shariff et al., 2015). These characteristics are at odds with the fascist, black and white, superstitious, and hostile characteristics of ARR that more closely resemble psychological defenses (Post, 1990; Jonas et al., 2014). Indeed, among ARR recruits there is often only a thin veneer of religious knowledge supporting their devotion (Atran, 2010; Barrett, 2014a). Pragmatic devotion would presumably be built on a more mature foundation.

The third reason to doubt the pragmatics of ARR is the risky self-destructiveness of its members. New recruits leave lives and loved ones behind and risk everything. A longer life of careful, strategic devotion to a solid cause would presumably accomplish more than a quick and dirty death for a dubious and sensational cause. The relish to join, fight, and

risk for an extreme cause seems to have more psychological appeal than instrumental benefit (Nash, unpublished doctoral dissertation; McGregor et al., 2013b; Black et al., 2014; Hogg, 2014).

# A Concrete and Social Approach

Our specific hypotheses below about the kinds of people and circumstances conducive to ARR are informed by a recently advanced taxonomy of phenomena people approach for relief from BIS-activation. People react to anxiety-inducing experiences by approaching phenomena that are either personal or social and either concrete or abstract. The four domains of phenomena people use to activate RAM are accordingly concrete personal (e.g., money, personal aggression, power, control); concrete social (e.g., group membership, group aggression, group power, group status); abstract personal (personal values, ideals, moral convictions); and abstract social (collective worldviews, ideologies). Phenomena in all quadrants can be eagerly pursued to activate approach motivated states (Jonas et al., 2014). Ostensibly religious phenomena can populate all four quadrants, e.g., ritual action in the concrete personal quadrant; coordinated group rituals, coalitional action, or intergroup hostility in the concrete social quadrant; idiosyncratic ideals and values in the abstract personal quadrant; and consensual worship of cultural symbols, worldviews, and meanings in the abstract social quadrant. Our working definition of religion emphasizes the idealistic aspects of religious devotion (see below) but the concrete aspects can serve as accessible sacraments for orienting toward the idealistic elements, especially (as we develop below) for people who may prefer concrete engagement. Our view is consistent with Armstrong's (2000, 2009) claim that fundamentalist ARR cleaves to the concrete in an arguably blasphemous attempt to remove the inherent mystery from its understanding of God. Our position here on the concrete nature of ARR is accordingly a departure from our past focus on more abstract aspects of idealistic religious devotion (e.g., McGregor et al., 2010b, 2012a).

Past investigations of phenomena people spontaneously use to activate RAM focused on abstract-personal levers. Random assignment to various, 2–5 min anxiety-related experiences (e.g., personal uncertainties, mortality salience, relationship insecurities, social exclusion, performance anxiety) caused people to become more extreme in their moral opinion conviction and consensus estimates, and in pursuit of idealistic goals, convictions, and meanings (McGregor et al., 2009b). These abstract-personal reactions appear to relieve anxious distress by activating approach-motivated states (Jonas et al., 2014). The same threats also heighten abstract-personal varieties of religious devotion—more confident certainty in the objective truth of selfidentified religious beliefs, more determination to live according to them, more identification with them, and more willingness to argue in defense of them (McGregor et al., 2008, 2010b, 2013b).

People with eager, idealistic, and approach-motivationcorrelated traits have been most inclined toward use of abstractpersonal levers for RAM (reviewed in Jonas et al., 2014). Based on this we had initially assumed that eager, idealistic, and confident kinds of people should be especially inclined toward ARR, under the assumption that ARR was essentially an abstract ideology (McGregor et al., 2008, 2010b, found that eager, idealistic people were most inclined toward reactive, abstract religious extremes). Our most recent research, however, suggests that the kind of abstract-personal religious zeal that our previous research focused on is *not* what ARR usually revolves around. Like ethnocentrism and outgroup derogation, ARR is usually idealistically impoverished, and seems to revolve more around opportunity for concrete participation in hostile authoritarian groups that mobilize aggressive action disguised by a veneer of oversimplified ideology (Post, 2005). We propose that it is through this concrete participation in simplified, black and white ideologies that identity-weak people (i.e., morally bewildered) are able to acquire an externally referenced sense of "identity, purpose, belonging or spiritual fulfillment" for a "greater sense of purpose and meaning in their lives" (Barrett, 2014a, p. 18; Barrett, 2014b, p. 7; Atran, 2015). Groups, aggression, and action can activate approach motivate states, and approach motivated states feel meaningful (McGregor et al., 2012b). Given the seeming concrete and authoritarian characteristics of ARR, we here hypothesize different predisposing personality traits than in our past research. Whereas identity-strong people have tended toward abstractpersonal levers for RAM, identity-weak people should be most inclined toward ARR.

Without the guiding and constraining influence of mature personal or religious identity, which tend to be prosocial, the more concrete, risky, aggressive, hostile, and coalitional kinds of levers for activating RAM could be especially alluring. Indeed, markers of concrete coalitional but not intrinsic religion predict support for suicide bombers and prejudice (Allport and Ross, 1967; Ginges et al., 2009). A limitation of concrete, extrinsic religion, however, may be that it is less reliable and efficient in maintaining approach-motivated relief. It requires involved physical engagement because it may not be as easily summoned as abstract ideals in private imaginations (McGregor et al., 2012b). Further, despite temporary relief engagement in ARR might provide, its antisocial hostility presumably impedes harmonious social functioning and brings more social conflict and anxious insecurity over time. Indeed, internal conflict is predicted to be the downfall of IS (Barrett, 2015).

Once it becomes clear to identity-weak people that ARR doesn't work as well as hoped, they might either amplify zeal, or withdraw altogether from life so as to activate unmitigated avoidance motivation. Anxiety is a function of the BIS response to simultaneous approach and avoidance cues. It can accordingly be relieved by either singular approach or singular avoidance (Hayes et al., 2015). Withdrawal from concern with life outcomes would make hostile and antisocial levers for RAM easier to engage without regard for possible consequences. In sum, the appeal of ARR may be that it offers opportunity for toggling between concrete approach (aggressive, powerful, hostile, coalitional) and fatalistic withdrawal from life through self-immolating extremes. Both are anxiety-relief strategies available to identity-weak people.

# Components of ARR

# Aggressive

By aggressive we mean a tendency to assert ones will against others, oneself, or any symbolic or concrete target in a way that can augment the feeling of power, status, or control vis-àvis the target. Belittling, overpowering, or destroying others, the self, institutions, or properties are aggressive by this definition. Violence is a concrete manifestation of aggression.

# Religious

As reviewed above, philosophical and goal-regulation perspectives propose that humans need moral ideals to guide concrete goals. Without moral ideals humans can become mired in conflict among all the imaginable possibilities for action. Following Fromm's (1973, p. 260–261) view that a primary existential need for humans is "an object of total devotion*...*to be a focal point of all his [sic] strivings," we view religion as a vehicle for moral values that is often bolstered by consensual rituals and symbols (Geertz and Banton, 1966). This morality-focused view of religion is compatible with Durkheim's (1976/1912) seminal claim that religions are not necessarily about gods (though they often are—Gods are potent symbols), but rather that they revolve around group convictions about what is to be valued as sacred, above any particular temporal concern. Worship and ritual involve active group affirmations of these sacred phenomena as worthy of veneration. From this perspective, the essence of religion is less about superstitious belief in existence of supernatural entities and their intra-group moral policing function (cf., Shariff and Norenzayan, 2007) than about eager devotion to moral commitments that function as arbiters for coordinating enthusiastic action within individuals and groups.

Some historians of religion have similarly concluded that the essence of religious devotion revolves around functional meanings that help people set priorities for effective living (Smith, 1986; Armstrong, 2006). Armstrong argues that the jagged evolution toward compassionate values across religious traditions has been occurring because prosocial values are the only kind that can sustain consensus and cooperation, and constrain costly inter-group violence. Drawing on existential, psychological, and religious-historical perspectives we accordingly define the essence of religion as a moral orientation toward action that is often but not necessarily anchored by ideas of God, that is often but not necessarily bolstered by consensual ideology and ritual, and that is usually prosocial but can sometimes endorse aggressive and fascist extremes.

The vulnerability toward fascist extremes exists because devotion to abstract moral ideals can be difficult for solitary individuals. Ideals and values have no concrete referents and so rely on consensus for confidence (Festinger, 1950; Wicklund and Gollwitzer, 1982; Hardin and Higgins, 1996). "Conservative," coalitional moral foundations relating to ingroup consensus (authority, loyalty, purity) sometimes co-occur along with the more universal social justice moral foundations (fairness and care; see Graham et al., 2009 for evidence of conservative vs. universal moral foundations). Among people who are vulnerable to anxiety (Jost et al., 2003; Olvet and Hajcak, 2008) and in frustrating circumstances conducive to aggressive reactions and ingroup cohesion and consensus (Dollard et al., 1939; Shah et al., 1998; Kruglanski et al., 2002; Miller et al., 2003; McGregor et al., 2005; Burke et al., 2010), the balance between social justice idealism and the more concrete coalitional and authoritarian moral foundations can list toward coalitional/authoritarian. When they eclipse social justice, ARR can flourish.

# Radicalization

Radicalization is a shift from mainstream to anti-normative or comparatively extreme ideological convictions that animate eagerness to challenge the status quo. Radicalization is neither necessarily aggressive nor religious. Martin Luther King Jr. and Mahatma Ghandi were radicals, as was Tommy Douglass, the politician who won universal health coverage for Canadians in the 1960s. Barack Obama's quest for more universal healthcare in the US is still considered radically subversive by half of the US population. All of the above might be considered religious radicals to the extent that their radical determination for social justice was girded by their own religious values. Their somewhat religious radicalization was pragmatic and constructive, however, not aggressive.

From our perspective ARR is usually motivated more by the psychological appeal of the radical and aggressive lifestyle than by a constructive assessment of what the radical agenda is likely to accomplish or by thoughtful religious integrity (see also Post, 1990; Atran, 2010, 2015, for more on the superficial religious knowledge and idealistic impoverishment of ARR recruits). We propose that people are drawn to ARR simply because it feels right. Here we develop a theory for why something that seems bizarre and abhorrent to most people could be attractive for some. Based on past theory and research we hypothesize that specific personality, threat, and affordance factors combine to make ARR feel right.

# Factors Conducive to ARR: Personality, Threat, and Affordance

When episodes of ARR occur they are typically met with astonished exclamations of "why him," "why there." Why, for example, did one sleepy Norwegian town spawn eight IS recruits (Higgins, 2015), one pre-university school in Montreal 11 (Perreaux, 2015), and one elementary school in Morocco five of the seven Madrid train bombers (Atran, 2010)? Part of the reason may be that the individuals belonged to tightly knit, actionoriented social networks, such as neighborhood groups or soccer teams that set a group-action template for them to engage in heroic adventure together in another context (Atran, 2010). But most people in buddy-groups are not drawn to ARR. Blaming personality predispositions or demographic characteristics is also little help because perpetrators of ARR are surprisingly normal (Post, 2005). Environmental threats, frustrations, and anxieties may similarly be ingredients (McGregor et al., 2010b), but most anxious and frustrated people do not turn to ARR.

Difficulty identifying causal factors in past research may have arisen from failure to simultaneously consider combined personality, threat, and affordance interactions. Here we identify 10 ARR-facilitating influences across the three factors (personality, threat, and affordance). For each influence to be above average in prevalence would occur by chance with a 2<sup>10</sup> probability of only 1/1024. It is not surprising, then, if our view is correct, that isolated personality traits or threat influences often fail empirical tests of relevance. Some laboratory studies (described below) have found combinations of a few of these 10 influences can cause self-reported movement toward aspects of ARR. Real life ARR surely requires more influences to align, however, because barriers to ARR in real life are higher than for self-reported opinions or intentions.

The personality factor includes three influences from normalrange personality traits that lean toward: (a) oppositional personality traits related to aggression (disagreeableness, hostility, anger, narcissism, or low self-control); (b) high BIS personality traits that incline people toward the experience of anxiety; and (c) identity-weak personality traits that undermine capacity to assert personal value priorities, take initiative, or maintain self-regulatory control. The threat factor includes two influences: (a) external control threats that undermine faith in the social system (that identity-weak people are especially inclined to rely on; and (b) life circumstances that underscore hopelessness. The affordance factor includes five influences: (a) opportunity for immediate, concrete engagement with active ARR groups, (b) a consensual injustice narrative that condenses conspiracy-theory blame for system and self-dissatisfaction onto the outgroup, (c) narratives that convey heroic participation in a cosmic battle between good (own group) vs. evil (outgroup), (d) justifying narratives for religious aggression, and (e) unfalsifiable religious arguments.

# Personality Oppositional

A history of delinquency or criminal activity prior to ARR is a common but not a clear predictor (Post, 1990; Atran, 2010). Its incidence may arise from intercorrelated dispositional leanings toward aggression, narcissism, disagreeableness, and low selfcontrol that have been linked to ARR-related phenomena in other research (and that are all correlated at around *r* = 0.3, McGregor, unpublished data). Trait-aggressive and narcissistic people are especially receptive to aggressive media exposure and inclined toward displaced hostile reactions to frustration, perceived provocation, and rejection (Bushman, 1995; Bushman and Baumeister, 1998; Anderson and Dill, 2000; Twenge and Campbell, 2003). Narcissistic and the other "dark tetrad" traits of psychopathy Machiavellianism and sadism (Paulhus, 2014) are significantly correlated with each other and with high agency, low agreeableness, punitiveness, and callous low empathy (Watson et al., 1984; Campbell et al., 2002; Paulhus and Williams, 2002; Vernon et al., 2008; Jones and Paulhus, 2010; McGregor et al., 2013a). Aggressive reactions to threats are approach-motivated (Carver and Harmon-Jones, 2009), and should accordingly focus one on tenacious personal goals with diminished regard for others' perspectives (Hogeveen et al., 2014; Sassenrath et al., 2014). Indeed, people with low agreeableness scores are mistrusting, devious, selfish, stubborn, arrogant, and callous (Costa and McCrae, 1991). Low self-control is also associated with delinquency, criminal behavior, physical and verbal aggression, self-directed aggression, and extreme and risky reactions to various environmental threats (Tagney et al., 2004; Nash, unpublished data; see also links with low conscientiousness and the dark personality traits, Jakobwitz and Egan, 2006). High scores on these intercorrelated and highly heritable oppositional traits should accordingly predispose people to the appeal of ARR, if other traits, threats, and affordances are also conducive (Costa and McCrae, 1991; Miles and Carey, 1997; Baker et al., 2008; Vernon et al., 2008; Beaver et al., 2009).

It is important to emphasize that these predisposing tendencies need not be in the abnormal range. The vast majority of ARR perpetrators have normal range traits. Our hypothesis is that even normal range tendencies should be enough to combine with the other factors to make ARR appealing. There are also several different kinds of ARR for which the conducive personality traits are likely differentially important. For example, lone-wolf perpetrators and leaders of ARR movements should be most likely to score highly on oppositional traits. In contrast, the foot soldiers and joiners of established movements may be less likely to be oppositional and social dominance oriented, and are more likely to be submissive authoritarians who participate in the oppositional tendencies by association (Son Hing et al., 2007).

#### Anxious

By anxious traits we mean those associated with a predisposition toward BIS activity that may or may not be reflected in consciously self-reported state anxiety (which can be attenuated by defenses and which is often out of sync with physiological indicators). The BIS generates early signals conducive to anxiety, vigilance, and caution in uncertain or conflict-laden circumstances, and people with anxious traits are more inclined than others toward these responses (Hirsh and Inzlicht, 2008; Proulx et al., 2012). Correlational studies show reliable links between anxious and aggressive traits (e.g., *r*s of 0.4 and 0.5 between neuroticism and measures of aggression and hostility in our recent sample of 299 culturally diverse college students; McGregor, unpublished data; see also Jakobwitz and Egan, 2006).

Anxious arousal also mediates defensive reactivity. After threats, if participants can misattribute their anxious arousal to a mundane external cause (e.g., a placebo, or uncomfortable room) they no longer react defensively to threats by rationalizing or becoming extreme (reviewed in Jonas et al., 2014). Anxietyrelated need for structure also mediates the effect of threats on worldview defense (Agroskin and Jonas, 2013). Anxietyrelated traits and states including felt uncertainty, trait neuroticism, uncertainty-aversion, need for structure, and sense of victimhood, also moderate defensively extreme lifestyle choice, worldview defense, violence, and religious zeal reactions to uncertainty, mortality, control deprivation, and relationship insecurity threats (Hirschberger et al., 2009; Juhl and Routledge, 2010; McGregor et al., 2010b, 2013b; Agroskin, unpublished doctoral thesis).

The interconnected and anxiety-linked constructs of low implicit self-esteem and relationship attachment insecurity (DeHart et al., 2006) similarly moderate distressed, aggressive, extreme, and worldview zealous reactions to failure, relationship, insecurity, and mortality threats (Mikulincer and Florian, 2000; McGregor and Marigold, 2003; Jordan et al., 2005; McGregor et al., 2005; McGregor and Jordan, 2007; Schmeichel et al., 2009; Laurin et al., 2014; Nash et al., 2014).

Recent advances in techniques for indirect assessment of *states* related to anxiety have also begun to implicate anxious distress in ARR reactions. A wide variety of threats that have caused ARR-related defenses in past research only inconsistently arouse self-reported anxiety. However, almost all of them have been shown to heighten electrical activity source-localized to the anterior cingulate cortex of the brain which has been associated with anxious distress and BIS activation (Proulx et al., 2012). These same threats also elevate self-reported anxious distress that is delayed or retrospective, presumably because delay or retrospection evades the defenses that can obscure self-reports of anxious distress immediately after threats (McGregor et al., 2001; Nash et al., 2011; Agroskin, unpublished doctoral thesis). Taken together in light of recent RAM theorizing (Jonas et al., 2014), the links between ARR-related phenomena and BIS-related states and traits suggest that anxious traits related to high BIS activity should predispose people towards using ARR to mask their distress.

High BIS personalities may also be drawn to ARR because of their discomfort with abstraction and their attraction to concrete engagement in low-level thoughts and goals. High BIS personalities feel energized and mobilized by immersion in details of concrete action and are averse to focus on abstract reasons for "why" which can cue distressing rumination. Indeed, concreteness manipulations (e.g., being randomly assigned to write about the "how" vs. "why" of various goals) not only relieve distress after anxious experiences (Watkins et al., 2008) but also causes high BIS participants to run harder and burn more calories on a treadmill endurance test, squeeze with more tenacity on a hand-gripper, persist to better performance in a speeded data-entry task, report higher eager excitement on a self-report questionnaire, to feel more optimism about personal goals, and show an increase in left frontal brain activity, characteristic of approach motivation (Tran et al., unpublished manuscript).

High BIS personalities may prefer the clarity of concrete action to the ambiguity of abstraction because focus on concrete steps regarding what to do alleviates the potentially bewildering tangle of uncertainties about values and identity, especially for identityweak people (Vallacher and Wegner, 1989; Baumeister, 1991). Developmentally, however, patience with uncertain abstraction may be required if one is to identify, simulate, and hone reliable personal values to identify with (Adorno et al., 1950; Erikson, 1959; Marcia, 1980). Values need to be test-driven in comparison with other existential options before they can be adopted as an autonomous and intrinsically motivating part of personal identity (Ryan and Deci, 2000; La Guardia, 2009). This uncertain discovery process is more likely to feel threatening for people with high BIS reactivity whose anxiety circuits are most easily overloaded by uncertainty (Hirsh and Inzlicht, 2008). For them, defenses that allow escape from anxious conflicts by either unmitigated approach or avoidance motivation would be appealing (cf; Marcia, 1980; Jonas et al., 2014; Hayes et al., 2015).

When anxious people choose the approach route, they should accordingly be especially drawn to concrete and immediate commitments that limit exposure to the anxious burdens of abstract selfhood (Baumeister, 1986). They should also be drawn to the fascist structure of ARR groups that can provide direction and a sense of meaning without requiring self-clarity. Indeed, when faced with mortality reminders, mildly depressed people were most likely to bolster meaning with jingoistic judgments (Simon et al., 1998). Concrete and active engagement with fascist groups should accordingly hit the motivational sweet spot for anxious people by allowing them to restore approach motivated states without having to think abstractly.

#### Identity-Weak

At any choice point, people can imagine multiple possibilities for action. Classic and contemporary theories of human choice and goal regulation hold that identity-strong people constrain the potential for uncertainty and conflict among imagined alternatives by using idealistic abstractions (i.e., highest values) for guidance (reviewed in McGregor, 2004, 2006b, 2007). Committed identification with high values can help guide choices and thereby allow people to function with confidence and efficacy through frustrating or uncertain circumstances (Lydon and Zanna, 1990; Kroger and Marcia, 2011). Put another way, clarity about how to be can effectively guide what to do. Committed values function like abstract goals and can also activate approach motivation directly and further relieve anxious distress by mere reflection (McGregor et al., 2001, Study 1; Creswell et al., 2006; McGregor, 2006a; McGregor et al., 2012b). Indeed, for highly meaning-seeking people, even brief reflection on personal values activates approach motivated states (neural, perceptual, and selfreport evidence in McGregor, unpublished data).

In contrast, identity-weak people who lack clear value identifications to guide action and relieve anxious distress are less able to cope constructively with frustrating circumstances. Indeed, identity-weak people (i.e., with low scores on a trait measure of idealism) reacted to mortality, control deprivation, failure, and relationship threats by becoming especially anxious, bewildered, and demotivated (McGregor and Marigold, 2003; Ferriday, unpublished master thesis; Prentice et al., unpublished data). Self-doubt also predicts materialistic reactions to uncertainty (Chang and Arkin, 2002). Moreover, when simultaneously confronted with multiple vulnerabilities and threats (life-dissatisfaction, goal frustration, mortality salience) identity-weak people become fatalistically withdrawn from personal goals, report being depressed, and report that they wish to live shorter lives (Hayes et al., 2015). In contrast, identity-strong people (i.e., high self-esteem, high scores on trait idealism) cope with anxious distress relatively easily by focusing on personal ideals about self-worth or value conviction that activate approach motivated states and thereby relieve the anxiety (Dodgson and Wood, 1998; McGregor and Marigold, 2003; McGregor, 2006a; McGregor et al., 2007, 2009a; Schmeichel et al., 2009; Nash et al., 2010; Schumann et al., 2014; see also McGregor et al., 2010a for evidence that experimentally priming ideals can also activate this process).

When high personal values (which are usually prosocial; Crocker et al., 2008; Schumann et al., 2014) are not available or salient, on the other hand, we propose that people will be more likely to revert to concrete, angry, controlling, and jingoistic foci. All can activate palliative approach motivation in a concrete way (Keltner et al., 2003; Carver and Harmon-Jones, 2009; McGregor, unpublished data). Participants with low scores on a dispositional idealism scale also reacted to social-order and relationshipsecurity threats by becoming especially bewildered, and by becoming extremely devoted to the concrete and social (angry, group-related) aspects of religious zeal but not the abstract personal. After threat they rated their personal goals in life as being more random and out of control, and reported more willingness to support war and die for their religious beliefs, more allegiance to their own religious group, more hostility and less openness to people with different religious beliefs, more confidence that God would give them power and take care of them, and more willingness to go to extremes for God (Ferriday, unpublished master thesis).

Both high self-esteem and idealism are significantly correlated with each other and with other constructs related to personal agency (McGregor et al., 2007, Study 2). People with low scores on either self-esteem or idealism report significantly lower power, self-control, drive, hope, and efficacy, and higher anxiety, depression, and rumination (McGregor, unpublished data). The low personal agency arising from difficulty with ideals is consistent with classic and contemporary theories of the selfregulatory role of ideals and values. Accordingly, when people lack personal agency they cleave to sources of group-based and external control, through heightened allegiance to active groups, external agencies, and religious authority (Fritsche et al., 2008, 2011, 2013; Kay et al., 2010; Landau et al., 2015; Stollberg et al., 2015).

On the surface it may seem perverse to argue that aggressive religious radicals are identity-weak when their rhetoric brims with moral certainty and megalomanic conviction. Theories of narcissism, however, hold that such overt entitlement and importance is a reaction to weak capacity for nuanced and viable ideals. The rigid narcissistic shell shields against anxious distress of a vulnerable and lost soul (Kohut, 1971; Kernberg, 1975; see also Kroger and Marcia, 2011, for related research on identity foreclosure). Indeed, reactive narcissistic rage, punitiveness, and callous disregard for others is especially prevalent among people whose entitled grandiosity is belied by vulnerability and shame (McGregor et al., 2005, 2013a; Krizan and Johar, 2015). From this perspective, ARR is a kind of narcissistic response that appeals to identity-weak people. They should be most vulnerable to the dubious and grandiose religious ideals that animate ARR because they have no clear opposing identity and are most in need of bolstering and escaping the problematic self. Accordingly, they should be most inclined both toward extremes of bolstering the self by joining ARR groups (Fritsche et al., 2013; Swann and Buhrmester, 2015), and de-individuated escaping of the self by cleaving to group authority (Postmes and Spears, 1998). Both processes would orient them toward group loyalty and angry, concrete action without pangs of conscience.

In sum, clear values provide resilience in the face of anxious distress. They aid instrumental coping by providing agile capacity for assessing, prioritizing, and adjusting goals. They also provide a ready focus for activating approach motivation and relieving anxiety. They thereby help people cope with life's frustrations, choices, and uncertainties in instrumental and palliative ways. Without authentic and confident personal value identifications, some other arbiter for making choices and soothing anxious distress is necessary. Rigid (concrete, literal) and extreme commitment to hostile, fascist, and authoritarian groups may be a default alternative for the dispositionally inclined.

# Threat

# External Threat

For people like those just described who are dispositionally reliant on external sources of agency, threats that undermine external sources of agency should make ARR especially appealing. When personal control is weak (i.e., for anxious and identityweak people) or unavailable, and a source of external agency (e.g., ingroup, god, government) is threatened, people should be inclined to switch allegiance to another seemingly viable source of external agency (Kay et al., 2008, 2010). Feeling excluded or alienated, or like one's ingroup was unfairly treated or humiliated, or one's country was out of control should make people with weak capacity for personal agency turn to other sources of external agency, such as ARR.

Preliminary support for this hypothesis comes from research showing that whereas high dispositional idealism predicts reactive *personal* agency (for personal goals and personal beliefs) after a personal agency threat (i.e., zeal for personal beliefs after a failure experience; McGregor et al., 2007, 2010b); low dispositional idealism predicts reactive allegiance to external religious agency after an external control threat (i.e., willingness to go to extremes for God and religious groups, and claims to derive strength and safety from a powerful God after threats to important relationships or to economic stability; Ferriday, unpublished master thesis). Based on these results and our extension of compensatory control and group-based control theories (Kay et al., 2010; Fritsche et al., 2013) we expect that because identity-weak people rely on external sources of agency in uncertain circumstances they should be especially bewildered and drawn to the external agency of ARR when their other external sources of agency are threatened.

Turning to ARR as a way to restore approach motivated relief from distress could be precipitated by various threats to institutional or relational sources of agency, such as war, sanctions, economic instability, high unemployment, system injustice, system incompetence, corruption, hypocrisy, relative deprivation, and cultural marginalization, ridicule, prejudice, ostracism, unfair social policies that relegate disadvantaged people to inescapable cycles of humiliation and hardship. Relational, domestic abuse, conflict, unfairness, boredom, or uncertainty could have a similar effect leaving people feeling overwhelmed and frustrated. External threats, whether societal or domestic should accordingly heighten the appeal of ARR. Consistent with this external-threat interpretation, the first three reasons suggested for the rise of IS relate to perceptions of systemic injustice against Muslims: (1) Shia (Iran, Iraq, Syria) oppression of Sunnis, (2) lack of confidence in governmental ability to protect social justice, and (3) the perception of a "Western-led onslaught" against Muslims by the West and their coalition (Barrett, 2014b).

# No Hope

If the present is grim and frustrations or chaos make the near future seem hopeless, one can still look to the distant future for redemption. Hope is an eager state, closely aligned with approach motivation, and is negatively correlated with anxiety (McGregor et al., 2012b). Indeed, when people are confronted with anxiety-inducing threats, they respond by exaggerating hopeful commitment in alternative domains at near and far temporal distances as a way to activate RAM and suppress anxiety (reviewed in Jonas et al., 2014). But if all temporal horizons for hope, near and far, seem blocked, then people simmer in impotent anxiety, get depressed, and disengage from life. The combination of present life-dissatisfaction, salience of near-future failure, and mortality salience (that undermines distant future hope) made participants became depressed, demotivated, disinterested in personal goals, and less interested in living (Hayes et al., 2015).

Such across-the board withdrawal from life goes against the primal human motivation to strive, live, and love, however, (Ryan and Deci, 2000). Accordingly, such no-hope predicaments could be expected to vacillate between depression and fatalistic withdrawal on the one hand, and anxiety arising from frustrated efforts to re-engage with life on the other. Such vacillation would be conducive to ARR because the withdrawal would make people more willing to forsake normal goals and even die, but the anxiety would also orient them toward active participation in extreme ARR as a source of group-based, external agency to activate RAM for relief. ARR could be even more alluring because its utopian elements transcend the frustrating limits of the hopeless temporal world (Cohen et al., 2011).

It is important to note that neither threat nor hopelessness is synonymous with objective personal failure, disadvantage, or low SES. Hopelessness can refer to the plight of a group one identifies with (Wohl et al., 2010). It can also refer to a sense of futility about reaching whatever standard one has for oneself, high or low. Indeed, hopelessness may become particularly acute when one has ostensibly succeeded according to normative standards, but still feels dissatisfied. Being wealthy and vocationally successful, for example, but still feeling ostracized, socially excluded, alienated, or meaningless might make one feel especially hopeless and cynical about the viability of culturally available opportunities for a good life. Similarly, even in affluent circumstances domestic dysfunction could trigger hopelessness about social goals.

#### Affordance

# Salient Narratives and Opportunities Conducive to

Immediate, Concrete Engagement with Active Groups The aforementioned combination of traits and threats should make RAM via concrete engagement in active groups particularly appealing. People intuitively turn to groups and consensus in anxious circumstances because doing so can help people either bolster or escape from their vulnerable or uncertain selves (e.g., Festinger et al., 1956; Solomon et al., 1991; Gardner et al., 2002; McGregor et al., 2005; Fritsche et al., 2008). Fusing one's personal identity with a group identity can powerfully validate and bolster confidence in personal agency, liberating sacrifice and risky extremes for the sake of the group (Swann and Buhrmester, 2015). Groups can also support de-individuated identification as an escape from personal selfhood, and thereby promote enactment of group norms and beliefs even if they are anti-normative (Postmes and Spears, 1998). Authoritarian groups may further facilitate conformity and obedience beyond personal inclinations (Martin and Hewstone, 2003). Cleaving to group norms as a way to escape self-uncertainty is a theme long identified by existential philosophers and classic developmental psychological perspectives as described in the introduction. Authoritarian groups allow one to follow concrete orders, focusing on the concrete topics of what to do and how to do it instead of being responsible for grappling with uncertain moral quandaries about why. Authoritarian groups that explicitly specify immediate engagement in concrete action should be especially appealing for anxious and identity-weak people (Vallacher and Wegner, 1989; Baumeister, 1991; Hogg and Adelman, 2013; Tran et al., unpublished manuscript).

Indeed, people with barriers to personal agency react to threats by cleaving specifically to active and agentic groups, presumably because the groups' agency can be internalized as a surrogate or splint for personal agency (Hogg and Adelman, 2013; Stollberg et al., 2015; cf. Landau et al., 2015). Consistent with this group-based agency view, short experimental manipulations of ingroup affirmation or outgroup derogation can activate approach motivated states (McGregor, unpublished data). Approach motivation is an eager, confident, and resilient state resistant to anxious distress (McGregor et al., 2005, 2012b; Drake and Myers, 2006; Nash et al., 2011).

Concrete group-identification allows anxious and identityweak people to escape the self in two ways—by eliminating the ambiguities of moral abstraction by immersing in concrete and immediate experience, and by splinting their problematic personal identity with the authoritarian group identity. The preference for active engagement is consistent with the finding that previous involvement in active groups, like soccer teams, service groups, or religious groups is a predictor of engagement in ARR that is surprisingly superficial in idealism beyond the hostile and extremely black and white morality of us = good vs. them = bad (Atran, 2010). Physical action is also more concrete than ideas and values, and so would further appeal to the anxious and identity-weak. IS training manuals that advise recruits on the minutia of concrete planning practicalities like bringing knee and elbow pads in readiness for immediate action may thus feel exhilarating for anxious and identity-weak people.

Concrete engagement, even in suicidal aggression could feel like a welcome adventure for oppositional, anxious, and identity-weak people, fused to or highly identified with the power of active groups (Baumeister, 1991; Hogg, 2014; Hayes et al., 2015; Swann and Buhrmester, 2015). This may explain the explicitly concrete declaration of Jihad against tyrants in the al-Qaeda terrorism manual that states (from Post, 2005):

*"The confrontation that we are calling for with the apostate regimes does not know Socratic debates, Platonic ideals, nor Aristotelian diplomacy. But it knows the dialog of bullets, the ideals of assassination, bombing and destruction, and the diplomacy of the cannon and the machine gun."*

The concrete appeal of ARR may gel when narratives rationalizing participation in ARR become consensually endorsed in already-fused buddy groups (Atran, 2010) with members searching for greater purpose and meaning in life (Atran, 2015). As described next, we propose that ARRcompelling narratives involve conspiracy theory grievances, exaggerated cosmic significance of participating in resistance, rationalization for hostile retribution, and perceived legitimacy of unfalsifiable truth claims (Armstrong, 2000, 2014; Barrett, 2014a; Ali, 2015).

#### Conspiracy Narratives

Experimentally manipulated threats related to anxiety (mortality salience, low power, low control, frustration, and uncertainty) increase belief in superstitions and conspiracy theories (reviewed in Landau et al., 2015). The prevailing explanation for such experimentally induced threat-to-conspiracy effects is that belief in externally controlling phenomena like conspiracy plots or superstitious agencies helps make the world seem less random and more orderly. Compensatory control theorists see the need for order as so fundamental that people are even willing to accept the existence of (delusional) enemies and evil forces if it can reduce anxiety by restoring a perception of order. Even sinister order is preferable to chaos (Sullivan et al., 2010).

Another possible interpretation of such anxiety-induced superstitious/conspiracy effects is that anxiety-related threats cause a BIS-induced anxious vigilance that makes people overnotice strange and malevolent possibilities, and to experience the world as *more* out of control and random (as found by Agroskin and Jonas, 2013). If anything could happen, then maybe superstitions, evil forces, and conspiracies are true. Moreover, the same combination of external threats and identityweak personality traits (e.g., low hope, low self-esteem, low efficacy, and low idealism) that result in persistent anxiety and depression (Hayes et al., 2015) also increase superstitious and supernatural belief (Prentice, unpublished master thesis; Prentice et al., unpublished data; Ferriday, unpublished master thesis). We are currently investigating whether this specific combination of personality traits and external threats will also predict belief in conspiracy theories, and if so whether this heightened belief is a palliative antidote to anxiety (following compensatory control theory), a symptom of anxiety (following the vigilant anxiety view), or both.

Whatever the mechanism, anxiety-related threats heighten paranoid belief in conspiracy theories and enemyship. Salience of consensually endorsed conspiracy theories should accordingly be an important risk factor for ARR. In the Islamist context, ubiquitous narratives about the hypocritical and callous abuses of the Great Satan West and/or Zionist or Shia forces rationalize Islamist ARR, and make it seem legitimate. Cultural endorsement reifies the conspiracies (e.g., Hardin and Higgins, 1996) and creates a ready focus for anxious vigilance aroused by hopeless circumstances, and a convenient scapegoat target to fight against with palliative zeal. Conspiracy narratives were similarly used by Christians to justify pogroms against Jews and Crusades against Muslims during historical periods where Christians were the ones experiencing cultural and existential threat (Durant, 1950; Armstrong, 2000, 2014). The prevalence and cultural acceptance of conspiracy theories should therefore be an affordance that promotes ARR.

#### Cosmic Moral Struggle

Once identified, the same combination of factors that makes conspiracy theories attractive can also create extreme moral conviction about polarized forces of good (us) vs. evil (them). An ARR perpetrator described his motivation for ARR as follows (from Sim, 2013):

*"Jihad is a war purely to enforce Allah's word...to protect Islam and Muslims...widen Islamic power... and spread kindness and the truth... It seemed that there was no effective way to effect change and to stop all the destruction except to wage jihad—a war between good and evil. When there is a war between good and evil, sinners will be afraid of the swords of the mujahidin. Thus, cruelty will be demolished gradually. My involvement with bombing has a personal purpose and hope. We bomb those whom we think of as the enemy and also in order to fulfill the obligation of jihad for Allah... On the battlefield, it will be seen who is good and who is evil ... Jihad ... is a pure war which is ordered by Allah. I always pray to be given the ability and opportunities to carry out my obligation to do jihad."*

As reviewed in Jonas et al. (2014) hundreds of experiments have demonstrated that anxiety-related threats make people go to moral extremes. The direction of the extremist urge under threat can be guided by salient cultural norms, whether antisocial or prosocial. Salient cultural or religious narratives that promote polarized views of goodness (us) vs. evil (them) and make every small action by an individual on behalf of the fight seem significant may thus provide especially attractive refuge (Pyszczynski et al., 2006; Kruglanski et al., 2014). Such oversimplified cosmic struggle narratives are like ultra-conspiracy theories with opportunity for actionoriented participation. Frustrations and conflicts on the ground become infused with universal significance, unencumbered by complicating details of concrete reality. Aggressive justice restoration becomes a simple and easily accessible ideology that the identity-weak can readily harbor in private imagination. For people seeking relief from temporal frustrations and uncertainties, exiting the temporal realm and concretely fighting for the sake of transcendent abstractions is intoxicating. Pure ideals and anger can activate sanguine, approach motivated states economically and anonymously in one's own mind (McGregor, 2007; Carver and Harmon-Jones, 2009). These oversimplified idealistic surrogates may be especially attractive to identityweak people who feel morally bereft. Indeed, idealistically bereft converts with little religious education often become among the most zealous perpetrators of ARR (Barrett, 2014a,b; Atran, 2015). For them, the utopian appeal of IS may be particularly alluring.

#### Violence Justification

Aggression is a primal reaction to frustration, especially if aggressive cues are salient (Berkowitz, 1989). This may be because anger is a powerful lever for approach-motivation (Carver and Harmon-Jones, 2009), and approach motivated states mute anxious distress (Corr, 2008; Nash et al., 2012). Imagining and enacting aggressive revenge should therefore feel rewarding in frustrating and anxious circumstances. Classic and new research on reactive aggression emphasizes that just as ambient cues moderate the direction of antisocial vs. prosocial ideology after threats, the ambient cues moderate the extent to which frustration causes aggression (Berkowitz, 1989; reviewed in Jonas et al., 2014). Salient narratives that prime and justify ARR are therefore essential for its reactive appeal. Enthusiasm for punitive justice may initially get provoked by outrage over ingroup suffering, and become amplified by conspiracy theories with cosmic significance that ignites angry eagerness to fight against the evil unfairness. Narratives that excuse ARR violence as holy, necessary, and just—'they deserve it' (e.g., as described in Sim, 2013) may be necessary to release this eager fury from normal social constraints against aggression. Indeed, group consensus can provide validation and legitimization for murderous activities that would otherwise be considered universally immoral and more stressful to participate in (Webber et al., 2013).

The same factors that make aggressive narratives appealing may also heighten willingness to subject oneself to risky commitments, reduce capacity for consideration of what one has to lose, and reduce capacity for awareness of others' suffering. Threat-induced RAM makes people take more risks and become insensitive to distress. Eager approach cues predominate, and the salience of aversive stimuli fade to seem less motivationally relevant (Nash, unpublished doctoral dissertation). Eager religious narratives bent around the psychological reward for martyrdom, and courageous aggression against infidels with callous disregard for their suffering would therefore resonate with the motivational configuration of ARR. The recent spate of Islamist ARR shares a common narrative originating from Wahabi/Salafist thought that justifies extreme violence against apostates and infidels (Barrett, 2014b). Such callous and violenceencouraging narratives have gained traction in radical sects of all major religious traditions during periods of cultural anxiety and hopelessness (Armstrong, 2000, 2014; cf. Ali, 2015).

#### Unfalsifiability

Supernatural authority also helps preserve dubious distortions and rationalizations that can make ARR seem legitimate. Anxiogenic threats cause anxious and identity-weak people to more strongly endorse transcendent religious explanations for personal well-being and world events (McGregor et al., 2010b, Study 1; Ferriday, unpublished master thesis). An advantage of supernatural religious narratives rooted in transcendent and historical authority beyond physical proof is that they can be unfalsifiably asserted without the inconvenience of evidence and without the possibility of conflict. They can be bolstered merely by appeals to traditional authority and social consensus. Accordingly, experimentally manipulated threats cause people to spontaneously frame their moral arguments as more unfalsifiable, and randomly assigned unfalsifiability causes arguments to become more extreme and aggressive against dissenters (Friesen et al., 2015). Religious unfalsifiability may therefore be an important affordance that both consolidates zeal and spurs confident and militant action for ARR (cf. Ali, 2015).

# Hypothesis Testing In the Laboratory

To test our multifactor hypothesis blocks of experiments could separately test the personality, threat, and affordance variables while holding the active ingredients from the other blocks constant. We would predict a three way interaction with highest endorsement of ARR among participants whose personalities are most oppositional, anxious, and identity-weak (with conducive threat and affordance circumstances primed for all). The best constellation of specific traits for each of the latent personality variables could be modeled with structural equation modeling. The ARR scenario participants would be asked to rate for the dependent variable would be plausibly tailored to the sample (as in Pyszczynski et al., 2006).

The threat block of experiments could experimentally manipulate externally imposed threat with mortality, relationship-distress, system-collapse, or injustice salience (vs. neutral) manipulations. Hopelessness could be manipulated by randomly assigning participants to write about a high effort commitment that did not pay off vs. one that did pay off (as in Hayes et al., 2015) or about an ongoing ingroup injustice with no hope vs. hope for abatement. Participants preselected with conducive personalities and primed with the affordance variables should show highest endorsement of ARR in the external threat/ hopeless circumstances condition.

To test the affordance block of variables, participants preselected for conducive personality traits and primed with the conducive threat variables would rate the appeal of six different randomly assigned versions of an ARR scenario. In five of the conditions only four of the affordance variables would be woven into the scenario. In the sixth condition, all five affordance variables would be present. ARR should be most prevalent in a sixth condition with all five affordance variables present.

Once each factor was established, a final study could test the entire model in a large 2 (personality) × 2 (threat) × 2 (affordance) between subjects design. Participants would be preselected based on combined high vs. low personality predisposition and the threat and affordance factors would be experimentally manipulated. Highest endorsement of ARR would be expected in the conducive personality/threat/affordance cell.

# Interview Studies

Biographical analysis should also reveal personality × threat × affordance factors among ARR perpetrators. Thousands of foreign fighters are returning home from participation in ARR, many who are disillusioned and might be willing to participate in research designed to help understand the phenomenon. They could be comprehensively tested and interviewed to assess the adequacy of the proposed multifactor hypotheses if researchers could get access to them (Sim, 2013; Sageman, 2014). Interviews could also be conducted in international hotbeds that produce inordinately high numbers of ARR recruits to test for prevalence of conducive factors (The Soufan Group, 2015).

## Real World

Interventions aimed at mitigating personality, threat, or affordance factors could be targeted or observed while controlling for the other two factors. Many aspects of personality trait dispositions are stable across time, but two aspects of ARR conducive personality might be amenable to intervention. Traits rooted in anxiety-related insecurities are affected by relationship responsiveness and security (e.g., Bokhorst et al., 2003; DeHart et al., 2006; Fearon et al., 2006). Policy or programs that provided access to affordable daycare or other resources to reduce stress for caregivers could help. Public education about importance of responsive connection and autonomy support for children and ill-effects of authoritarian parenting could also be promoted (Ryan and Deci, 2000).

A second aspect of ARR conducive personality amenable to intervention is identity-weakness. Provision of adequate structure to support youth's developing sense of efficacy and autonomy, and interventions that promote pro-social values and belonging in experientially engaging groups could help bolster authentic intrinsic motivation to stabilize identity and inoculate against ARR (Ryan and Deci, 2000; Walton and Cohen, 2007, 2011; cf. Cohen and Sherman, 2014). The same manipulations could also decrease anxious reactivity (Creswell et al., 2006).

A set of quasi-experimental and intervention studies could also focus on factors affecting threat variables. They could assess whether ARR varies with externally threatening and hopeless social conditions, e.g., exposure to economic disruption, sanctions, invasions, occupations, mass-displacements, wars, political instability, institutional corruption, natural disasters, racial profiling, police brutality, judicial unfairness, unemployment, social ostracism vs. integration, accessibility of social services and acculturation support. Intervention experiments could follow up by enriching one of two matched communities with social programs that build infrastructure and systems to help culturally vulnerable immigrants cope, feel welcome, and able to hope. ARR incidence could be compared before and after interventions as well as to incidence in matched communities not receiving the interventions.

Similar studies could assess prevalence of the affordance factors and ARR. Countering ARR narratives with accessible and credible alternatives has been identified as a crucial intervention point (The Soufan Group, 2013; Atran, 2015). Effects of providing opportunity for active engagement in meaningful group activities suffused by prosocial narratives could be tested. Non-radical religious engagement might be particularly effective. Religious devotion is a powerful and mutable human response to the pervasive appeal of consensual moral clarity in action. Anxious humans yearn for moral devotion. The best defense against narratives that support the darker mutations might be opportunity for participation in narratives that uphold the lighter ones, and intrinsic religiosity is universally prosocial (Jonas and Fischer, 2006; Armstrong, 2009; Schumann et al., 2014; Shariff et al., 2015). To be effective, the counter-narratives would need to equally address local grievances, and have as much opportunity for experiential linkage to active engagement, adventure, and intimate group involvement as ARR narratives do (The Soufan Group, 2013; Atran, 2015).

Intervention studies in countries where most young people go to school could test the effects of counter-ARR education. Comparative religion curricula could reveal the history of world religions and weakness of violent religious arguments. Objectively debunked conspiracy theories, such as The Protocols of the Elders of Zion, could also be exposed, and examples of inter-religious cooperation could be featured (political apologies for actual historical injustices could also help; Blatz et al., 2009). The history and psychology of extreme, rigid belief systems could also be taught, summarizing the now 100s of experiments showing that intuitive moral certainty can be defensive and is an unreliable index of objective truth. Developmental and historical aberrations toward zealous extremism and aggression could be revealed as such and the centrality of mystery, humility, and compassion as fundamental to religious piety across traditions could be taught (Armstrong, 2000, 2009).

Importantly, effects of these real-world interventions would need to be carefully evaluated. De-radicalization and riskreduction programs that try to mitigate threat and affordance factors for at-risk individuals are already in place in the United Kingdom, Indonesia, Singapore, Sweden, Netherlands, Spain, Denmark, and elsewhere. They already combine elements of integration and support for immigrants, community resilience resources, and religious education with tools to counter ARR propaganda. What is missing is systematic program evaluation of what works and what does not work (Qatar International Academy for Security Studies, 2010). We submit that our hypotheses, derived from psychological theory and research, could help guide interventions and program evaluations in the

# References


real world. Given their current uncertainty about root causes of ARR, it is not surprising that politicians are reluctant to wade into that theoretical murkiness. Strong-arm punishment is a safer strategy. The Conservative Prime Minister of Canada recently dismissed interest in root causes of ARR with the unapologetic quip "this is not a time to commit sociology" after the arrest of religious radicals who had plotted to blow up a Canadian train (Fitzpatrick, 2013). He then announced a new bombing campaign against IS, and pushed through new anti-terror laws with unprecedented violation of civil rights.

Our hypotheses predict that punitive and fairness-violating social policies are as likely to exacerbate as quell ARR. ARR groups are fueled by an "ideology of protest" rooted in local grievances about maltreatment (The Soufan Group, 2013; The Economist, 2015). Brutal crackdowns fuel this fire. Post (2005) concluded that religious violence is driven by the psychological appeal of displaced aggression, for people who feel oppressed and hopeless. He further emphasized how cultural narratives across generations breed deep ideological hatred. His view converges with ours, in its conclusion that the appeal of ARR is psychological, and should be countered with psychological, at least as much as military intelligence. Atran (2015) similarly identified a psychological motivation for ARR as primarily idealistic:

*Most have had no traditional religious education, and are often "born again" into a socially tight, ideologically narrow but worldspanning sense of religious mission... Violent extremism represents not the resurgence of traditional cultures, but their collapse, as young people unmoored from millennial traditions flail about in search of a social identity that gives personal significance and glory.*

He further notes the importance of action groups that provide opportunity to act with "deep conviction of moral virtue," and that "sacred value must be fought with other sacred values." "When, as now, the focus is on military solutions and police interdiction, matters have already gone way too far. If that focus remains, we lose the coming generation." We hope that the recent empirical advances in personality and social psychological knowledge about root causes of ARR together with new knowledge derived from the hypotheses described above will augment converging insights from other disciplines, and help make wisdom drawn from psychological science a credible option for leaders grappling with ARR.


Baumeister, R. F. (1991). *Escaping the Self*. New York, NY: Basic Books.


Rogers, C. R. (1951). *Client-Centered Therapy*. Cambridge: Riverside.


Sassenrath, C., Sassenberg, K., and Scholl, A. (2014). From a distance*...* the impact of approach and avoidance motivational orientation on perspective taking. *Soc. Psychol. Personal. Sci.* 5, 18–26. doi: 10.1177/1948550613486672


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2015 McGregor, Hayes and Prentice. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# **Reminders of behavioral disinhibition increase public conformity in the Asch paradigm and behavioral affiliation with ingroup members**

*Kees van den Bos 1,2 \*, E. A. Lind <sup>3</sup> , Jeroen Bommelé <sup>1</sup>† and Sebastian D. J. VandeVondele <sup>1</sup>*

*<sup>1</sup> Department of Social and Organizational Psychology, Utrecht University, Utrecht, Netherlands, <sup>2</sup> School of Law, Utrecht University, Utrecht, Netherlands, <sup>3</sup> Fuqua School of Business, Duke University, Durham, NC, USA*

#### *Edited by:*

*Eva Jonas, University of Salzburg, Austria*

#### *Reviewed by:*

*Joseph Hayes, York University, Canada Immo Fritsche, University of Leipzig, Germany*

#### *\*Correspondence:*

*Kees van den Bos, Department of Social and Organizational Psychology, Utrecht University, Heidelberglaan 1, 3584 CS Utrecht, Netherlands k.vandenbos@uu.nl*

*†Present address:*

*Jeroen Bommelé, IVO Addiction Research Institute, Rotterdam, Netherlands*

#### *Specialty section:*

*This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology*

*Received: 09 October 2014 Accepted: 02 June 2015 Published: 29 June 2015*

#### *Citation:*

*Van den Bos K, Lind EA, Bommelé J and VandeVondele SDJ (2015) Reminders of behavioral disinhibition increase public conformity in the Asch paradigm and behavioral affiliation with ingroup members. Front. Psychol. 6:837. doi: 10.3389/fpsyg.2015.00837* This paper argues that being in the Asch situation, where there is a felt need to conform to others' faulty behaviors, poses a social threat to people. Furthermore, participating in a psychology experiment in which you will have to interact with other participants might trigger sense-making processes. The paper proposes that these assumed threats or sense-making processes are likely to activate the behavioral inhibition system, making people respond in more inhibited ways than they normally would be inclined to do. As a result, people's tendency to affiliate behaviorally with persons who are similar to them can be inhibited. The implication is that lowering behavioral inhibition (by experimentally reminding people about having acted without behavioral inhibitions) should lead to more public conformity in the Asch situation and stronger behavioral affiliation with ingroup members than not being reminded about behavioral disinhibition. Findings of four experiments support this line of reasoning. These findings are discussed in terms of behavioral inhibition and behavioral affiliation. Alternative accounts of the data that focus on social belongingness threats and optimal distinctiveness are also considered.

**Keywords: threats, social interaction, behavioral disinhibition, affiliation, conformity, behavior, sense-making**

# **Introduction**

In the present paper we examine the dynamics of how people respond to potentially threatening social interactions. In particular, we focus on reactions in the Asch (1951, 1955, 1956) situation, in which there is pressure to conform to the faulty answers given by other research participants. We also examine responses in psychology experiments in which people expect to interact with other participants. We argue that these kinds of social interaction situations may contain threatening or sense-making aspects for research participants and that the social interaction threats or sensemaking processes may activate the behavioral inhibition system (BIS: Carver and White, 1994). We test behavioral implications of this line of reasoning by assessing how lowering behavioral inhibition by means of experimental manipulation may affect public conformity and behavioral affiliation with the other participants in the experiments. We ground our predictions by building on work on behavioral affiliation and associated literatures and aim to integrate these insights with our recently developed perspective on behavioral inhibition and disinhibition (Van den Bos and Lind, 2013).

# **Behavioral Affiliation**

Humans are social animals, creatures that are highly interactive with, and responsive to, other members of their species (Aronson, 1972). One of the core needs of humans is the need for affiliation. "Affiliation is the act of associating or interacting with one or more other people" (Leary, 2010, p. 865). Thus, people who want to affiliate seek to be in the company of others and want to interact with these other persons.

Theorists have suggested that being with or interacting with other people is a fundamental social behavior (e.g., Murray, 1938), especially when this involves individuals who are similar to us (Schachter, 1959). Therefore, peers or others who are similar to us serve a special function in fulfilling the need for affiliation and, more generally, people's pursuit of interpersonal connection (Sherif and Sherif, 1964; Schwarz, 1973; Wolf, 2008; Sundar et al., 2009; see also Erikson, 1968).

Peers are those individuals who share a similar or equal status, are usually of roughly the same age, and often have similar interests and backgrounds, bonded by the premise of sameness (Wolf, 2008). These individuals have a significant influence on the behaviors of people, especially when people are high in need for affiliation (Leary, 2010). For example, members inside peer groups learn to develop relationships with others in a social system. Furthermore, peers, and in particular ingroup peers, constitute important social referents for conveying customs and social norms (Clausen, 1968). Moreover, research suggests that peers exert stronger influence on what people do than do other important figures such as authorities. For example, a correlational study by Schwarz (1973) suggested that peers have stronger effects on inmates' behaviors than do prison authorities. An experiment on consumer attitudes by Sundar et al. (2009) showed that peer cues are generally more persuasive than are cues received from authorities or experts with high source credibility. Similarly, Harris (1995) concluded in her review article that peers may be more important for the socialization of children than are parents and other authorities.

With the phrase "The Social Animal," Aronson (1972) highlighted that we humans have a social nature and that we have a strong tendency to affiliate with others around us, including (and probably especially) with those who are similar to us (Schachter, 1959; see also Murray, 1938; Sherif and Sherif, 1964; Clausen, 1968; Erikson, 1968; McClelland, 1987; Baumeister and Leary, 1995; Wolf, 2008). As a result of this social quality of humans, people's behaviors tend to be influenced heavily by their social surroundings. In other words, a great many human behaviors are, at their core, socially oriented behaviors.

This does not mean, though, that socially oriented behaviors are always good or benign. In fact, the notion that our susceptibility to social influence can yield both positive and negative effects on what we do forms a central part of classic and modern social psychology. The potentially deleterious effect of peer pressure is well-known and includes instances where an individual feels directly or indirectly pressured into conforming with the group to make their behavior match that of their peers, even when conformity has a less than positive impact (Erikson, 1968; Sherif and Sherif, 1968). And in his APA-medal winning book Aronson (1972) clearly pointed out that social behaviors include not only prosocial behaviors (such as helping in bystander situations or fighting injustice), but also less benevolent behaviors (such as behaviors pertaining to prejudice, aggression, and conformity with wrong answers given in the Asch paradigm). The current paper aims to address both aspects of social influence.

A key feature of the argument we present here is that humans have a natural tendency to want to affiliate with people similar to them and that when confronted with threats people often want even more to affiliate with similar others. However, to fully understand people's responses to threats, we also need to consider those threats that result from those situations in which we have the feeling we do not belong in the group or in which our feelings of optimal distinctiveness (e.g., Brewer, 1991) are threatened. There is a truly substantial amount of research on these kinds of threats in social psychology. For example, people feel threatened if they are socially excluded (e.g., Eisenberger et al., 2003) and react with all sorts of defenses to social exclusion (see, e.g., Baumeister and Tice, 1990; Twenge et al., 2001; Abrams, 2005; DeWall and Baumeister, 2006; Bernstein et al., 2008; Lakin et al., 2008; Molden et al., 2009; Aydin et al., 2010; Gunther Moor et al., 2011; Riva et al., 2012; Schaafsma and Williams, 2012). Furthermore, Simon et al. (1997) have shown that a mortality salience threat can lead people to want to be similar or dissimilar to others depending on whether their optimal distinctiveness to others had been threatened (i.e., whether their uniqueness or their similarity to others had been threatened). Thus, many issues need to be considered to provide a complete picture of the need for affiliation and people's responses to threat.

Obviously, the current paper cannot address all aspects of responses to social threats. Therefore, based on notions such as peer group affiliation (Sherif and Sherif, 1964), affiliation motivation (McClelland, 1987), and the social animal (Aronson, 1972), the present paper notes that a core issue in classic and contemporary social psychology is trying to understand when people want to be involved with their fellow companions in their surroundings, and what different forms of behaviors people may engage in when they want to be involved with these peers. The current paper focuses on these issues by examining the effects of reminders of behavioral disinhibition on conforming and affiliating with peers.

# **Responding to Threats in Social Interactions**

One reason why people affiliate with others is to obtain relief from stressful or fearful situations (Hill, 1987). Thus, behavioral affiliation is a response often seen when people are responding to social threats. Furthermore, Schachter (1959) proposed that people who are uncertain about the nature of a situation and how they should react desire to affiliate with other people to find out (see also Leary, 2010). Therefore, following the literature on behavioral affiliation, we focus in the present paper on how people respond to threatening situations and situations in which they are at least somewhat uncertain as to how they should behave exactly.

We examine these issues by relying on recent insights that suggest that in many situations people can be surprised by what is happening and do not know how to respond to the situation at hand (see, e.g., Van den Bos et al., 2011b; Van den Bos, 2013; Van den Bos and Lind, 2013). We argue here that in these confusing situations the BIS will be activated such that people will inhibit behavioral action because they are seeking first to find out what is going on and what behavior is appropriate in the situation at hand. After people have made sense of the situation the inhibition system is deactivated and the behavioral activation system is turned on so that people can perform the behavior that they think is appropriate in the current situation (Van den Bos, 2013). We ask what implications this line of reasoning can have for our understanding of how people affiliate with and conform to peers or fellow research participants.

Asch (1951, 1955, 1956) showed that participants in his classic conformity experiments were trying to sort out what was going on in the experiments and why their fellow research participants suddenly gave wrong answers to objectively simple questions. Given that people devalue, dislike, and reject those who do not conform to their judgments, decisions, and behaviors (Schachter, 1951), people understandably conform to others' views (Cialdini et al., 1991; Leary, 2010). Furthermore, consider the situation of a participant entering the psychology laboratory in which they are told that they will have to interact with other participants. It is a well-known fact that people who do this are trying to sort out what is going on in the experiment in which they are participating and to make sense of the situation in which they now find themselves, in particular when they will have to interact with an experimenter and other participants in the experiment. As a result of these sense-making processes, research participants are susceptible to how they are evaluated by important persons present in the lab setting. These important others may include the experimenter (Cottrell et al., 1968; Cottrell, 1972) but may also include the participants' peers (Innes and Young, 1975).

We assume that the social threats encountered in the Asch situation as well as the more general sense-making processes triggered in psychology experiments in which you will have to participate with other participants inhibits your reactions. Our assumption is based in part on the insight that evaluation apprehension involves anxiety (Christensen, 1982) and fear of negative evaluation (Rosenberg, 1980), which are concepts that are related to the activation of the BIS (Gray, 1987; Gray and McNaughton, 2000). Asch (1951, 1955, 1956) showed that participants in his conformity experiments were trying to sort out what was going on in the experiments and why their fellow research participants suddenly gave wrong answers to objectively simple questions. Thus, in addition to anxiety and fear of negative evaluation, more general processes of sense-making play a role in how research participants act in (at least some) psychology experiments, particularly those experiments in which participants interact with others.

Here we acknowledge that there are different perspectives on the functioning of the BIS in the research literature (see, e.g., Latané and Nida, 1981; Gray, 1987; Monteith, 1993; Carver and White, 1994; Gable et al., 2000; Gray and McNaughton, 2000; Nigg, 2000; Sawyer and Behnke, 2002; Carver, 2005; Knyazev et al., 2006; Amodio et al., 2008). This noted, there is good evidence that the BIS is activated when people are faced with anxiety-triggering stimuli (e.g., Carver and White, 1994; Gray and McNaughton, 2000) or, more generally, with social situations that instigate processes of sense-making (e.g., Gable et al., 2000; Van den Bos, 2013). For example, Carver and White (1994) argue that the BIS regulates people's responses to anxiety-related cues and inhibits behavior that can lead to negative or painful consequences. Furthermore, the BIS has also been used to explain self-regulation and inhibition of prejudiced responses (Monteith, 1993). Moreover, the BIS has also been linked to more general sense-making processes in social contexts, such as how people deal with novelty in their environments (Gable et al., 2000) or how they interpret and react to puzzling situations (Van den Bos et al., 2011b; Van den Bos, 2013).

Importantly, as explained in detail in Van den Bos and Lind (2013), our ideas about inhibition and disinhibition focus on behavioral (dis)inhibition in public contexts. We note that an important notion in social psychology is the idea that in public settings the presence of others can constrain people from following their personal inclinations. Thus, we argue that issues of *public* and *behavioral* inhibition are important elements in the psychology of inhibition and sense-making. *Public* because the inhibition of primary importance is often instigated by thoughts of what others will think of our actions in non-private and fundamentally social contexts, and *behavioral* because the main consequence of interest in our line of work will be the effects of inhibition on the behaviors that people subsequently show. In the current research we examine how this analysis may contribute to insights about when people affiliate with and conform to their fellow research participants.

# **The Current Research**

In the present paper we aim to combine the insights on conformity (Asch, 1951, 1955), behavioral affiliation (Schachter, 1959; Leary, 2010), and associated literatures (Murray, 1938; Sherif and Sherif, 1964; Clausen, 1968; Erikson, 1968; Aronson, 1972; McClelland, 1987; Wolf, 2008) with the idea that people try to make sense of their surroundings, including psychology experiments in which they are taking part with other participants (Cottrell et al., 1968; Rosenberg, 1980; Christensen, 1982; Geen, 1983, 1985; Van den Bos, 2013). Specifically, we attempt to integrate these insights with recent work that suggests that people in many social situations are inhibited from showing important social behaviors (Van den Bos, 2013). That is, we argue that if participants in psychology experiments in which they are expecting to interact with others indeed are inhibited from showing their social behaviors, as has been suggested in recent papers (Van den Bos et al., 2009, 2011b; Van den Bos, 2013), and if young people such as university students are indeed oriented toward their peers, as important scholars have argued (Schwarz, 1973; Harris, 1995; Sundar et al., 2009), then it should be the case that lowering behavioral inhibition will lead people to show increased affiliation with peers or others who are close or similar to them. Our previous research shows that behavioral inhibition can be lowered by reminding people of times in the past when they acted without inhibitions (Van den Bos et al., 2009, 2011b). Thus, reminding people of past disinhibited behaviors should lead them to affiliate more (not less) with their peers.

In four studies we examine the implications of this hypothesis on the actual behavior of research participants. To connect our research directly to the influence of social threats we focus in Studies 1 and 2 on people's behavior in theAsch (1951, 1955, 1956) paradigm. That is, in Studies 1 and 2 we argue that if reminders of behavioral disinhibition indeed lead people to affiliate more with their peers, they should be willing to conform more with what their peers do. Indeed, we reveal in Studies 1 and 2 that reminding people of having acted without inhibitions leads them to conform more (not less) with the wrong answers given by fellow research participants in the Asch paradigm.

We then use Studies 3 and 4 to generalize the effects of disinhibition to other measures of peer affiliation. In particular, in Studies 3 and 4 we note that increased affiliation with peers should be shown in university students wanting to sit closer to a fellow student from their university (cf. Macrae et al., 1994; Van den Bos et al., 2007). Indeed, in Study 3 we reveal that reminding university students of having acted without inhibitions leads them to sit closer to a fellow research participant, and not closer to the experimenter. Furthermore, in Study 4 we show that reminders of behavioral disinhibition lead students to sit closer to a student from their own university, and not closer to a student from a rival other university. Thus, taken together, our four studies reveal that reminders of behavioral disinhibition increase public conformity in the Asch paradigm and behavioral affiliation with ingroup members.

In all four studies we use a behavioral disinhibition manipulation that we developed and validated in earlier research (see Van den Bos et al., 2009, 2011b). Our manipulation asks participants in the disinhibition condition to answer three simple open-ended questions that remind them about their thoughts and feelings about having behaved without inhibitions. In the control condition participants answer similar questions that do not remind participants about disinhibited behaviors.

Van den Bos et al. (2009)showed that this way of reminding (vs. not reminding) participants of having acted without behavioral inhibitions successfully lowers behavioral inhibition as assessed by a state version of the popular and well-validated measure of BIS sensitivity by Carver and White (1994). Specifically, after completing the three disinhibition questions or the three control questions, participants completed the following seven state BIS items. Following Carver and White (1994) these items asked participants to indicate to what extent they agreed or disagreed with the following statements: "At this moment, I worry about making mistakes"; "At this moment, criticism or scolding would hurt me quite a bit"; "At this moment, I would feel pretty worried or upset when I think or know somebody is angry at me"; "At this moment, I do not experience fear or nervousness, even when something bad is about to happen to me" (reverse coded); "At this moment, I would get pretty worked up when I would know that something unpleasant is going to happen"; "At this moment, I would feel worried when I would think I have done poorly at something"; "At this moment, I have very few fears compared to my friends" (reverse coded). All items were answered on 7-point scales (1 = *strongly disagree*, 7 = *strongly agree*). Reliability of the resulting state scale BIS scale was good (Cronbach's alpha = 0.76). Results reported in Van den Bos et al. (2009) showed that the disinhibition manipulation successfully lowered behavioral inhibition such that participants in the disinhibition condition experienced significantly lower levels of state behavioral inhibition than participants in the nodisinhibition condition.

Furthermore, the disinhibition manipulation yields effects comparable to differences on Carver and White's (1994) measure of trait BIS (see Van den Bos et al., 2011a,b). In addition, the manipulation does not trigger behavioral activation [no effects were found on state versions of Carver and White, 1994, behavioral activation scales (BAS)] nor does it influence positive or negative affective states [no effects were found on the positive and negative subsets of the Positive and Negative Affect Schedule (PANAS) by Watson et al., 1988; see Van den Bos et al., 2009, 2011b]. Moreover, the manipulation does not affect self-monitoring nor experienced accountability or selfawareness (Van den Bos et al., 2011b). Participants in studies using this manipulation typically indicate no suspicion of the procedures employed during the disinhibition manipulation nor do they suspect a direct relationship between the manipulation and their subsequent reactions in other parts of the experiments (Van den Bos et al., 2009, 2011b). Furthermore, the effects of the disinhibition manipulation can be found both among students in the psychology laboratory and in non-student samples outside the psychology laboratory (Van den Bos et al., 2009, 2011b). And in all the studies that have used this manipulation, gender did not interact with the effects of the disinhibition manipulation. Gender also does not affect the findings we will report here.

Thus, the reminders of behavioral disinhibition that we use in our studies have been pretested extensively. These earlier tests show that this is a manipulation that is conceptually related to the BIS as defined by Carver and White (1994; see also Van den Bos, 2013), significantly lowers state behavioral inhibition (Van den Bos et al., 2009), yields comparable effects as associated individual difference variables (Van den Bos et al., 2011a,b), and does so without affecting alternative concepts such as behavioral activation, affective states, self-monitoring, or accountability (Van den Bos et al., 2009, 2011b). What effects do these reminders of behavioral disinhibition have on conformity and affiliation with peers?

# **Study 1**

In Studies 1 and 2 we examine whether reminding people of having acted without inhibitions lead them to conform more in public with the wrong answers given by other participants in the Asch (1951, 1955, 1956) paradigm. In Study 1, participants completed reminders of disinhibition or no disinhibition, after which they were asked to participate in a human perception task. In this task, participants were asked to indicate publicly which of three lines was equal in length to stimulus lines. In the condition in which confederates were present, four other supposed participants gave wrong answers in 10 critical trials. We assessed how many wrong answers the actual participants gave during the critical trials. In Study 1, we compared these responses with answers given in a condition where no confederates were present.

# **Method**

#### Participants and Design

Eighty-six students (31 men and 55 women) at Utrecht University participated in the study and were randomly assigned to one of the cells of the 2 (confederates: present vs. absent) *×* 2 (behavioral disinhibition: disinhibition vs. no disinhibition) factorial design.1,2 Participants received 3 Euros for their participation in the study.

# Procedure

The experiment was presented to the participants as consisting of two unrelated parts. In the first part, the disinhibition manipulation took place. This manipulation used the same procedures developed and extensively pretested in earlier research (for details, see Van den Bos et al., 2009, 2011a,b). Specifically, participants were asked to complete a short questionnaire of three open-ended questions. Participants in the disinhibition condition were instructed as follows:

The purpose of this questionnaire is to assess how people react to being disinhibited, that is, how people behave when they do not care about what others think of their reactions and what feelings they then experience. To this end, please complete the following three questions: Please briefly describe a situation out of your own life in which you acted without inhibitions. Please briefly describe how you behaved in the situation in which you acted without inhibitions. Please briefly describe the emotions that you experienced when you acted without inhibitions.

In the no-disinhibition condition participants completed a short questionnaire of three open-ended questions pertaining to public transportation. Specifically, participants in the nodisinhibition condition received the following instruction:

The purpose of this questionnaire is to assess how people react to using public transportation, that is, how people behave when they use public transportation and what feelings they then experience. To this end, please complete the following three questions: Please briefly describe a situation out of your own life in which you used public transportation. Please briefly describe how you behaved in the situation in which you used public transportation. Please briefly describe the emotions that you experienced when you used public transportation.

After the disinhibition manipulation, participants were informed that the first part of the study had ended and that the second part now would begin. In this part, participants were asked to participate in a study on human perception. Based on the meta-review by Bond (2005), which shows that when three or more confederates are present the tendency to conform tends to be stable, participants in the condition in which confederates were present took part in this study together with four other participants (in reality confederates who were blind to conditions). In the condition in which confederates were absent there were no other participants.

Participants were presented a total of 17 sets of vertical lines, projected on a big white screen. Each set consisted of one stimulus line and three other lines (A, B, and C). To make our stimulus materials a bit different from the original Asch materials (which consisted of horizontal lines) we used vertical lines.<sup>3</sup> The stimulus line was presented at the top of the screen and the three other lines beneath the stimulus line. After the presentation of each set of lines, participants were asked to indicate out loud which of the three other lines was equal in length to the stimulus line.

In the condition in which confederates were present, three confederates first gave their answers, after which the actual participant gave his or her answer, followed by the answer of the last confederate. As in the original Asch experiment, the confederates started by answering a few questions correctly but eventually began providing incorrect responses. That is, during 7 of the 17 trials (Trials 1, 2, 5, 8, 11, 14, and 17) the confederates gave the correct answers. During the 10 other trials the confederates gave a uniformly wrong answer. Our dependent variable assessed how many wrong answers (0–10) the actual participants gave during the 10 critical trials.

At the end of the experiment, participants were thoroughly debriefed. During debriefing, participants indicated no suspicion of the procedures employed nor did they suspect a direct relationship between the disinhibition manipulation and their reactions in the perception study.

# **Results**

A 2 (confederates) *×* 2 (disinhibition) analysis of variance on our conformity measure (the number of wrong answers given by the participants during the critical trials) revealed a main effect of confederates being present or absent, *F*(1,82) = 62.39, *p <* 0.001, η 2 *<sup>p</sup>* = 0.43, a main effect of disinhibition, *F*(1,82) = 10.11, *p <* 0.01, η 2 *<sup>p</sup>* = 0.11, and a significant interaction between the confederates and disinhibition manipulations, *F*(1,82) = 8.28, *p <* 0.01, η 2 *<sup>p</sup>* = 0.09. **Figure 1** shows the effects together with the respective standard errors. In the condition in which confederates were present, participants gave more wrong answers when they had been reminded about disinhibited behavior (*M* = 4.75, SD = 3.19) than when they had not been reminded about disinhibited behavior (*M* = 2.35, SD = 2.16), *F*(1,84) = 6.34, *p <* 0.02, η 2 *<sup>p</sup>* = 0.07. In the condition in which confederates were absent, there was no significant effect of the disinhibition manipulation, *F*(1,84) = 0.03, *p* = 0.86, η 2 *<sup>p</sup>* = 0.00. Participants in this condition did not gave many wrong answers following

<sup>1</sup> In all studies of this paper, gender was proportionally distributed among conditions. Furthermore, gender did not interact with the hypotheses under consideration and hence was dropped from the analyses.

<sup>2</sup>We report all manipulations, all data exclusions, and all measures in our studies (Simmons et al., 2012), so we note that in Study 1, 14 extra participants took part in the experiment and were removed from the analyses reported: Three participants knew about the Asch experiments, one participant indicated suspicion about the experimental procedure used, eight participants had to omitted because faults in the experimental procedures were made when running these participants, and two participants from the no-disinhibition control condition were removed from the analyses because inspecting Cook's (1977) distance measure in our main analysis (Cohen et al., 2003) revealed that they showed a distance score of more than 3.50 SDs above the mean. We further note that after assessing conformity in Studies 1 and 2, and measuring distance in Studies 3 and 4, participants completed some questionnaires. The results of these questionnaire findings are not discussed here and are available on request.

<sup>3</sup>Debriefing indicated that participants of Studies 1 and 2 did not see a relationship between our studies and the original Asch (1951, 1955, 1956) experiments, with the exception of the participants mentioned in Footnotes 2 and 5 who were omitted from the analyses presented.

the presence of reminders of behavioral disinhibition (*M* = 0.48, SD = 0.77) or following the absence of these reminders (*M* = 0.36, SD = 0.70).<sup>4</sup>

In addition, please note that when participants had not been reminded about disinhibited behavior, they gave more wrong answers following the presence as opposed to absence of confederates, *F*(1,84) = 8.81, *p <* 0.01, η 2 *<sup>p</sup>* = 0.09. This replicated the original Asch finding. Furthermore, supporting our predictions, in the condition in which participants had been reminded about disinhibited behavior, the effect of confederates being present or absent was three times as large, *F*(1,84) = 37.37, *p <* 0.001, η 2 *<sup>p</sup>* = 0.31, than when participants had not been reminded about disinhibited behavior.

# **Study 2**

Study 1 reveals that reminding people of having acted without inhibitions lead them to conform more in public with the wrong answers given by other participants in the Asch (1951, 1955, 1956) paradigm. Reminders of disinhibition do not affect participants' line size perceptions when no other participants are present, suggesting that the effect of disinhibition reminders is not a perceptual phenomenon, but instigates increased conformity with peers.

Study 2 attempted to replicate the effect of reminders of behavioral disinhibition on conformity reactions. We did this in an experiment in which there were always three confederate participants present. After all, Study 1 showed that the effect of the disinhibition reminders was only there in the condition in which four confederates were present, and in Study 2 we wanted to see whether we could replicate the effects of the disinhibition reminders in the presence of only three confederates.

Study 2 also sought to refine our understanding of what it is in our disinhibition manipulation that causes the effect. Note that the disinhibition manipulation used in the first study reminds people of how they "react to being disinhibited, that is, how people behave when they do not care about what others think of their reactions." This is a rather general manipulation of public behavioral disinhibition (Van den Bos et al., 2009, 2011a,b). Of course, evidence for our line of reasoning would be stronger if we could specify what this manipulation entails in somewhat more detailed terms.

In a pilot study we determined that most of our participants (Utrecht University students), when asked to indicate what they did when they acted without concerns for others present in their situation, pointed out that they voiced their own opinions in the presence of others. Voicing of opinions is an important issue to people (e.g., Hirschman, 1970; Folger, 1977; Van den Bos, 1999). In Study 2, therefore, before participants took part in the Asch paradigm we exposed them to reminders of general behavioral disinhibition, no disinhibition, or reminders of voicing their own opinions without much concerns for others present. If our reminders of general behavioral disinhibition are predominantly about disinhibition pertaining to voicing one's own opinions, as our pilot study suggested, then the general disinhibition condition should yield about the same level of conformity as the condition in which people were reminded about disinhibited behaviors regarding voicing of their own opinions.

Building on Study 1, we again expected that the lowest levels of conformity would be shown in the absence of reminders of

<sup>4</sup>Controlling for heterogeneity using the Welch–Satterthwaite approach (Satterthwaite, 1946; Welch, 1947) yielded the same results: a significant interaction between the confederates and disinhibition manipulations, *F*(1,27.85) = 5.69, *p <* 0.03, indicating a significant effect of the disinhibition manipulation when confederates were present, *F*(1,25.30) = 6.62, *p <* 0.02, and a non-significant effect when confederates were absent, *F*(1,47.57) = 0.33, *p >* 0.56. Thus, heterogeneity does not affect the conclusions regarding Study 1. Furthermore, heterogeneity of variance was not an issue in the other experiments reported in this paper.

disinhibition. To make conforming a more attractive and easier to choose option we made the stimulus lines and the other lines more comparable to each other. Compared to Study 1 this should lead to more conformity in the no-disinhibition condition, hence providing a tougher test of our prediction that there should be more conformity in the disinhibition conditions (either general or voice disinhibition) than in the no-disinhibition condition.

# **Method**

#### Participants and Design

Sixty-two students (15 men and 47 women) at Utrecht University participated in the study and were randomly assigned to one of the conditions of the behavioral disinhibition manipulation (general disinhibition, voice disinhibition, no disinhibition).<sup>5</sup> Participants were paid 3 Euros for their participation.

#### Procedure

As in Study 1, the disinhibition manipulation took place in the first part of the study. The no-disinhibition condition was the same as in Study 1. The instructions in the "general disinhibition" condition were the same as those used in the disinhibition conditions of Study 1. In the condition in which participants were reminded about "disinhibition regarding voice," participants were instructed as follows:

The purpose of this questionnaire is to assess how people react to being disinhibited, that is, how people voice their own opinions in the presence of others such that they do not care about what others think of their reactions and what feelings they then experience. To this end, please complete the following three questions: Please briefly describe a situation out of your own life in which you felt no inhibitions to voice your own opinions. Please briefly describe how you behaved in the situation in which you voiced your own opinions without inhibitions. Please briefly describe the emotions that you experienced when you voiced your own opinions without inhibitions.

The second part of Study 2 was the same as in Study 1, with this time three confederate participants present in all conditions and (as in the original Asch experiment) the actual participant always being the last to answer which line resembled the stimulus line.

Participants were thoroughly debriefed at the end of the experiment. Again, participants indicated no suspicion of the procedures employed and did not suspect a direct relationship between the disinhibition manipulation and their reactions in the perception study.

### **Results**

An analysis of variance showed a significant effect of the disinhibition manipulation on our conformity measure (the number of wrong answers given by the participants during the critical trials), *F*(2,59) = 3.31, *p <* 0.05, η 2 *<sup>p</sup>* = 0.10. **Figure 2** shows the effect together with the respective standard errors. When participants had been reminded about general disinhibited behavior they conformed more with the wrong answers given by the confederate participants (*M* = 4.80, SD = 2.48) than when they had not been reminded about disinhibited behavior (*M* = 3.24, SD = 1.58), *F*(1,60) = 4.72, *p <* 0.04, η 2 *<sup>p</sup>* = 0.07. Furthermore, when participants had been reminded about disinhibition regarding voice they also conformed more (*M* = 4.86, SD = 2.71) than when they had not been reminded about disinhibited behavior, *F*(1,60) = 5.14, *p <* 0.03, η 2 *<sup>p</sup>* = 0.08. Conformity did not differ between the general disinhibition and voice disinhibition conditions, *F*(1,60) = 0.00, *p >* 0.91, η 2 *<sup>p</sup>* = 0.00.

# **Study 3**

Studies 1 and 2 focused on the dynamics of how people respond to threats in social interactions, in particular Asch experiments in which there is pressure to publicly conform with faulty answers of fellow participants. Study 1 demonstrated that reminding people about having acted without inhibitions lead them to conform more with the faulty answers their fellow research participants give. Study 2 replicated this effect and in addition demonstrates that similar levels of enhanced conformity are found following reminders of general behavioral disinhibition and following reminders of having voiced one's own opinions without inhibitions.

Our line of reasoning suggested that in social threatening situations people want to behaviorally affiliate with those who are similar to them (such as their fellow research participants) but can be inhibited in showing their behavioral affiliation tendencies. After all, both behavioral affiliation (e.g., Hill, 1987) and behavioral inhibition (e.g., Gray, 1987; Gray and McNaughton, 2000) are responses that people frequently show when responding to social threats. Studies 3 and 4 aim to generalize the effects of disinhibition reminders to direct measures of peer affiliation. In particular, the student participants in Studies 3 and 4 are told that they will take part in psychology experiments with other students and we assess behavioral affiliation with peers by measuring how close our participants will sit to fellow students (Macrae et al., 1994; Van den Bos et al., 2007).

Studies 3 and 4 also extend Studies 1 and 2 by focusing on participant reactions to taking part in psychology experiments in which they merely expect to interact with other people. After all, there are many instances in which people seek to interact with others that do not involve stressful or distressing circumstances (Leary, 2010) and that involve more general sensemaking processes (Van den Bos, 2013). The kind of psychology experiments on which we focus in Studies 3 and 4 may well resemble those situations that Schachter (1959) explored when noting that people who are uncertain about the nature of a situation will be motivated to affiliate with similar others to find out what to expect and how to behave in the new situation at hand.

In Study 3 we reminded our participants about times they had acted without inhibitions (disinhibition conditions) or how they act on normal days (no-disinhibition conditions). We did

<sup>5</sup> In Study 2, 14 extra participants took part and were removed from the analyses presented: Five participants knew about the Asch experiments, three participants had to omitted because faults in the experimental procedures were made when running these participants, five participants had difficulty understanding the questions asked to them, and one participant from the no-disinhibition control condition was removed from the analyses because inspecting Cook's distance measure in our main analysis indicated that this participant showed a distance score of more than 2.75 SDs above the mean.

this using the method of earlier studies that extensively pretested this manipulation of behavioral disinhibition (Van den Bos et al., 2009, 2011a,b). After this, in a separate part of the experiment, participants were brought to a big room where they saw a small desk and a row of seven chairs. Building on and extending the method used by Macrae et al. (1994; see also Van den Bos et al., 2007), the desk at the left was where the experimenter would sit and the chair on the right was where another participant would sit. The chair in which participants sat down was our dependent variable, providing an indication of how much participants wanted to be closer to the other participant or to the experimenter.

Thus, the dependent variable was the distance, in number of chairs, between the chair with the belongings on it and the chair that the participant chose to sit on. This task measures interpersonal social distance (see Holland et al., 2004). Indeed, physical and social distances have been shown to be conceptually related (Bar-Anan et al., 2007). If our hypothesis was true that behavioral disinhibition would lead participants to want to affiliate with their peers, then we should see that reminding our student participants of disinhibited behaviors would lead them to sit closer to the other participant. In other words, we should see that reminders of behavioral disinhibition should lead to behavioral affiliation with a peer, not with an authority such as an experimenter.

Another advantage of this experimental set-up was that it allowed us to assess *behavioral* affiliation. Social psychology has always been aware that it is important to show effects of its concepts on people's behavioral reactions (instead of only showing effects on cognitive responses, perceptions, affective reactions, or intentions), yet frequently our research does not provide behavioral data (Greenberg, 1987; Jones, 1998; Baumeister et al., 2007). Furthermore, from an applied point of view it is interesting to see whether just asking people to complete three questions that remind them of their disinhibited behaviors has behavioral consequences on where they sit down in a room.

# **Method**

### Participants and Design

Sixty students (17 men and 43 women) at Utrecht University were randomly assigned to either the disinhibition or no-disinhibition conditions. Participants received 3 Euros for their participation.

## Procedure

The experiment was presented to the participants as two separate studies. In the first study, the disinhibition manipulation was induced. The instructions in the disinhibition condition were the same as in Studies 1 and 2. Following earlier studies (Van den Bos et al., 2009, 2011a,b), participants in the no-disinhibition condition completed a short questionnaire of three open-ended questions pertaining to how they experience a normal day. Specifically, instructions were as follows:

The purpose of this questionnaire is to assess how people experience a normal day in their lives, that is, how people usually behave on a regular day and what feelings they then experience. To this end, please complete the following three questions: Please briefly describe a situation out of your own life in which you acted in a normal way like you do on a regular day. Please briefly describe how you behave when you act in a normal way like you do on a regular day. Please briefly describe the emotions that you experience when you act in a normal way on a regular day.

After the disinhibition manipulation the experimenter told participants that the first study had ended and that the second study now would begin. This second study would take place in another room across the hall. The experimenter, carrying some papers, escorted one participant at a time to this room. Upon entering the room, the participant saw a small desk and a row of seven chairs. The desk was placed at the left of the room. The experimenter put down his papers on the desk and pointed out that the desk was the place where he would sit during the second part of the study. On the right-hand chair hang a coat and below the chair there was a bag. The experimenter said: "You will participate in this study together with another student. You see the student is already there [pointing at the right-hand chair]. This student is now in the bathroom and will be back in a moment. Please seat yourself at one of the chairs and wait till the other student gets back. I will check how the other student is doing. After this, we will start the study." The experimenter then ostensibly started walking out of the room but did not actually leave the room until the participant had sat down on one of the chairs. In this way, the experimenter could assess in which chair the participant sat (1 = *immediately next to the other participant's chair*, 6 = *immediately next to the experimenter's desk*) and this constituted our dependent variable.

After participants sat down, they were thoroughly debriefed. Participants indicated no suspicion of the procedures employed. Furthermore, they did not suspect a direct relationship between the disinhibition manipulation in the first study and the chair in which they sat in the second study in which they participated.

# **Results**

An analysis of variance revealed a significant effect of the disinhibition manipulation on our distance measure (sitting close to or distant from the other participant), *F*(1,58) = 5.97, *p <* 0.02, η 2 *<sup>p</sup>* = 0.09. When participants had been reminded about disinhibited behavior they sat closer to the other participant (*M* = 2.07, SD = 0.83) than when they had not been reminded about disinhibited behavior (*M* = 2.87, SD = 1.59).

# **Study 4**

In accordance with our line of reasoning Study 3 reveals that reminding people of having acted without inhibitions lead them to sit closer to a fellow research participant. The reminders of behavioral disinhibition did not lead our student participants to sit closer to the experimenter. These findings suggest that disinhibition leads to behavioral affiliation with peers, not with authorities.

The aim of Study 4 was to replicate the finding that reminding people of having acted without inhibitions lead them to affiliate behaviorally with those who are close or similar to them, and not with those who are less similar to them. To this end, participants (students at Utrecht University) again first completed the reminders of disinhibition or no disinhibition, and then were asked to take a seat in a row of seven chairs. To rule out possible alternative explanations, there was no experimenter desk in Study 4 and we varied whose belongings were on the right-hand chair in the room: These belongings were said to be from another student at Utrecht University or were from a student from a rival university. If our hypothesis was true that behavioral disinhibition would lead to behavioral affiliation especially with similar people, then we should find that reminders of behavioral disinhibition will lead our participants to sit closer to the student from their own university, but not closer to the student from the rival university.

# **Method**

#### Participants and Design

Eighty students (25 men and 55 women) at Utrecht University were randomly assigned to one of the cells of the 2 (university affiliation of other student: same university vs. other university) *×* 2 (behavioral disinhibition: disinhibition vs. no disinhibition) factorial design. They received 3 Euros for their participation.

# Procedure

The experiment was presented as two separate studies. In the first study, the disinhibition manipulation was induced in the same way as in Study 3. After this, the first study ended and the second study began. Walking to the room in which the second study would take place, the experimenter informed participants that they would participate in the second study together with another participant. The university affiliation manipulation varied whether the experimenter told our Utrecht University participants that the other participant was from Utrecht University (ingroup affiliation condition) or was from Leiden University (outgroup affiliation condition). When entering the room, participants saw a row of seven chairs. On the right-hand chair hang a coat and below the chair there was a bag. The experimenter said: "As I told you, you will participate in this study together with the other student. This student will be back in a moment. Please seat yourself at one of the chairs and wait till the other student gets back." The chair in which participants sat down (1 = *immediately next to the other participant's chair*, 6 = *furthest away from the other participant's chair*) served as the dependent variable of Study 4.

After participants sat down, they were thoroughly debriefed. Participants indicated no suspicion of the procedures employed and did not suspect a direct relationship between the disinhibition manipulation and the chair on which they sat down.

# **Results**

A 2 (university affiliation of other student) *×* 2 (disinhibition) analysis of variance on the distance measure showed only a significant interaction effect between the university affiliation and disinhibition manipulations, *F*(1,76) = 5.39, *p <* 0.03, η 2 *<sup>p</sup>* = 0.07. **Figure 3** illustrates the effect together with the respective standard errors. When interacting with the student from their own university, participants sat closer to the other participant when they had been reminded about disinhibited behavior (*M* = 2.55, SD = 0.89) than when they had not been reminded about disinhibited behavior (*M* = 3.35, SD = 1.31), *F*(1,76) = 4.41, *p <* 0.04, η 2 *<sup>p</sup>* = 0.05. When interacting with the student from the other university, being reminded about disinhibited behavior (*M* = 3.55, SD = 1.32) or not being reminded about disinhibited behavior (*M* = 3.10, SD = 1.25) did not significantly affect where participants sat down, *F*(1,76) = 1.40, *p >* 0.24, η 2 *<sup>p</sup>* = 0.02.

In addition, it is worth mentioning that in the disinhibition condition participants sat closer to the student from the same university than from the other university, *F*(1,76) = 6.89, *p <* 0.02, η 2 *<sup>p</sup>* = 0.08. The effect of university affiliation was not statistically

significant in the no-disinhibition condition, *F*(1,76) = 0.43, *p >* 0.51, η 2 *<sup>p</sup>* = 0.01.

# **General Discussion**

Four experiments focused on the dynamics of how people respond to threats in social interactions. We examined this issue by means of Asch experiments in which there was pressure to publicly conform with faulty answers of fellow participants (Studies 1 and 2). We also studied behavioral affiliation by means of seating distance measures in psychology experiments in which people were expecting to interact with other participants (Studies 3 and 4). All studies demonstrate that reminding people of having acted without inhibitions leads them to show behaviors that are more oriented toward their peers, in these studies peers being fellow research participants who are similar or close to them.

In particular, Study 4 reveals that reminding people about having acted without inhibitions leads them to sit closer to fellow research participants from their own university, and not closer to those who are from a rival other university. This suggests that reminders of disinhibited behaviors lead people to affiliate behaviorally with people who are similar to them. This effect is in line with Study 3 in which we found that participants who had been reminded about behavioral disinhibition sat closer to a fellow research participant and not closer to the experimenter. Thus, Studies 3 and 4 suggest that disinhibited individuals want to affiliate with their peers, that is, that they want to be with those who are similar to them and not with those who have authority over them (Study 3) and not with those who are members from an outgroup (Study 4). Taken together, the current experiments reveal a pioneering finding that the disinhibited individual wants to affiliate behaviorally and conform his or her behaviors with those who are similar to them.

# **Possible Implications**

A noteworthy aspect of all our four experiments is that we obtained our effects on the actual behavior of our participants. In this way the four studies that we report contribute to pleas that social psychology should provide behavioral data (not just cognitive responses, perceptions, affective reactions, or intentions; see, e.g., Greenberg, 1987; Jones, 1998; Vohs et al., 2006; Baumeister et al., 2007). In addition, the behavioral effects we obtained on participants' behavior are especially remarkable, we think, because we obtained them using a manipulation that consisted only of completing three questions. The findings we report here reveal that this somewhat modest manipulation yields reliable and consistent effects on participants' actual behavior.

The findings in our first Asch experiment were obtained by contrasting reactions given in the presence of confederates who gave wrong answers with reactions given in the absence of those confederates. Importantly, the effects of our disinhibition manipulation were only found in the presence of confederates and hence only when pressure to conform to fellow research participants was high and not when this pressure was absent. Pressure to conform constitutes an important threat in social interactions and is an important reason why people affiliate with others (Hill, 1987), so this is one way in which we studied the dynamics of social threats in the current paper.

Following Schachter (1959) and others (e.g., Leary, 2010) we also examined the effects of reminders of behavioral disinhibition in situations in which we assumed that participants would be at least somewhat uncertain as to how they should behave exactly. We studied this issue in psychology experiments in which participants were expecting to interact with other participants. We note explicitly that we did not have conditions in those studies (nor in Study 2) in which the assumed social threats or uncertainties about how to behave were contrasted with conditions in which threats and/or uncertainties clearly were absent. Furthermore, the assumed threatening or uncertaintyprovoking aspects of these situations were not measured in our studies. One reason for this is because it may be very difficult to reliably measure perceived threat or uncertainties (see, e.g., Van den Bos, 2009; Van den Bos and Lind, 2009). This noted, we would applaud it when future research would include more control conditions as well as sensitive measures of threat or uncertainty that would show robust evidence for the line of reasoning proposed here, for example by means of moderation or mediation analyses.

When we inspected what participants wrote down when answering the questions that asked them about their disinhibited behaviors we found that they were describing situations in which they did not feel strong public constraints on their behaviors, such as when they were attending big dance parties or other events in which they felt they could do whatever they wanted to do without others constraining their behaviors (see also Van den Bos et al., 2009, 2011a,b).

Importantly, our disinhibition manipulation does not ask participants to think about situation in which they did not follow group norms, rather it is a more general manipulation that ask them to think back about situations in which they did not care about what others were thinking of their reactions. Thus, the manipulation is not a group-related manipulation *per se*. Furthermore, when we inspected what participants wrote down when answering the disinhibition questions we did not find strong evidence that participants thought about groups and their not following group norms. Thus, we do not think the disinhibition manipulation is strongly or directly related to group behavior or group norms. We think it is better viewed of as a manipulation of interpersonal disinhibited behavior, thus behavior against other people (not necessarily groups or group members).

Previous findings have shown that our disinhibition manipulation is conceptually related to the BIS (Carver and White, 1994; Van den Bos, 2013), significantly lowers state behavioral inhibition (Van den Bos et al., 2009), yields comparable effects to those of individual differences in trait behavioral inhibition (Van den Bos et al., 2011a), and does so without engendering experimenter demands or affecting alternative concepts such as behavioral activation, affective states, selfmonitoring, or accountability (Van den Bos et al., 2009, 2011a,b). Study 2 extends these findings by showing that one important component of the effect of disinhibition manipulations may have to do with people feeling free to voice their own opinions in public. The findings we present here, together with earlier research (Van den Bos et al., 2009, 2011a,b), suggest that reminders of behavioral disinhibition have conceptually meaningful and statistically significant effects on what people actually do.

In developing our ideas about behavioral disinhibition, we built our theorizing not only on work on the BIS as developed by Gray (1987; Gray and McNaughton, 2000) and Carver and White (1994), but also on the work on public inhibition as defined by Latané and Nida (1981). Latané and Nida (1981) note that in public settings the presence of others can restrain people from showing their personal inclinations. For example, in a bystander dilemma a person may want to engage in helping behavior but may be restrained from doing so because of the presence of others (bystanders) who are not helping. Similarly, we think that important elements in the psychology of inhibition and sensemaking involve the issues of *public* and *behavioral* inhibition. *Public* because the inhibition of primary importance seems often to be instigated by thoughts of what others will think of our actions, and *behavioral* because the main consequence of interest in our line of work are the effects on the behaviors that people subsequently show. The studies we presented here are in line with this public and behavioral perspective on disinhibition. For example, our Studies 1 and 2 reveal that reminders of behavioral disinhibition lead to more public behavioral conformity. These findings extend insights derived from Asch's classic experiments on public conformity and contradict common sense by revealing that it is the disinhibited participant who shows more conformity.

Earlier research has highlighted the pernicious effects of behavioral disinhibition (e.g., Newman et al., 2005) and depicted behavioral disinhibition as antisocial (Lilienfeld, 1992), psychopathological (Nigg, 2000), and a source of unwanted acts (Peters et al., 2006). Along the same lines, an important theme in moral and political philosophy has been that humans should refrain from disinhibited behavior and that it would be better for the greater good if people acted with more inhibition than they normally do (e.g., Kant, 1959). In contrast, our research program thus far has highlighted more benign effects of behavioral disinhibition. For example, we showed that following reminders of behavioral disinhibition people do not suffer from the usual bystander effects that limit helping (Van den Bos et al., 2009) and are more likely to resist advantageous but unfair outcomes (Van den Bos et al., 2011b).

Going beyond these insights, the current studies provide a more nuanced perspective on behavioral disinhibition. Yes, reminders of behavioral disinhibition can lead people to conform with faulty answers given by fellow research participants (Studies 1 and 2), but our work suggests that is it is not just the case that disinhibition provokes conformity. Rather, the link between disinhibition and conformity should be understood in light of the fact that following disinhibition reminders people want to affiliate with those who are close or similar to them (Studies 3 and 4). Thus, behavioral disinhibition is best not viewed as unequivocally bad (e.g., Kant, 1959) or antisocial (Lilienfeld, 1992), but rather should be viewed of as triggering peer-oriented responses. This can lead to benign effects, such as helping of peers in need (Van den Bos et al., 2009) or rejection of outcomes that are unfairly better than outcomes of peers (Van den Bos et al., 2011b), but also to conformity with faulty behaviors of those peers (Studies 1 and 2). Behavioral disinhibition as a trigger of increased peer affiliation yields a new, more precise, and more nuanced understanding of behavioral disinhibition than seen previously in the research literature (e.g., Lilienfeld, 1992; Nigg, 2000; Suler, 2004; Peters et al., 2006; Van den Bos et al., 2009, 2011b).

# **An Alternative Account of the Data: Belongingness Threat and Optimal Distinctiveness**

The empirical observation that reminders of behavioral disinhibition are related to more conformity and more group affiliation is an intriguing effect. In fact, we think that the counterintuitive quality may be part of what makes this effect so interesting. This noted, the exact psychological processes instigated by our reminders of behavioral disinhibition should be examined in more detail in future research. Although earlier evidence revealed that the disinhibition manipulation attenuates a state version of the Carver and White (1994) behavioral inhibition scale, and does not influence state versions of the Carver and White BAS and also does not have reliable effects on positive and negative affective states, selfmonitoring, accountability, or self-awareness (Van den Bos et al., 2009, 2011b), more insights into these issues is needed. After all, there are different conceptualizations and associated measures of behavioral inhibition and activation out there and these conceptualizations and measures may well yield better insight into the exact psychological processes triggered by our disinhibition manipulation.

An important possibility that needs to be examined carefully in future research is whether being reminded of one's own socially deviant behavior can constitute a threat to social belonging and optimal distinctiveness. After all, it could be argued that being reminded about not having cared about what others think of your reactions might lead participants to realize that there have been instances in which they acted too individualistically and did not pay enough attention to important social connections. Arguably, this might threaten the balance between people wanting to belong to important social groups and form meaningful social connections and their desire to be unique individuals who stand out a bit (but not too much) from other persons, other groups, and other social connections.

Optimal distinctiveness theory (e.g., Brewer, 1991) suggests that reminders of being individuated should increase affiliative needs and thus the enhance motivation to belong to social groups. There is certainly evidence for this effect, beginning with the work presented in Brewer's (1991) initial article on that prominent model. Thus, our reminders of behavioral disinhibition might in fact have disturbed the balance of optimal distinctiveness and might have instigated belonging threats to at least some of the participants. Viewed in this way the effects reported in the present article would conceptually replicate studies demonstrating that after reminders of possible social exclusion people show affiliative tendencies, such as mimicry or norm conformity (see, e.g., Leary, 2010).

Thus, optimal distinctiveness and social belongingness threats may provide important alternative accounts of the findings we presented here. After all, an intriguing aspect of the current findings is that if the disinhibition manipulation induced people to feel free to voice their own opinions in public why they did not stick to the correct answer in the Asch paradigm? On the contrary, the manipulation seemed to have caused people to behave in such a way as to show a great deal of concern about social evaluation and a great deal of caring about what others think of their reactions. In short, optimal distinctiveness and social belonging may constitute important alternative accounts for explaining the effects of our manipulation.

We assumed that people are naturally inclined to affiliate with others but their natural inclination to affiliate with others can be inhibited. We further argued that the disinhibition manipulation allows participants to break free of this inhibited state and follow their natural inclination to affiliate with others. Importantly, our disinhibition manipulation (which involves having people think of a time in which they did not care about what others were thinking of their reactions) might cause an important affiliation threat to participants, which may partly help to explain our results.

Exploring the psychology of peer relations as well as group psychology may also be important in this regard. These concepts may share important similarities but may also differ in important ways from each other and understanding the similarities and differences between these concepts and related issues such as affiliation and belonging (Leary, 2010) may help to better understand the effects presented in the present paper.

To conclude this section, we strongly advocate for future research that would show whether and how our disinhibition manipulation is related to processes of social belonging and optimal distinctiveness. Obtaining these kinds of findings would elucidate the psychological processes underlying the behavioral effects reported here and elsewhere (see, e.g.,Van den Bos et al., 2009, 2011a,b; Van den Bos and Lind, 2013). In earlier research we observed that our disinhibition manipulation successfully lowers a state version of the Carver and White (1994) BIS scale, so we also suggest that our manipulation is conceptually and empirically related to behavioral inhibition in social contexts as we defined it here and elsewhere (see, e.g., Van den Bos and Lind, 2013). More fine-grained insight into the psychological processes discussed here would add a significant contribution to theory building and would help to identify a fascinating field of research.

# **Other Possible Limitations**

The concept of approach motivation may also help to interpret the findings reported here. For example, Harmon-Jones et al. (2013) conceptualize approach motivation as the urge to move toward something. Perhaps this suggests that our results suggest that people who overcome inhibition show approach-related behavior such that in Studies 1 and 2 people approach the opinion of their peers and that in Studies 3 and 4 people approach peers, in particular ingroup members. Thus, future research may want to focus on how approach-related behavior might be involved in the results presented here. Different operationalizations of approach motivation (see, e.g., Coan and Allen, 2003; Harmon-Jones et al., 2013) may be relevant here.

Another issue that should be examined is whether behavioral inhibition and activation are independent of each other. Many social psychologists have good reasons to consider the BIS and BAS as constituting independent systems (e.g., Carver and White, 1994; Gable et al., 2000; Gray and McNaughton, 2000), but current cognitive psychologists also tend to focus on the interaction between the BIS and BAS (e.g., Knyazev et al., 2006).

Of course, the findings of Studies 1 and 2 are limited to public conformity, which is not the same as private conformity (see Asch, 1951, 1955, 1956). This noted, Studies 1 and 2 suggest that following reminders of behavioral disinhibition people actively affiliate more with those who are similar to them, and it is certainly possible that this desire for greater affiliation will affect private, as well as public, conformity. Future research is needed to examine the effects of behavioral disinhibition on private conformity as well as its effects on additional other-oriented reactions, including relevant cognitions, feelings, and other internalized responses.

Although we believe that the effects of reminders of behavioral disinhibition are better studied using a chain of experiments (Spencer et al., 2005), rather than by attempting to tap intervening variables that may disturb the effects of the reminders, we do want to note explicitly that future research is needed to examine relevant moderators and mediators of the processes suggested by our findings. For example, in earlier research we found that differences in social value orientations moderate the effects of reminders of behavioral disinhibition on reactions to being overpaid (Van den Bos et al., 2011b). In contrast, though, social value orientations do not moderate the influence of behavioral disinhibition on reactions to bystander situations (Van den Bos et al., 2009), moral dilemmas (Van den Bos et al., 2011a), or the findings we presented in this paper. It appears that some processes, such as responses to bystander situations, moral dilemmas, behavioral affiliation settings, and conformity are so robust that they are not moderated by social value orientations, while other reactions, such as responses to being overpaid or to experiencing other mixedmotive situations, are susceptible to the moderating influence of social value orientations.

Do the findings we presented here imply that disinhibited people will seldom or never be influenced by authorities, but rather only by peers? Of course not. Research clearly shows that authorities can have strong influence on what people do (see, e.g., Cottrell et al., 1968; Milgram, 1974; Tyler and Lind, 1992). But our findings do suggest that the disinhibited individual is more likely to affiliate with their peers than with authorities

**References**


(see, e.g., Study 3). Future research should examine under what conditions affiliation with authorities becomes more likely. Future research should also explore other antecedents of behavioral affiliation and conformity, such as physical similarity (Mackinnon et al., 2011) or being mimicked by others (Van Baaren et al., 2003).

# **Conclusion**

Building on and extending earlier work on behavioral inhibition (e.g., Latané and Nida, 1981; Carver and White, 1994; Gray and McNaughton, 2000) and behavioral disinhibition (e.g., Suler, 2004; Van den Bos et al., 2009, 2011a,b) the aim of this paper was to examine the dynamics of how people make sense of and respond behaviorally to threats in social interaction experiments. To this end, we delineated some important and unexplored effects of reminders of disinhibited behavior. In particular, we reasoned that reminders of behavioral disinhibition would want to affiliate with their peers more. Supporting this line of reasoning we found that reminders of disinhibition lead people to show more conformity with faulty answers given by their peers in the Asch paradigm (Studies 1 and 2). Our findings also revealed increased behavioral affiliation following reminders of behavioral disinhibition (Studies 3 and 4). These effects were obtained on actual behavior in both modern and classic experimental paradigms oriented toward the understanding of human behavior pertaining to public conformity (Asch, 1951, 1955, 1956) and behavioral affiliation (Macrae et al., 1994). Taken together, our studies portray the disinhibited individual as someone who in potentially threatening social interactions affiliates and conforms with his or her peers.

behavior? *Curr. Dir. Psychol. Sci.* 16, 396–403. doi: 10.1111/j.1745-6916.2007. 00051.x


eds S. T. Fiske, D. T. Gilbert, and G. Lindzey (Hoboken, NJ: Wiley), 864–897.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2015 Van den Bos, Lind, Bommelé and VandeVondele. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*