Automatic Evaluations and Exercising: Systematic Review and Implications for Future Research

The general purpose of this systematic review was to summarize, structure and evaluate the findings on automatic evaluations of exercising. Studies were eligible for inclusion if they reported measuring automatic evaluations of exercising with an implicit measure and assessed some kind of exercise variable. Fourteen nonexperimental and six experimental studies (out of a total N = 1,928) were identified and rated by two independent reviewers. The main study characteristics were extracted and the grade of evidence for each study evaluated. First, results revealed a large heterogeneity in the applied measures to assess automatic evaluations of exercising and the exercise variables. Generally, small to large-sized significant relations between automatic evaluations of exercising and exercise variables were identified in the vast majority of studies. The review offers a systematization of the various examined exercise variables and prompts to differentiate more carefully between actually observed exercise behavior (proximal exercise indicator) and associated physiological or psychological variables (distal exercise indicator). Second, a lack of transparent reported reflections on the differing theoretical basis leading to the use of specific implicit measures was observed. Implicit measures should be applied purposefully, taking into consideration the individual advantages or disadvantages of the measures. Third, 12 studies were rated as providing first-grade evidence (lowest grade of evidence), five represent second-grade and three were rated as third-grade evidence. There is a dramatic lack of experimental studies, which are essential for illustrating the cause-effect relation between automatic evaluations of exercising and exercise and investigating under which conditions automatic evaluations of exercising influence behavior. Conclusions about the necessity of exercise interventions targeted at the alteration of automatic evaluations of exercising should therefore not be drawn too hastily.

Introduction

Health behavior is not solely based on the reflective processing of information. It is also driven by automatic processes (Kahneman, 2003). This duality in the processing of information is theoretically described in dual-process theories (e.g., Strack and Deutsch, 2004; Gawronski and Bodenhausen, 2006; Evans and Stanovich, 2013) that distinguish between reflective (also termed “explicit, propositional”) and automatic (also termed “implicit, associative”) processing of information. Reflective processes are characterized by conscious deliberation on available information. Resulting reflective evaluations can be cognitively (e.g., “exercise is healthy”) or affectively (e.g., “exercise is fun”) shaped. Automatic evaluations as an output of automatic processes represent the affective, spontaneous and often unconscious reaction to a present stimulus or event. The reaction arises from the (often previously learned) mental associations between a target concept (e.g., exercise) and affective attributes (e.g., pleasant, tiring) and is based on the mental representation of one's core affective reactions (Ekkekakis, 2013) to the stimulus or the event. According to dual-process theories, behavioral decisions are the result of interactions between both processes and thus between automatic and reflective evaluations (Strack and Deutsch, 2004; Gawronski and Bodenhausen, 2006). Automatic evaluations can be, but are not necessarily, in agreement with the reflective evaluations of a given object (e.g., Friese et al., 2011). One often referred to dual-process model, the associative-propositional model (APE), offers theoretical assumptions about mutual influences of automatic and reflective evaluations (Gawronski and Bodenhausen, 2011). For example, when automatic and reflective evaluations are inconsistent the initial reflective evaluation can be resolved in order to avoid aversive feelings. Furthermore, dual-process theories provide a framework for possible relations between automatic evaluations and exercise behavior. For example, according to the reflective-impulsive model (RIM, Strack and Deutsch, 2004) automatically activated associations trigger behavioral schemata that contain a motivational orientation toward approaching or avoiding the respective behavior. Automatic evaluations can thus for example help in understanding the paradoxical phenomenon of nonexercising. Individuals often know that exercising is good for them (reflective cognitive evaluation) but might automatically feel bad about it and thus decide against it (Bluemke et al., 2010).

Theorizing on health behavior and health behavior change (e.g., Deci and Ryan, 1980; Prochaska and DiClemente, 1984; Ajzen, 1985; Bandura, 1986) focused for more than half a century on reflective processes and almost completely neglected the potential relevance of automatic antecedents of behavior. Major social-cognitive theories and the consequential health interventions assume that by providing information, intentions are changed, which will lead to respective changes in behavior. However, meta-analysis revealed only a weak relation between altered intentions and exercise behavior (d = 0.15; Rhodes and Dickau, 2012) and concluded that this relation has very low practical value. Others commented that moving away from the underlying social-cognitive “information processing paradigm” (Ekkekakis, 2017), which is based on the meta-theoretical assumption that human beings act as rational information collectors who reflectively arrive at their decisions, could offer the opportunity to re-establish a new way of looking at a well-known phenomenon (i.e., why a large proportion of people fail to adapt to related health behavior interventions).

Since the effects of the mentioned interventions are rather sobering, Ekkekakis (2017, p. 86) anticipated that exercise psychology might be “undergoing a transition to dual-process theoretical models for conceptualizing the mechanisms that shape behavioral decisions about participation or nonparticipation in exercise and physical activity” and proposed that exercise psychology is already within a meta-theoretical crisis. Marteau et al. (2012) offer a more applied perspective when acknowledging that health interventions that additionally target “automatic bases of behaviors may be more effective” (p. 1492) than interventions that address reflective processes solely. Following this line of argumentation Conroy and Berry (2017) consider automatic evaluations of exercise and physical activity as “potentially modifiable targets” (p. 236) for specific interventions.

Taking the described—possible—meta-theoretical transition into account, the first necessary step in order to better understand the potential of automatic process theorizing for exercise psychology is the systematic summary and evaluation of related, so far accumulated, empirical evidence.

A growing body of literature already underlines the relevance of automatic evaluations for health behaviors (Friese et al., 2011; Hagger, 2016) such as eating (Friese et al., 2008), smoking (Payne et al., 2007), alcohol intake (Houben and Wiers, 2008), and physical activity (Conroy et al., 2010). There is ample evidence that automatic evaluations are related to exercise decisions and exercise behavior as well (e.g., Bluemke et al., 2010; Brand and Schweizer, 2015). Until now there has been no systematic review on automatic evaluations of exercising (AEE) and relations to exercise behavior. This lack of systematically gained, summarized and evaluated findings hinders progress in the field.

The Focus of This Review

This review focuses research on AEE and exercise. AEE can be related to physiological or psychological correlates of actual exercise behavior (e.g., exercise-related decisions) or exercise behavior itself (e.g., exercise amounts, exercise adherence). For the purpose of this review, we will use henceforth the umbrella term exercise indicators when referring to the various different measures of exercise-related constructs.

There are a variety of different terms when speaking about AEE. While some researchers seem to prefer the term “implicit attitudes” (e.g., Calitri et al., 2009; Markland et al., 2015), others have used the term “affective associations” (e.g., Sala et al., 2016). While the synonymous use of these terms is generally non-admissible, in this review, we checked thoroughly that all included studies targeted the affective core of the assessed evaluations in order to subsume the identified evaluations as AEE.

Literature on automatic evaluations describes various implicit measures for the assessment of automatic evaluations (for an overview see Gawronski and De Houwer, 2014). Prominent examples of these implicit measures are the Implicit Association Test (IAT; Greenwald et al., 1998) and the Evaluative Priming Task (EP; Fazio et al., 1986). One common aspect of these methods is avoiding verbal or written self-reports (e.g., questionnaires) but rather employing computerized tools that indirectly assess AEE through within-person response time latencies, interpersonal reaction time differences and error rates. Resulting scores are interpreted as indices of the strength of associations between a target concept (“exercise”) and affective attributes (valence). This review includes an overview of the most commonly used methods to measure AEE. Their expedient use is evaluated and implications for future research discussed.

The aim of this systematic review is to summarize and structure existing findings on AEE and relations to exercise indicators. In order to appraise the magnitude of the impact of AEE on exercise indicators, effect sizes will be extracted from the identified studies where possible.

Methods

Search Strategy

A systematic literature search was carried out in May 2017 and conducted according to the PRISMA guidelines (Liberati et al., 2009). The following databases were used: PsycINFO, PSYNDEX, PsycARTICLES, SPORTDiscus and PubMed (see footnote¹ for the used search terms). The search included the articles' titles, abstracts and keywords. We exclusively searched for peer-reviewed articles. There was no restriction on year of publication. Additionally, the reference lists of the identified articles were screened for further articles.

Eligibility Criteria and Study Selection

The authors executed the initial screening of all retrieved studies on the basis of titles and abstracts independently from each other. The following inclusion and exclusion criteria were applied. Only original study reports in English or German were considered for integration in this review. Studies had to include an implicit assessment of AEE and some kind of exercise indicator to be eligible. Due to the use of diverse terms when referring to AEE we thoroughly checked if the assessed automatic evaluations in the identified studies fit to the described definition of automatic (also called “spontaneous,” “associative,” “uncontrolled,” or “implicit”) affective evaluations of exercising. Additionally, studies that did not examine exercise indicators were excluded. Exercise was defined as a “subset of physical activity that is planned, structured, and repetitive” (Caspersen et al., 1985, p. 126). Selected studies had to include human participants, be interventional or observational and gather cross-sectional or longitudinal data. Structured consultations between the two authors were carried out to eliminate disagreement regarding the eligibility of contentious studies. Afterwards, the two authors independently performed full-text reviews (based on the described inclusion and exclusion criteria) of all (pre-)selected articles. Finally, 19 records were included in further evaluation.

Referring to the proceeding of Rebar et al. (2016a), all studies' grades of evidence based on study design were evaluated. Cross-sectional studies were considered to provide first-grade evidence, prospective and longitudinal studies were taken to represent the second grade of evidence and experimental studies were considered to provide third-grade evidence. Ratings could have been lowered when risks of bias were present. Risks of bias were classified on study level following the guidelines from the Qualitative Assessment Tool for Quantitative Studies (Effective Public Health Practice Project, 2009). More specifically, the following sources of bias were evaluated: selection bias (when individuals selected to take part in the study were not likely to be representative of the target population), inadequate blinding of study participants (e.g., high risk of bias when participants were aware of the research question), unreliable data collection methods (e.g., high risk of bias when data relied on self-reports only), existence of confounders (e.g., high risk of bias when important differences between groups were present prior to manipulations), and insufficient reports of withdrawals and dropouts (for further information see Effective Public Health Practice Project, 2009). Only one rating was lowered from second to first-grade of evidence (Craeynest et al., 2008). Two cross-sectional studies showed risks of biases too (Berry et al., 2011; Sala et al., 2016). Since both studies were already rated as providing first-grade of evidence the rating was not adjusted.

Data Extraction and Synthesis

We summarized the specific characteristics of the studies separately for nonexperimental (see Table 1) and experimental (see Table 2) studies. After coding each study, great heterogeneity in the applied implicit and outcome measures became apparent. Due to the incomparability of the studies it was decided that it was impossible to conduct a meta-analysis (Ioannidis et al., 2008), so we carried out a systematic review instead. Nevertheless, we report significant effect sizes when available in the included studies. For reasons of clarity and comprehensibility we use Cohen's d as the effect size that expresses the degree of difference between groups (or means). Originally reported effect sizes (e.g., ${\eta }_{part}^2$ or η²) were transformed into Cohen's d where necessary (Lenhard and Lenhard, 2016). To express the degree of association between variables, the correlation coefficient r or the unstandardized and standardized regression coefficients b and β were used. The determination coefficient R² indicates the proportion of variance in one variable that is explained by another variable.

TABLE 1

Table 1. Nonexperimental studies investigating automatic evaluations of exercising and exercise indicators (k = 14).

TABLE 2

Table 2. Experimental studies investigating automatic evaluations of exercising and exercise indicators (k = 6).

Results

The literature search identified a total of 1,331 records (see Figure 1 for PRISMA flow chart). No additional studies were identified through records' reference lists. After excluding duplicates and records not meeting the described eligibility criteria by checking the titles and abstracts, 31 records remained for full-text analysis of eligibility. In total, 19 records, including one that was a multi-study report of two separate studies, fulfilled the eligibility criteria and were included in this review. Of these, six studies had an experimental design and 14 were nonexperimental (cross-sectional or prospective).

FIGURE 1

Figure 1. Study selection flow chart according to PRISMA statement (Moher et al., 2009).

Study Characteristics

Non-experimental Studies

From the included nonexperimental studies, 11 were cross-sectional and a further three had a prospective design, ranging from 3.5 months to 1 year and three to four times the measurement. The overall sample size of all included nonexperimental studies was N = 1,157 participants, ranging from 19 to 188 participants per study. Studies either explored male and female participants (k = 13) or focused on female participants only (k = 1). The samples included university students (k = 7) or groups of center exercisers or exercise course participants (k = 2), aircraftsmen (k = 1), cancer patients (k = 1), or obese children and adolescents (k = 3; which were compared to normal-weight children and adolescents). Average age ranged from M = 12.79 (SD = 2.68) to M = 57.0 (SD = 11.01). Due to their nonexperimental design, the majority of studies were rated as first-grade evidence (k = 12). Two studies reached a second grade of evidence rating (see Tables 1, 2 for the rating of each study).

Experimental Studies

All experimental studies investigated the effects of different interventions on the alteration of AEE (k = 6). One study additionally examined the effect of altered AEE on subsequent exercise behavior. The design of the interventions was experimental (k = 4) or quasi-experimental (k = 2), ranging from two to five experimental conditions. All studies applied a single-session intervention to alter AEE. Interventions were diverse, using Evaluative Conditioning (EC; see Hofmann et al., 2010 for an overview; k = 2), video clips depicting exercising individuals (k = 1) or an exercise advertisement (k = 1), a text with targeted exercise-related information (k = 1) or a guided imagery intervention (k = 1). The overall sample size was 771 participants, ranging from 41 to 213 participants per study. The participants comprised undergraduate psychology students (k = 4), undergraduate sport students (k = 1) or under- and postgraduate students of different subjects (k = 1). Average age ranged from M = 19.06 (SD = 1.96) to M = 23.51 (SD = 4.36). Studies included either male and female participants (k = 5) or investigated female participants only (k = 1). Three studies were rated as second-grade evidence and a further three studies were rated as third-grade evidence.

AEE Measurement

A first finding of this review concerns an immense diversity in the applied methods to measure AEE. Most of the identified studies used variants of the IAT (k = 7), which measure the strength of associations between semantic concepts stored in the memory. Standard IATs (k = 4; Berry et al., 2011; Markland et al., 2015; Chevance et al., 2016; Endrighi et al., 2016) are lexical sorting tasks in which stimuli from the target (e.g., “exercise”) vs. a comparison category (e.g., “nonexercise”) and from two evaluative categories (e.g., “good” vs. “bad”) are mapped to the same computer key. The sorting task is easier for respondents when the concepts sharing the same response key are closely associated than when they are not. A person with negative AEE will respond more quickly when, for example, the words “exercise” and “bad” are mapped to the same key than when “exercise” and “good” are mapped to the same key. IAT scores are usually calculated using the difference between response times for related and unrelated word pairs (Greenwald et al., 2003). Most IATs used pictures of exercise equipment or people engaging in exercise to represent the category “exercise.” Therefore the depicted exercises differed in their intensity within and between the studies (e.g., running on a treadmill, stretching, performing resistance exercise, using an exercise ball; Markland et al., 2015). Astonishingly, all studies used different comparison categories (e.g., “couch potato,” “inactivity” or “not exercise”) represented by pictures of diverse activities (e.g. reading a book, resting, relaxing, watching television).

Two studies (Antoniewicz and Brand, 2016a; Brand and Antoniewicz, 2016) used the Single-Target IAT (ST-IAT; Bluemke and Friese, 2008), one study (Sala et al., 2016) used the Single-Category IAT (SC-IAT; Karpinski and Steinman, 2006) and another one (Antoniewicz and Brand, 2014) used the Brief IAT (BIAT; Sriram and Greenwald, 2009). In comparison to the original IAT, the ST-IAT, SC-IAT and BIAT emphasize the strength of associations between one focal target category (e.g., “exercise”) and its evaluation (“good” or “bad”) so that the participants only have to pay attention to one concept. In addition, the BIAT is an abbreviated version of the IAT and is thus timesaving. The two ST-IAT studies used the same eight pictures to represent “exercise” (two different gymnastic exercises, strength training, running, swimming, volleyball, soccer, and tennis). For the SC-IAT, words representing exercise in different intensities were used. For the BIAT, diverse exercise-related pictures represented the “exercise” category. The BIAT study and one of the ST-IAT studies used symbols representing smiling or frowning faces (so-called “emoticons”) to depict the evaluative categories “good” and “bad.”

Three studies (Craeynest et al., 2005, 2008; Calitri et al., 2009) applied the Extrinsic Affective Simon Task (EAST; De Houwer, 2003) to measure AEE. The EAST is conceptually related to the IAT but differs procedurally. In comparison to the IAT, participants respond to evaluative target words written in white and (exercise-specific) target words written in different colors (e.g., blue and green) by pressing one of two computer keys. Participants have to respond to white attribute words in terms of their valence whereas they have to ignore (the irrelevant) valence for colored words and respond only in terms of color. For example, participants are asked to respond with one key for “good” and “blue words” and with another for “bad” and “green words.” Participants with positive AEE perform more quickly when the target color of the (colored) exercise word and positive valence are sorted under the same response key. This effect relies on the congruence of the task-irrelevant valence of the target word and valence response. The intensity of exercises represented by the exercise words was diverse, ranging from walking to rowing and sprinting. Four studies (Eves et al., 2007; Bluemke et al., 2010; Brand and Schweizer, 2015; Antoniewicz and Brand, 2016a) applied an EP task, which assesses automatic evaluative associations through a basic procedure of sequential priming (Fazio et al., 1986). In this task, a prime stimulus (e.g., exercise word or control word) is briefly presented and followed by a positive or negative word as target stimulus (e.g., “pleasant,” “delightful” vs. “repulsive,” “disgusting”). Participants are asked to categorize positive target words to one key and negative target words to another. It is assumed that the presentation of the prime activates associated evaluative attributes. For people with positive AEE, the response to positive target words after an exercise prime is thus facilitated (in comparison to a control prime) and leads to faster response times. All these studies used exercise words representing various exercise intensities and, as the comparison or control category, words representing nonexercise activities, nonsense words, or food words. Three studies (Bluemke et al., 2010; Brand and Schweizer, 2015; Antoniewicz and Brand, 2016a) explicitly used exercise-specific positive and negative words (e.g., “relaxed” vs. “exhausted”) as target stimuli.

Another five studies deployed implicit methods that do not rely on reaction time differences. The Go/No-go Association Task (GNAT; Nosek and Banaji, 2001) was applied in three studies (Berry et al., 2013; Berry and Shields, 2014; Berry, 2016). The GNAT is utilizable for the assessment of associations involving a single target category. Participants are asked to show a go response, by pressing a key, to target stimuli (e.g., exercise words and positive words) and a no-go response, by not pressing a key, to distracter stimuli (e.g., negative words). Unlike the IAT, differences in error rates (analyzed as sensitivity scores) of the responses to target words (e.g., “exercise”) indicate the strength of association between the target concept and a related evaluation. Positive AEE are assumed to ease the accuracy in discriminating “exercise” and “good” from distracters. All three GNAT studies assessed AEE with the target category “exercise” and another “generic” category. To represent “exercise,” diverse words were used (e.g., “workout,” “active,” “sports”). Two studies (Karpen et al., 2012; Antoniewicz and Brand, 2014) made use of the Affect Misattribution Procedure (AMP; Payne et al., 2005). The AMP applies a sequential priming procedure to assess affective associations through evaluative responses (more pleasant or less pleasant). This method differs from the previously mentioned methods in several ways. The evaluated stimulus is an ambivalent Chinese character, which has to be classified as “more pleasant” or “less pleasant” than an average Chinese character. Prior to the Chinese character, exercise pictures or pictures from a “neutral” category are presented as primes. The evoked affect is misattributed to the Chinese ideograph, which leads to shifts in the “more pleasant/less pleasant” ratings. Primes were shown supraliminally (k = 1) or subliminally (k = 1). Karpen et al. (2012) used images of household items as “neutral” primes whereas Antoniewicz and Brand (2014) used gray rectangles as control primes.

AEE and Associated Exercise Indicators

The identified studies examined hypotheses on the relation between AEE and exercise indicators and checked for AEE differences between different groups (see Tables 1, 2 for an overview). As for the applied methods, again, a respectable heterogeneity of the investigated exercise variables became apparent. The assessed exercise indicators can be categorized into three different domains. One group of studies addressed rather proximal exercise indicators, which directly represent quantitative (e.g., exercise volumes, exercise attendance) and qualitative (e.g., preferred exercise-setting) aspects of exercise behavior. A second group of studies targeted rather physiological or psychological variables that are associated with exercise behavior or decisions. These exercise indicators are summarized as distal exercise indicators (e.g., BMI changes, situated decisions to exercise or not to exercise). A third group of studies can be identified that assessed differences in AEE in specific target groups (e.g., obese individuals vs. normal-weight individuals).

Nonexperimental Studies

Automatic evaluations of exercising and proximal exercise indicators

Within the nonexperimental studies targeting AEE and proximal exercise indicators, several studies focused on the differences between highly active and less active individuals. Bluemke et al. (2010) showed that “exercisers” hold more positive AEE than “nonexercisers” (d = 0.59). Likewise, Berry et al. (2011) found a marginally significant (p < 0.06) difference between AEE in highly active individuals (i.e., more positive AEE) and less active individuals (d = 0.81). Moreover, AEE were identified as predicting the self-reported frequency, the duration of typical exercise sessions and the amount of habitual exercise behavior per week. Calitri et al. (2009) revealed a significant, positive, small-sized correlation between AEE and self-reported exercise behavior in the past week (r = 0.22). The study of Eves et al. (2007) contributed to these findings as well and demonstrated that their participants' AEE were associated with running behavior in the following week (d = 0.63), with those not running in the subsequent week having negative AEE.

Among the proximal exercise indicators that have been investigated are qualitative aspects such as the preferred exercise-setting. Fitness center exercisers had more positive AEE of fitness center exercising than a likewise physically active comparison group that exercised at other settings (d = 0.59; Antoniewicz and Brand, 2014). Antoniewicz and Brand (2016b) assessed adherence to a 14-week exercise course not only on the quantitative level (i.e., amount of overall participation) but also on the qualitative level (i.e., consideration of individual participation patterns such as returning to the course after a missed session). The three resulting adherence groups (i.e., maintainers, early, and late dropouts) differed in their AEE (d = 0.54) already at the beginning of the exercise course and highlighted the predictive power of AEE for exercise adherence. Positive associations discriminated particularly well between later exercise course “maintainers,” “early dropouts” and “late dropouts” at the beginning of the exercise course.

Not all studies found the expected associations between AEE and proximal exercise indicators. In a longitudinal prospective study with female, previously diagnosed cancer patients, baseline AEE did not predict daily minutes of exercise after 2, 4, or 6 months. The authors argued that this lack of association might be due to generally positive AEE with limited variability in the study sample (Endrighi et al., 2016). Furthermore, AEE were not significantly associated with self-reported exercise behavior (Chevance et al., 2016) and self-reported frequency of moderate- and high-intensity activities in the past week (Eves et al., 2007).

Automatic evaluations of exercising and distal exercise indicators

Furthermore, distal exercise indicators and their association with AEE have been investigated. One of the first explorations of AEE showed that AEE influence people's visual attention to exercise cues (β = 0.29). Extremely negative and positive AEE led to elevated visual attention to exercise words (“U-shaped relation”; Calitri et al., 2009). Berry et al. (2011) targeted exercisers' self-schema and illustrated that people who identify themselves as “exercisers” had more positive AEE than “non-exerciser” schematics (d = 0.77). Brand and Schweizer (2015) showed that AEE influence situated decisions to exercise (β = 0.15). The more positive the AEE, the more likely people were to decide in favor of exercise in the face of a behavioral alternative. Additionally, it was shown that the tendency to decide for or against exercise predicts the habitual amount of exercise per week.

Chevance et al. (2016) showed that AEE (β = 0.25) incrementally predicted exercise behavior in obese adults over and above Theory of Planned Behavior (Ajzen, 1985; β = 0.38) variables. These findings were not present in the general population. The Theory of Planned Behavior, as one typical representative of the social-cognitive “information-processing paradigm” (Ekkekakis, 2017), involves reflective parameters like outcome expectancies and tries to explain intentions and resulting behavior from those variables. The authors concluded that AEE might be especially influential for obese individuals, which might be due to differences in self-regulation between obese and non-obese persons. Other studies explored correlations between AEE and reflective evaluations. Thereby some studies focused on reflective affective evaluations of exercise (e.g., “How pleasant is exercising in a fitness center for you?,” Antoniewicz and Brand, 2014) whereas others measured the relationship between AEE and cognitive components of reflective evaluations (e.g., “Exercising is: useless-useful, unnecessary-necessary, foolish-wise,” Karpen et al., 2012).

With regard to the association between AEE and reflective affective evaluations, Antoniewicz and Brand (2014) found no significant correlations in fitness center exercisers or the comparison group. Calitri et al. (2009) showed that neither reflective cognitive evaluations nor reflective affective evaluations interacted significantly with AEE in their association with self-reported exercise behavior in the past week. Moreover, AEE were not correlated with a combined measure of cognitive and affective components of reflective evaluations (Brand and Schweizer, 2015).

Some studies addressed discrepancies between AEE and reflective (affective and cognitive) evaluations of exercise. AEE and reflective evaluations alone did not significantly predict changes in self-perception and reflective evaluations after a self-perception manipulation. However, it was shown that self-beliefs (β = 0.43) and reflective evaluations of exercise (β = 0.35) were more strongly affected by a self-perception manipulation in individuals with larger discrepancies between AEE and reflective evaluations (Karpen et al., 2012). Brand and Antoniewicz (2016) advanced the idea of discrepant AEE and reflective affective evaluations. They developed combined scores for a more accurate description of the variable pairs' (AEE and reflective affective evaluation) sum and discrepancy. The sum of AEE and reflective affective evaluations predicted the actual exercise frequency of fitness club exercisers (b = 3.83, R² = 0.10), whereas the discrepancy between AEE and reflective affective evaluations predicted the self-reported aspired exercise frequency per week (b = 4.74, R² = 0.10) and the ratio of actual to aspired exercise frequency (b = −0.10, R² = 0.12). Large discrepancies were associated with high self-reported aspired exercise frequencies. The authors interpreted these inflated goals as being a result of mistrust in the negative AEE and thereupon asserted very positive reflective affective evaluations (i.e., “idealized evaluations”; Brand and Antoniewicz, 2016). Low discrepancies predicted success in achieving the aspired exercise frequency. Also, AEE were positively associated with the ratio of actual to aspired exercise frequency (r = 0.32).

Not all expected associations between AEE and distal exercise indicators were found. For example, AEE neither predicted the affective response during moderate exercise nor the immediate post-exercise affective responses (Sala et al., 2016). Additionally, AEE were not associated with intention to run (Eves et al., 2007) or exercise self-efficacy (Endrighi et al., 2016).

Manifestations of AEE were also examined in specific target groups. Two studies focused on AEE and potential impact on obesity. They showed that obese youngsters neither had more negative AEE nor more positive automatic evaluations of sedentary behaviors than nonobese control individuals (Craeynest et al., 2005). A longitudinal study with obese children and adolescents in an obesity treatment setting found that favorable AEE of high-intensity exercise were a predictor of positive BMI change after obesity treatment (β = 0.51; Craeynest et al., 2008). Additionally, a within-person change in AEE of moderate-intensity exercise was a predictor of BMI change after 1 year of treatment (β = 0.89). A decrease in self-reported bodyweight was associated with increasingly negative AEE. Due to the small sample size (n = 19) and contextual inferences (some participants completed the EAST during treatment sessions, other at home or at university) these rather unexpected results should be interpreted with caution.

Automatic evaluations of exercising in specific target groups

Included studies assessed their data in diverse samples (e.g., students, general population), which also differed in age and gender. Two previously mentioned studies explicitly investigated the effects of AEE in specific samples. Eves et al. (2007) examined the effects of AEE on brisk walking (p > 0.05) and running (i.e., intention to run, p > 0.05; running behavior, d = 0.63) together with other moderate exercises (p > 0.05) in military trainee aircraftsmen. Another study used female previously diagnosed cancer patients as participants. There were no significant effects of baseline AEE on exercise self-efficacy changes or exercise behavior after 2, 4, and 6 months (Endrighi et al., 2016). Unlike the studies above (Craeynest et al., 2005, 2008) these studies did not compare specific samples to another sample (e.g., the general population). Due to the very specific populations the main study results should not be generalized.

Experimental studies

Experimental studies examined the possibility of altering AEE. Two studies used a computerized EC task (Antoniewicz and Brand, 2016a). In this task, pictures of exercising (conditioned stimuli) were repeatedly paired with positive or negative picture stimuli (unconditioned stimuli). The EC task resulted in differences in AEE between the group that should acquire positive AEE and the control group (d = 0.77) but not between the group intended to acquire negative AEE and the control group. Additional analyses revealed that changes in the group that acquired positive AEE were mainly driven by changes in associative connections between exercising and negative associations, in contrast to a facilitation of the associative connection between exercising and positive associations.

Markland et al. (2015) successfully altered AEE through a guided imagery intervention: AEE were more positive in the exercise imagery group (d = 0.39) than in a control group that imagined preparing a meal. Independently from the experimental manipulation, Markland et al. (2015) found that frequent exercisers had more positive AEE than less frequent exercisers (d = 0.57) and that AEE significantly correlated with reflective affective evaluations (r = 0.32) and reflective cognitive evaluations (r = 0.24).

In another study, participants read targeted exercise information that was incompatible with their pretest-AEE or pretest-reflective (affective and cognitive) evaluations of exercise. Results indicated that participants with positive AEE in the pretest who read information targeting negative reflective cognitive evaluations (e.g., information about negative health effects of exercise) unexpectedly showed even higher positive AEE in the posttest (d = 0.71; Berry, 2016).

Two other studies explored the alteration of AEE by using short video clips. One study used a short sequence from the TV show the biggest Loser, which depicted a strenuous exercise bout. There was neither a significant difference in AEE in comparison to a control group, nor a significant within-subject effect (Berry et al., 2013). In a study by Berry and Shields (2014), participants watched a short health- or appearance-orientated exercise advertisement. There were neither group differences nor significant within-subject differences in AEE.

To date, only one study has investigated how altered AEE affect subsequent exercise behavior. Antoniewicz and Brand (2016a) showed that AEE altered by EC influenced the choice of exercise intensity on a bicycle ergometer (main effect: d = 1.28). The group that learned positive AEE selected significantly higher intensities than the control group (d = 0.88). There were no significant differences in selected intensities between the group that should acquire negative AEE and the control group.

Discussion

The main aim of this systematic review was to summarize and evaluate available research on AEE and exercising. Findings brought up small to large-sized correlations between AEE and exercise indicators in the vast majority of studies. Various implicit measures were used to assess AEE. Even when studies used the same measure, the specific application varied.

Implicit Measures to Assess AEE

To the best of our knowledge, so far five different implicit measures have been used to assess AEE in exercise-related studies. Although very different measures have been used and the temptation is high to hand out advice that one might be the “best” implicit measure for assessing AEE, it is neither possible nor advisable to recommend “a particular paradigm as the best” (Gawronski and De Houwer, 2014, p. 293). In order to decide which measurement tool to use, every researcher has to answer the questions (among others) what exactly should be measured, as well as how and when it should be assessed. Since each measure has characteristic features, the fit between the respective research question as well as the underlying theoretical assumptions and the implicit measure's specific procedure is decisive. For all of these questions, the identified studies can provide some, albeit no complete, answers.

In the identified studies the predominant implicit measure is the IAT and variants of it (used in 40% of exercise-related studies). This finding corresponds with the dissemination of implicit measures in social cognition research in general where standard IATs are used in nearly every second published study (Nosek et al., 2011). As described before, standard IATs require two opposing categories, i.e., the target category and one for comparison. The studies in this review deployed “couch potato,” “inactivity” and “not exercise” in the classical IAT, which nicely illustrates the difficulty in finding a clear conceptual opposite of “exercise” (Rebar et al., 2015). In order to maximize the conceptual overlap between the implicit measure and the research design (Gawronski and De Houwer, 2014), researchers interested in AEE should check carefully whether their research questions include a comparison of “exercising” with another behavior. If not, other implicit measures might be more suitable for the respective research question. Only very few authors directly stated in their studies why a specific measurement procedure was used. Brand and Schweizer (2015) used an EP task and argued that the underlying mechanism (i.e., spontaneous evaluation of a stimulus) was closely related to the task requirements in their dependent variable (i.e., spontaneous decisions for or against exercising). The fit between research question and used method was thus explicitly taken into account. Antoniewicz and Brand (2014) targeted the automatic characteristic in AEE. They applied an AMP with subliminal stimulus presentation in order to conclude more easily on the automatic basis of AEE.

Exercising itself is a very complex and diverse behavior, which has to be reflected in AEE assessment. This systematic review has enumerated many different kinds of stimuli that have been used to assess AEE. Some of these stimuli represent the diversity more, some less. Again, the fit between the behavior of interest and the selected stimuli representing this behavior has to be considered. Some studies applied a very narrow focus and selected stimuli representing one specific behavior (e.g., fitness center exercising) in order to explain this tangible behavior (Antoniewicz and Brand, 2014), others used a broad range of stimuli in order to explain exercising behavior in general (e.g., Bluemke et al., 2010; Berry et al., 2011; Brand and Schweizer, 2015). A good indication of the selection of stimuli is the set provided by Rebar et al. (2016b). They aimed to establish population-level evidence of the most common exercise stimuli and ranked the 20 most-named activities when asked to report words relevant to the term “exercise.”

When measuring AEE and exercise indicators, it is essential to bear the affective nature of AEE in mind. This focus led to the exclusion of some studies that assessed automatic associations in the context of exercising but highlighted aspects other than the affective one (e.g., health; Berry et al., 2016, and exercise importance; Forrest et al., 2016). This differentiation is especially important when correlating AEE with reflective evaluations that measure more (see Calitri et al., 2009; Brand and Schweizer, 2015) or fewer (see Karpen et al., 2012) affective components.

Moreover, decisions about how to measure AEE should be guided by careful theoretical deliberations on the basis of dual-process theories. For the included studies, surprisingly, only a limited amount of studies explicitly referred to a dual-process theory in the theory section. Even less studies described relevant theoretical assumptions made by the stated dual-process theory (e.g., Bluemke et al., 2010; Berry et al., 2011) or, beyond that, directly explained how the theoretical assumptions guided decisions on an appropriate method to measure AEE (e.g., Brand and Antoniewicz, 2016).

The question how to measure AEE even includes one more issue. When the appropriate measure and stimuli are selected, the gained AEE scores can be calculated (even within the same measure) in different ways. Social cognition research is growing and progressing fast, which, for example, has led to alternative scoring algorithms for IATs. The established IAT D-Score has been extended to the DW-Score (Chevance et al., 2017b) and the IP-Score (Rebar et al., 2015), which have already been tested in the context of physical activity.

Lastly, the question when to measure AEE has been handled very differently in the identified studies. Prospective or retrospective assessments and mean scores resulting from the assessment before and after exercising have been used. While none of these approaches is right or wrong per se, the implications from each study vary widely. The retrospective assessment after an exercise bout could thus differ from prospectively assessed ones, due to the triggered automatic associations of exercising. Sound theoretical deliberations should thus guide the decision on the measurement point, in order to prevent a blending of findings, which can a priori be expected to be very different.

The (in)stability of automatic evaluations (Gawronski et al., 2017), in the context of physical activity (Hyde et al., 2012) and exercise (Antoniewicz and Brand, 2016a), has been much debated. Gawronski et al. (2017) demonstrated that automatic evaluations are less stable (weighted average r = 0.54) than conceptually corresponding reflective affective evaluations (weighted average r = 0.75). This finding can have methodological reasons, but can also be theoretically expected. For example, APE (Gawronski and Bodenhausen, 2011) points out that automatic evaluations are activated and altered on the mere basis of feature similarity and spatiotemporal contiguity whereas reflective evaluations are validated on the basis of logical consistency. Again, clarity on the underlying theoretical foundations of the study and on the measured construct can explain the assessment of sometimes more transitory or more long-lasting associations.

In sum, many different measures have been used to assess AEE. Quoting Nosek et al. (2011), the last few years of research can be characterized as the “Age of Measurement because of a proliferation of measurement methods and research evidence demonstrating their practical value for predicting human behavior” (p. 152). We claim to choose carefully between the available measures. They cannot be treated as substitutable and come with particular advantages and disadvantages that could fit or not fit with the research aim.

AEE and Exercise Indicators

The current research was able to detect multiple associations between AEE and exercise indicators. As a first result of this systematic review, the examined exercise indicators were classified as proximal or distal in order to better systematize and evaluate the findings. Whereas, the applied proximal indicators are akin in some ways (e.g., assessment of exercise amounts), the distal exercise indicators we identified were very diverse, making it very difficult to relate the findings to each other.

Proximal Exercise Indicators

Quantitative aspects of exercising, such as the amount of exercise per week, have been examined repeatedly. In particular, the association between differently pronounced AEE in individuals with higher or lower exercise amounts has been confirmed in a number of studies. For the studies with similar exercise indicators, the seemingly comparable results need to be critically evaluated. In general, the literature underscores that frequent exercisers hold more positive AEE than less frequent exercisers and highlights the predictive power of these AEE. However, only a very limited number of studies used objective measures such as accelerometers to collect actual behavioral data. Since self-reports are susceptible to overreporting and recall bias (Duncan et al., 2001) and could be biased by specific self-concepts (especially when referring to exercising; Brewer et al., 1993), the collected data might not represent the actually executed exercise behavior. Although this inadequacy represents a systematic error within each study that does not necessarily influence the targeted association with AEE, the external validity and the comparability with other studies are limited. Using a more objective measurement for exercise behavior should become the rule rather than the exception in order to improve the precision and accuracy in future AEE research.

Taking a closer look at the different exercise intervals that have been associated with AEE, huge differences become apparent. While Antoniewicz and Brand (2016a) successfully demonstrated immediate behavioral differences after altering AEE, some studies used a time frame of 1 week (e.g., Calitri et al., 2009) and others referred to three- (Antoniewicz and Brand, 2016b) or 6-month periods (Endrighi et al., 2016). Even though not all of these studies detected the generally found significant relation between AEE and exercise behavior (Endrighi et al., 2016), AEE seem to be linked to both short- and long-term exercise behavior. These findings correspond with results from other research areas where automatic evaluations have successfully been used to alter immediate food choices (Hollands et al., 2011) or to explain weight gain over a year (Nederkoorn et al., 2010). However, due to the limited number of longitudinal studies included in this review the question whether AEE are better suited to predicting short- or long-term behavior can and should not be answered on the basis of current empirical evidence. Nevertheless, reflections on theoretical assumptions of dual-process theories can help to become an insight about the possible impact of AEE for long- and short-term exercise behavior. For example, in the APE model (Gawronski and Bodenhausen, 2011), AEE provide an evaluative basis (default-interventionist conception) for behavior. This could, on the one hand, explain the short-term approach and help to understand immediate decisions for or against exercising (Brand and Schweizer, 2015). On the other hand, the conception does not exclude the assumption that AEE can be associated with habitual or long-term behavior. Long-term behavior can be understood as repeated decisions to engage in exercise, which are, as this review has demonstrated, influenced by AEE. Repeated exercise experiences, such as exercising with a pleasurable feeling connected to it, can, according to learning theory, be understood as positive reinforcement that gradually changes or manifests AEE (Strack and Deutsch, 2004) and thus facilitates exercising.

A further point of discussion is the generally gained knowledge from most of the studies that assessed proximal exercise indicators. With the help of these studies we know how AEE differ between, for example, exercisers and nonexercisers. However, we do not know why AEE differ or under which conditions they are decisive for behavior execution. Another interesting question might be whether AEE vary or fluctuate among individuals whose exercise behavior is similar. Although the empirical findings do not yet provide answers to these questions, dual-process theories offer a wide range of explanations. Again, the APE model as an example (Gawronski and Bodenhausen, 2011) outlines specific operating principles and conditions for both processes. The operating conditions, for example, comprise issues such as intentionality, awareness, efficiency and controllability (the “four horsemen” of automaticity; Bargh, 1994) and describe exemplarily how the formation and expression of AEE takes place.

Another dual-process theory, the recently presented Affective-Reflective Theory (ART) of physical inactivity and exercise (Brand and Ekkekakis, 2017), describes under which conditions AEE have more or less impact on exercise behavior. One strength of ART is the explicit reference to the phenomenon of exercising and the unique bodily sensations related to it, which result in more or less pleasurable or displeasurable states during exercise. In the light of this theory, AEE are linked with action impulses of approaching or avoiding the bodily sensations associated with the behavior. As an operating condition, Brand and Ekkekakis (2017) designate the availability of self-control resources. Limited self-control resources (for an overview on the concept of self-control see Baumeister et al., 1998) would thus lead to a greater impact of AEE on exercise behavior. Some empirical findings might hint at this connection and outline the difficulties of individuals in adhering to an exercise regimen on days when self-control is lowered (Englert and Rummel, 2016), which could increase the impact of AEE. Additionally, it can be assumed that individuals with negative AEE require higher amounts of self-control in order to reach the same exercise behavior (e.g., amount of exercise per week) as those individuals with more positive AEE. Whereas, the exercise behavior is identical, the need to overcome negative AEE and corresponding behavioral schemata of avoiding is debilitating and may hinder to maintain the behavior over a longer period.

Distal Exercise Indicators

Distal exercise indicators comprise a great variety of variables (see Tables 1, 2). Among these, reflective affective evaluations and their interaction with AEE have been examined several times. Thereby studies either focused on correlations (e.g., Antoniewicz and Brand, 2014) or on the individual predictive power of AEE and reflective affective or cognitive evaluations (e.g., Karpen et al., 2012). Three studies did not find significant associations between AEE and reflective affective and/or cognitive evaluations (Calitri et al., 2009; Antoniewicz and Brand, 2014; Brand and Schweizer, 2015) while two studies found small-to-medium-sized associations between AEE and reflective affective and/or cognitive evaluations (Berry et al., 2013; Markland et al., 2015). The ambiguous findings can be explained by theoretical deliberations. As already mentioned dual-process theories postulate an interplay between AEE and reflective affective evaluations. Those interactions are, according to RIM (Strack and Deutsch, 2004) influenced by cognitive capacity, motivation, and the amount of attention. The APE posits that the interplay of automatic and reflective evaluations is mainly driven by the consistency or inconsistency of both evaluations (Gawronski and Bodenhausen, 2011). Further research is needed to evaluate the theoretical assumptions and examine how and under which conditions AEE and reflective evaluations interact.

Most of the studies assessed AEE and reflective affective evaluations distinctly. While this approach is necessary in order to get a better understanding of AEE and its unique impact on exercise behavior, it does not necessarily correspond to the theoretical assumptions of dual-process theories. According to dual-process theories, behavioral decisions are the result of interactions between both processes and thus between AEE and reflective affective evaluations (Strack and Deutsch, 2004; Gawronski and Bodenhausen, 2006). For example, according to APE, this interplay can either be described as a “bottom-up” influence (see Gawronski and Bodenhausen, 2011), in which automatic evaluations influence reflective evaluations, or as a “top-down” influence, which operates in the opposite way. One possible “bottom-up” influence could occur in the case of inconsistency between evoked AEE and consciously validated reflective affective evaluations. In order to avoid aversive feelings due to the resulting cognitive dissonance (Festinger, 1957), reflective affective evaluations would be adjusted. These kinds of adjustments should be taken into account when designing a study. Brand and Antoniewicz (2016) directly referred to this default-interventionist rationale and conducted a computerized sequential assessment of AEE and reflective affective evaluations, yielding dependent values for the two constructs and thus taking into account this pair of evaluations' temporal and functional relationship. “Top-down” influences can be characterized by processes of affirmation or negation. Negating the reflective evaluation “I dislike exercising” might strengthen the associative connection between “exercise” and “dislike” and translate into respective AEE.

In reference to ART (Brand and Ekkekakis, 2017), AEE and reflective affective evaluations can be understood to interact through reciprocal feedback. Those feedback loops are a prerequisite for learning and again highlight the default-interventionist connection between the two evaluations. Empirical examinations of the described mechanisms (such as by Berry, 2016) could help in understanding why AEE (and reflective affective evaluations) differ in some individuals and not in others.

In summary, the described state of evidence highlights the relevance of AEE for predicting proximal as well as distal exercise indicators. We want to emphasize that proximal and distal exercise indicators are very different things. Measuring people's intentions or exercise-related self-efficacy can and should not be mixed up with exercising behavior itself. It has to be treated as a variable that can, but does not necessarily have to, lead to exercising.

AEE Should Not Yet Be Targeted in Exercise Interventions

Some of the presented studies provide initial insights to the potential of targeting AEE in future exercise interventions (e.g., Calitri et al., 2009; Antoniewicz and Brand, 2014; Endrighi et al., 2016). However, before targeting AEE in exercise interventions, there has to be clarity on the causal connection between AEE and exercise behavior. Only six out of the 20 identified studies employed an experimental design and only one study addressed the AEE-exercise behavior link (Antoniewicz and Brand, 2016a), which would suggest a causal relation between AEE and exercising.

The general accessibility of AEE for interventions was addressed in five studies. These studies used different approaches to test the alterability of AEE: Either a theoretically driven or a more application-oriented approach was used. The theoretically driven approach (Antoniewicz and Brand, 2016a) used an EC task that, based on reflections of the formation of AEE in the APE model (Gawronski and Bodenhausen, 2006), systematically paired pictures of exercising (or nonexercising) with pictures eliciting positive or negative sensations. The encouraging result for possible interventions was the significant shift in AEE in the group acquiring positive AEE. Changes toward more negative AEE were not detected. In contrast to this approach, other experiments used materials such as advertisements or sequences from TV shows that might change AEE. Although the overall results are not consistent (e.g., Berry et al., 2013), an alterability of AEE by some of these techniques seems to be confirmed. Even though all experimental studies applied single-session interventions, they vary widely concerning contextual, procedural and temporal aspects. Whereas, the shortest manipulation included the presentation of 85 s long video clips (Berry and Shields, 2014), other manipulations took the participants several minutes to work on (Berry et al., 2013; Antoniewicz and Brand, 2016a). Longer manipulations, which might provide a bigger amount of associative learning possibilities, do not seem to be more successful (Berry et al., 2013) than rather short manipulations (Berry and Shields, 2014). Successful alterations of AEE could rather be due to the high density of the provided information (Antoniewicz and Brand, 2016a) or the personal involvement of the participants (Markland et al., 2015) in those studies. Replications of the described experiments may help to differentiate between effective and rather ineffective manipulations. It is important to note that there is no empirical evidence on the stability of the achieved changes. As the creation or modulation of associative links is based on the principle of contiguity, it is arguable how long-lasting the effects of these single-session interventions will be. Experimental designs that observe the sustainability of the manipulation effects or test the theoretical assumption that often co-activated mental representations lead to stronger AEE then singularly experienced contiguities, could serve as a basis for further deliberation about application in practice. More experimental evidence is urgently needed before we can ask for practical applications of the so far limited knowledge. The lack of research becomes particularly clear if one considers that only one experiment targeted the AEE-exercise behavior link. Antoniewicz and Brand (2016a) demonstrated that changes in AEE are connected to changes in actual exercise behavior. However, it is essential first of all to better understand what we are actually measuring with the implicit measures and what impact it has on the psychological system before considering whether, and how, we want to change that.

Unresolved Issues

This review provides the first systematic overview on the relation between AEE and exercise. While some questions concerning the relation can be answered, it is important to note limitations that require further research.

First, it should not go unmentioned that the results may represent a publication bias due to the preferential publication of statistically significant results in the last few decades. It has been noted that the selective publication of significant results represents a great threat to validity for meta-analyses and systematic reviews since the “published literature is systematically unrepresentative of the population of completed studies” (Rothstein et al., 2006, p. 1).

Second, the diverging use of related terms concerning the examined psychological construct (e.g., implicit vs. automatic vs. associative or attitudes vs. evaluations) and the inadequate distinction concerning the observed behavioral phenomenon (e.g., physical activity vs. exercise vs. sport) might have hindered the identification of all studies targeting the relation between AEE and exercise. In order to collect all studies fitting to our aimed-at research question, a large variety of (partly synonymous terms) was used. In the future, the extent of the use of different terms when referring to the same psychological construct should be reconsidered in order to avoid misunderstandings.

Third, the number of studies that achieved a grade-three evidence rating is very low (k = 3; 15% of all identified studies). This was predominantly driven by the considerable amount of studies that applied correlational designs. Correlational studies qualify for many differentiated conclusions and are doubtlessly necessary when starting to explore a new research area. However, in order to examine and understand the cause-effect relation, experimental designs are required. In general, the research field would profit from more experimental studies that target the mechanisms explaining the link between AEE and exercising and allow a causal connection between the two to be inferred.

Fourth, we started this review with the statement that AEE might help to understand the paradox phenomenon of nonexercising (despite the individual's reflective evaluation that exercising is e.g., healthy). Whereas, the review offers first insights on the relation between reflective evaluation and AEE (e.g., Calitri et al., 2009; Antoniewicz and Brand, 2014) and their respective impact on exercise decision and behavior (e.g., Brand and Schweizer, 2015), we want to point out that exercising and physical activity are two distinct behaviors that might be influenced by unique motivational factors (Biddle, 2011). Concluding from the provided findings on AEE and exercise behavior on the consequences for physical inactivity might be a shortsighted approach. None of the described studies directly assessed automatic evaluations of physical inactivity. We are aware of only one very recent study (Chevance et al., 2017a) employing two different SC-IATs to assess both, AEE and automatic evaluations toward sedentary behavior in obese individuals. They revealed that only AEE were related to exercise behavior, whereas automatic evaluations toward sedentary behavior did not predict exercising. These findings underline the necessity to understand exercising and sedentary behavior as two distinct behaviors with different motivational antecedents.

Conclusion

As a result of our systematic review, we conclude that AEE are relevant determinants of exercise behavior, and are deserving much more research attention than they are actually given. Far-reaching conclusions are difficult to draw because of the immense heterogeneity concerning the observed exercise indicators, the implicit measures used to assess AEE and the underlying dual-process theories. This is tolerable, bearing the early phase of AAE research in sport and exercise psychology in mind. However, we claim that this review and the concomitant reassurance of the empirical evidence should mark the end point of this explorative phase of research. In order to achieve progress, a revision of the theoretical basis is urgently needed. We previously referred to Ekkekakis (2017), who diagnosed exercise psychology as being in a meta-theoretical crisis. In order to accelerate the transition from the “information processing paradigm” to dual-process theoretical frameworks, dual-process theories have to be scrutinized more thoroughly. Hence, it is an immense deficit that only a limited number of studies sufficiently described the underlying theoretical deliberations. Moreover, short-sighted, selected imports of automatic variables into existing theories from the “information processing paradigm” (for example, adopting the IAT for the add-on measurement of implicit attitude within the framework of the Theory of Planned Behavior) might not go far enough and might thus not reflect the explanatory potential of dual-process theorizing. Before handpicking and integrating single parameters into established theories, more empirical evidence on the coherence and principles of the operation of automaticity is required. Consequently, as long as all these preconditions are not fulfilled, wide-ranging implications for exercise interventions should be postponed. Yet in sum, we fully agree that AEE constitute a worthwhile target for further basic research that might in the future, according to the findings, lead to practical implications.

Author Contributions

MS contributed to the conception and design of the review and performed the initial literature search. MS and FA independently rated the studies and extracted the key variables. MS extracted the effect-sizes and transformed them when necessary. All passages of the review have been written mutually by both authors.

Funding

We acknowledge the support of Deutsche Forschungsgemeinschaft (German Research Foundation) and Open Access Publication Fund of Potsdam University.

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.

Acknowledgments

We would like to thank Sinika Timme who helped to proof-read the text and Prof. Dr. Ralf Brand whose advises helped to improve the overall quality of the review.

Footnotes

1. ^The following search terms were used to identify the relevant articles: i. Exercis* OR Sport* OR physical activit*, ii. AND implicit OR automatic OR unconscious OR non-conscious OR associative OR impulsive AND, iii. Attitude* OR Evaluation*/Process* OR Cognition*/Attitude* OR Evaluation* AND “Dual process theor*”/Attitude* OR Evaluation*AND “Dual process model*”/Attitude* OR Evaluation*AND Process* OR Cognition*.

References

Ajzen, I. (1985). “From intentions to actions: a theory of planned behavior,” in Action Control: From Cognition to Behavior, eds J. Kuhl and J. Beckmann (Berlin; Heidelberg: Springer), 11–39.

Google Scholar

Antoniewicz, F., and Brand, R. (2014). Automatic evaluations and exercise setting preference in frequent exercisers. J. Sport Exerc. Psychol. 36, 631–636. doi: 10.1123/jsep.2014-0033

PubMed Abstract | CrossRef Full Text | Google Scholar

Antoniewicz, F., and Brand, R. (2016a). Dropping out or keeping up? Early-dropouts, late-dropouts and maintainers differ in their automatic evaluations of exercise already before a fourteen-week exercise course. Front. Psychol. 7:838. doi: 10.3389/fpsyg.2016.00838

PubMed Abstract | CrossRef Full Text | Google Scholar

Antoniewicz, F., and Brand, R. (2016b). Learning to like exercising: Evaluative conditioning changes automatic evaluations of exercising and influences subsequent exercising behavior. J. Sport Exerc. Psychol. 38, 138–148. doi: 10.1123/jsep.2015-0125

PubMed Abstract | CrossRef Full Text | Google Scholar

Bandura, A. (1986). Social Foundations of Thought and Action: A Social Cognitive Theory. Prentice-Hall Series in Social Learning Theory. Englewood Cliffs, NJ: Prentice-Hall, Inc.

Bargh, J. A. (1994). “The four horsemen of automaticity: awareness, intention, efficiency, and control in social cognition,” in Handbook of Social Cognition, 2nd Edn., Vol. 1–2, eds R. Wyer and T. Srull (Hillsdale, NJ: Lawrence Erlbaum Associates, Inc), 1–40.

Baumeister, R. F., Bratslavsky, E., Muraven, M., and Tice, D. M. (1998). Ego depletion: is the active self a limited resource? J. Pers. Soc. Psychol. 74, 1252–1265. doi: 10.1037/0022-3514.74.5.1252

PubMed Abstract | CrossRef Full Text | Google Scholar

Berry, T. R. (2016). Changes in implicit and explicit exercise-related attitudes after reading targeted exercise-related information. Psychol. Sport Exerc. 22, 273–278. doi: 10.1016/j.psychsport.2015.09.001

CrossRef Full Text | Google Scholar

Berry, T. R., McLeod, N. C., Pankratow, M., and Walker, J. (2013). Effects of Biggest Loser exercise depictions on exercise-related attitudes. Am. J. Health Behav. 37, 96–103. doi: 10.5993/AJHB.37.1.11

PubMed Abstract | CrossRef Full Text | Google Scholar

Berry, T. R., Rodgers, W. M., Markland, D., and Hall, C. R. (2016). Moderators of implicit–explicit exercise cognition concordance. J. Sport Exerc. Psychol. 38, 579–589. doi: 10.1123/jsep.2016-0174

PubMed Abstract | CrossRef Full Text | Google Scholar

Berry, T. R., and Shields, C. (2014). Source attribution and credibility of health and appearance exercise advertisements: relationship with implicit and explicit attitudes and intentions. J. Health Psychol. 19, 242–252. doi: 10.1177/1359105312468190

PubMed Abstract | CrossRef Full Text | Google Scholar

Berry, T. R., Spence, J. C., and Clark, M. E. (2011). Exercise is in! Implicit exercise and sedentary-lifestyle bias held by in-groups. J. Appl. Soc. Psychol. 41, 2985–2998. doi: 10.1111/j.1559-1816.2011.00857.x

CrossRef Full Text | Google Scholar

Biddle, S. J. H. (2011). Fit or sit? Is there a psychology of sedentary behaviour? Sport Exerc. Psychol. Rev. 7, 5–10.

Google Scholar

Bluemke, M., Brand, R., Schweizer, G., and Kahlert, D. (2010). Exercise might be good for me, but I don't feel good about it: Do automatic associations predict exercise behavior? J. Sport Exerc. Psychol. 32, 137–153. doi: 10.1123/jsep.32.2.137

PubMed Abstract | CrossRef Full Text | Google Scholar

Bluemke, M., and Friese, M. (2008). Reliability and validity of the Single-Target IAT (ST-IAT): assessing automatic affect towards multiple attitude objects. Eur. J. Soc. Psychol. 38, 977–997. doi: 10.1002/ejsp.487

CrossRef Full Text | Google Scholar

Brand, R., and Antoniewicz, F. (2016). Affective evaluations of exercising: the role of automatic-reflective evaluation discrepancy. J. Sport Exerc. Psychol. 38, 631–638. doi: 10.1123/jsep.2016-0171

PubMed Abstract | CrossRef Full Text | Google Scholar

Brand, R., and Ekkekakis, P. (2017). Affective-Reflective Theory of physical inactivity and exercise: foundations and preliminary evidence. German J. Exerc. Sport Res. 47, 1–11. doi: 10.1007/s12662-017-0477-9

CrossRef Full Text | Google Scholar

Brand, R., and Schweizer, G. (2015). Going to the gym or to the movies? Situated decisions as a functional link connecting reflective and automatic evaluations of exercise with exercising behavior. J. Sport Exerc. Psychol. 37, 63–73. doi: 10.1123/jsep.2014-0018

PubMed Abstract | CrossRef Full Text | Google Scholar

Brewer, B. W., Van Raalte, J. L., and Linder, D. E. (1993). Athletic identity: hercules' muscles or Achilles heel? Int. J. Sport Psychol. 24, 237–254.

Google Scholar

Calitri, R., Lowe, R., Eves, F. F., and Bennett, P. (2009). Associations between visual attention, implicit and explicit attitude and behaviour for physical activity. Psychol. Health 24, 1105–1123. doi: 10.1080/08870440802245306

PubMed Abstract | CrossRef Full Text | Google Scholar

Caspersen, C. J., Powell, K. E., and Christenson, G. M. (1985). Physical activity, exercise, and physical fitness: definitions and distinctions for health-related research. Public Health Rep. 100, 126–131.

PubMed Abstract | Google Scholar

Chevance, G., Caudroit, J., Henry, T., Guerin, P., Boiché, J., and Héraud, N. (2017a). Do implicit attitudes toward physical activity and sedentary behavior prospectively predict objective physical activity among persons with obesity? J. Behav. Med. doi: 10.1007/s10865-017-9881-8. [Epub ahead of print].

PubMed Abstract | CrossRef Full Text | Google Scholar

Chevance, G., Caudroit, J., Romain, A. J., and Boiché, J. (2016). The adoption of physical activity and eating behaviors among persons with obesity and in the general population: the role of implicit attitudes within the Theory of Planned Behavior. Psychol. Health Med. 22, 319–324. doi: 10.1080/13548506.2016.1159705

PubMed Abstract | CrossRef Full Text | Google Scholar

Chevance, G., Héraud, N., Guerrieri, A., Rebar, A., and Boiché, J. (2017b). Measuring implicit attitudes toward physical activity and sedentary behaviors: test-retest reliability of three scoring algorithms of the Implicit Association Test and Single Category-Implicit Association Test. Psychol. Sport Exerc. 31, 70–78. doi: 10.1016/j.psychsport.2017.04.007

CrossRef Full Text | Google Scholar

Conroy, D. E., and Berry, T. R. (2017). Automatic affective evaluations of physical activity. Exerc. Sport Sci. Rev. 45, 230–237. doi: 10.1249/JES.0000000000000120

PubMed Abstract | CrossRef Full Text | Google Scholar

Conroy, D. E., Hyde, A. L., Doerksen, S. E., and Ribeiro, N. F. (2010). Implicit attitudes and explicit motivation prospectively predict physical activity. Ann. Behav. Med. 39, 112–118. doi: 10.1007/s12160-010-9161-0

PubMed Abstract | CrossRef Full Text | Google Scholar

Craeynest, M., Crombez, G., Deforche, B., Tanghe, A., and De Bourdeaudhuij, I. (2008). The role of implicit attitudes towards food and physical activity in the treatment of youth obesity. Eat. Behav. 9, 41–51. doi: 10.1016/j.eatbeh.2007.03.002

PubMed Abstract | CrossRef Full Text | Google Scholar

Craeynest, M., Crombez, G., De Houwer, J., Deforche, B., Tanghe, A., and De Bourdeaudhuij, I. (2005). Explicit and implicit attitudes towards food and physical activity in childhood obesity. Behav. Res. Ther. 43, 1111–1120. doi: 10.1016/j.brat.2004.07.007

PubMed Abstract | CrossRef Full Text | Google Scholar

Deci, E. L., and Ryan, R. M. (1980). Self-determination theory: when mind mediates behavior. J. Mind Behav. 1, 33–43.

Google Scholar

De Houwer, J. (2003). The extrinsic affective simon task. Exp. Psychol. 50, 77–85. doi: 10.1026//1618-3169.50.2.77

PubMed Abstract | CrossRef Full Text | Google Scholar

Duncan, G. E., Sydeman, S. J., Perri, M. G., Limacher, M. C., and Martin, A. D. (2001). Can sedentary adults accurately recall the intensity of their physical activity? Prevent. Med. 33, 18–26. doi: 10.1006/pmed.2001.0847

PubMed Abstract | CrossRef Full Text | Google Scholar

Effective Public Health Practice Project (2009). Dictionary for the Effective Public Health Practice Project Quality Assessment Tool for Quantitative Studies. Available online at: http://www.ephpp.ca/tools.html (Accessed September 2016).

Ekkekakis, P. (2013). The Measurement of Affect, Mood, and Emotion. New York, NY: Cambridge University Press.

Google Scholar

Ekkekakis, P. (2017). People have feelings! Exercise psychology in paradigmatic transition. Curr. Opin. Psychol. 16, 84–88. doi: 10.1016/j.copsyc.2017.03.018

PubMed Abstract | CrossRef Full Text | Google Scholar

Endrighi, R., Basen-Engquist, K., Szeto, E., Perkins, H., Baum, G., Cox-Martin, M., et al. (2016). Self-reported and automatic cognitions are associated with exercise behavior in cancer survivors. Health Psychol. 35, 824–828. doi: 10.1037/hea0000385

PubMed Abstract | CrossRef Full Text | Google Scholar

Englert, C., and Rummel, J. (2016). I want to keep on exercising but I don't: the negative impact of momentary lacks of self-control on exercise adherence. Psychol. Sport Exerc. 26, 24–31. doi: 10.1016/j.psychsport.2016.06.001

CrossRef Full Text | Google Scholar

Evans, J. S., and Stanovich, K. E. (2013). Dual-process theories of higher cognition: advancing the debate. Perspect. Psychol. Sci. 8, 223–241. doi: 10.1177/1745691612460685

PubMed Abstract | CrossRef Full Text | Google Scholar

Eves, F. F., Scott, E. J., Hoppé, R., and French, D. P. (2007). Using the affective priming paradigm to explore the attitudes underlying walking behaviour. Br. J. Health Psychol. 12, 571–585. doi: 10.1348/135910706X153775

PubMed Abstract | CrossRef Full Text | Google Scholar

Fazio, R. H., Sanbonmatsu, D. M., Powell, M. C., and Kardes, F. R. (1986). On the automatic activation of attitudes. J. Pers. Soc. Psychol. 50, 229–238. doi: 10.1037/0022-3514.50.2.229

PubMed Abstract | CrossRef Full Text | Google Scholar

Festinger, L. (1957). A Theory of Cognitive Dissonance. Evanston, IL: Row Peterson.

Google Scholar

Forrest, L. N., Smith, A. R., Fussner, L. M., Dodd, D. R., and Clerkin, E. M. (2016). Using implicit attitudes of exercise importance to predict explicit exercise dependence symptoms and exercise behaviors. Psychol. Sport Exerc. 22, 91–97. doi: 10.1016/j.psychsport.2015.06.006

PubMed Abstract | CrossRef Full Text | Google Scholar

Friese, M., Hofmann, W., and Wänke, M. (2008). When impulses take over: moderated predictive validity of explicit and implicit attitude measures in predicting food choice and consumption behaviour. Br. J. Soc. Psychol. 47, 397–419. doi: 10.1348/014466607X241540

PubMed Abstract | CrossRef Full Text | Google Scholar

Friese, M., Hofmann, W., and Wiers, R. W. (2011). On taming horses and strengthening riders: recent developments in research on interventions to improve self-control in health behaviors. Self Identity 10, 336–351. doi: 10.1080/15298868.2010.536417

CrossRef Full Text | Google Scholar

Gawronski, B., and Bodenhausen, G. V. (2006). Associative and propositional processes in evaluation: an integrative review of implicit and explicit attitude change. Psychol. Bull. 132, 692–731. doi: 10.1037/0033-2909.132.5.692

PubMed Abstract | CrossRef Full Text | Google Scholar

Gawronski, B., and Bodenhausen, G. V. (2011). The associative-propositional model: theory, evidence, and open questions. Adv. Exp. Soc. Psychol. 44, 59–127. doi: 10.1016/B978-0-12-385522-0.00002-0

CrossRef Full Text | Google Scholar

Gawronski, B., and De Houwer, J. (2014). “Implicit measures in social and personality psychology,” in Handbook of Research Methods in Social and Personality Psychology, 2nd Edn., eds H. T. Reis and C. M. Judd (New York, NY: Cambridge University Press), 283–310.

Google Scholar

Gawronski, B., Morrison, M., Phills, C. E., and Galdi, S. (2017). Temporal stability of implicit and explicit measures: a longitudinal analysis. Pers. Soc. Psychol. Bull. 43, 300–312. doi: 10.1177/0146167216684131

PubMed Abstract | CrossRef Full Text | Google Scholar

Greenwald, A. G., McGee, D. E., and Schwartz, J. L. (1998). Measuring individual differences in implicit cognition: the implicit association test. J. Pers. Soc. Psychol. 74, 1464–1480. doi: 10.1037/0022-3514.74.6.1464

PubMed Abstract | CrossRef Full Text | Google Scholar

Greenwald, A. G., Nosek, B. A., and Banaji, M. R. (2003). Understanding and using the implicit association test: I. An improved scoring algorithm. J. Pers. Soc. Psychol. 85, 197–216. doi: 10.1037/0022-3514.85.2.197

PubMed Abstract | CrossRef Full Text | Google Scholar

Hagger, M. S. (2016). Non-conscious processes and dual-process theories in health psychology. Health Psychol. Rev., 10, 375–380. doi: 10.1080/17437199.2016.1244647

PubMed Abstract | CrossRef Full Text | Google Scholar

Hofmann, W., De Houwer, J., Perugini, M., Baeyens, F., and Crombez, G. (2010). Evaluative conditioning in humans: a meta-analysis. Psychol. Bull. 136, 390–421. doi: 10.1037/a0018916

PubMed Abstract | CrossRef Full Text | Google Scholar

Hollands, G. J., Prestwich, A., and Marteau, T. M. (2011). Using aversive images to enhance healthy food choices and implicit attitudes: an experimental test of evaluative conditioning. Health Psychol. 30, 195–203. doi: 10.1037/a0022261

PubMed Abstract | CrossRef Full Text | Google Scholar

Houben, K., and Wiers, R. W. (2008). Implicitly positive about alcohol? Implicit positive associations predict drinking behavior. Addict. Behav. 33, 979–986. doi: 10.1016/j.addbeh.2008.03.002

PubMed Abstract | CrossRef Full Text | Google Scholar

Hyde, A. L., Elavsky, S., Doerksen, S. E., and Conroy, D. E. (2012). The stability of automatic evaluations of physical activity and their relations with physical activity. J. Sport Exerc. Psychol. 34, 715–736. doi: 10.1123/jsep.34.6.715

PubMed Abstract | CrossRef Full Text | Google Scholar

Ioannidis, J. P. A., Patsopoulos, N. A., and Rothstein, H. R. (2008). Reasons or excuses for avoiding meta-analysis in forest plots. BMJ 336, 1413–1415. doi: 10.1136/bmj.a117

PubMed Abstract | CrossRef Full Text | Google Scholar

Kahneman, D. (2003). A perspective on judgement and choice. Mapping bounded rationality. Am. Psychol. 58, 697–720. doi: 10.1037/0003-066X.58.9.697

CrossRef Full Text | Google Scholar

Karpen, S. C., Jia, L., and Rydell, R. J. (2012). Discrepancies between implicit and explicit attitude measures as an indicator of attitude strength. Eur. J. Soc. Psychol. 42, 24–29. doi: 10.1002/ejsp.849

CrossRef Full Text | Google Scholar

Karpinski, A., and Steinman, R. B. (2006). The single category implicit association test as a measure of implicit social cognition. J. Pers. Soc. Psychol. 91, 16–32. doi: 10.1037/0022-3514.91.1.16

PubMed Abstract | CrossRef Full Text | Google Scholar

Lenhard, W., and Lenhard, A. (2016). Calculation of Effect Sizes. Bibergau: Psychometrica. Available online at: https://www.psychometrica.de/effect_size.html

Liberati, A., Altman, D. G., Tetzlaff, J., Mulrow, C., Gøtzsche, P. C., and Ioannidis, J. P. A. (2009). The PRISMA statement for reporting systematic reviews and meta-analyses of studies that evaluate health care interventions: Explanation and elaboration. Ann. Int. Med. 151, W-65. doi: 10.1371/journal.pmed.1000100

PubMed Abstract | CrossRef Full Text | Google Scholar

Markland, D., Hall, C. R., Duncan, L. R., and Simatovic, J. (2015). The effects of an imagery intervention on implicit and explicit exercise attitudes. Psychol. Sport Exerc. 17, 24–31. doi: 10.1016/j.psychsport.2014.11.007

CrossRef Full Text | Google Scholar

Marteau, T. M., Hollands, G. J., and Fletcher, P. C. (2012). Changing human behavior to prevent disease: the importance of targeting automatic processes. Science 337, 1492–1495. doi: 10.1126/science.1226918

PubMed Abstract | CrossRef Full Text | Google Scholar

Moher, D., Liberati, A., Tetzlaff, J., Altman, D. G., and The, P. R. I. S. M. A., Group (2009). Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. PLoS Med. 6:e1000097. doi: 10.1371/journal.pmed1000097

PubMed Abstract | CrossRef Full Text | Google Scholar

Nederkoorn, C., Houben, K., Hofmann, W., Roefs, A., and Jansen, A. (2010). Control yourself or just eat what you like? Weight gain over a year is predicted by an interactive effect of response inhibition and implicit preference for snack foods. Health Psychol. 29, 389–393. doi: 10.1037/a0019921

PubMed Abstract | CrossRef Full Text | Google Scholar

Nosek, B. A., and Banaji, M. R. (2001). The Go/No-Go association task. Soc. Cogn. 19, 625–666. doi: 10.1521/soco.19.6.625.20886

CrossRef Full Text | Google Scholar

Nosek, B. A., Hawkins, C. B., and Frazier, R. S. (2011). Implicit social cognition: from measures to mechanisms. Trends Cogn. Sci. (Regul. Ed) 15, 152–159. doi: 10.1016/j.tics.2011.01.005

PubMed Abstract | CrossRef Full Text | Google Scholar

Payne, B. K., Cheng, C. M., Govorun, O., and Stewart, B. D. (2005). An inkblot for attitudes: affect misattribution as implicit measurement. J. Pers. Soc. Psychol. 89, 277–293. doi: 10.1037/0022-3514.89.3.277

PubMed Abstract | CrossRef Full Text | Google Scholar

Payne, B. K., McClernon, F. J., and Dobbins, I. G. (2007). Automatic affective responses to smoking cues. Exp. Clin. Psychopharmacol. 15, 400–409. doi: 10.1037/1064-1297.15.4.400

PubMed Abstract | CrossRef Full Text | Google Scholar

Prochaska, J. O., and DiClemente, C. C. (1984). The Transtheoretical Approach: Crossing Traditional Boundaries of Therapy. Homewood IL: Dow Jones-Irwin.

Rebar, A. L., Dimmock, J. A., Jackson, B., Rhodes, R. E., Kates, A., Starling, J., et al. (2016a). A systematic review of the effects of non-conscious regulatory processes in physical activity. Health Psychol. Rev. 10, 395–407. doi: 10.1080/17437199.2016.1183505

PubMed Abstract | CrossRef Full Text | Google Scholar

Rebar, A. L., Ram, N., and Conroy, D. E. (2015). Using the EZ-diffusion model to score a single-category implicit association test of physical activity. Psychol. Sport Exerc. 16, 96–105. doi: 10.1016/j.psychsport.2014.09.008

PubMed Abstract | CrossRef Full Text | Google Scholar

Rebar, A. L., Schoeppe, S., Alley, S. J., Short, C. E., Dimmock, J. A., Jackson, B., et al. (2016b). Automatic evaluation stimuli: the most frequently used words to describe physical activity and the pleasantness of physical activity. Front. Psychol. 7:1277. doi: 10.3389/fpsyg.2016.01277

PubMed Abstract | CrossRef Full Text | Google Scholar

Rhodes, R. E., and Dickau, L. (2012). Experimental evidence for the intention–behavior relationship in the physical activity domain: a meta-analysis. Health Psychol. 31, 724–727. doi: 10.1037/a0027290

PubMed Abstract | CrossRef Full Text | Google Scholar

Rothstein, H. R., Sutton, A. J., and Borenstein, M. (eds.). (2006). Publication Bias in Meta-Analysis: Prevention, Assessment and Adjustments. Chichester, UK: John Wiley & Sons.

Google Scholar

Sala, M., Baldwin, A. S., and Williams, D. M. (2016). Affective and cognitive predictors of affective response to exercise: examining unique and overlapping variance. Psychol. Sport Exerc. 27, 1–8. doi: 10.1016/j.psychsport.2016.07.005

CrossRef Full Text | Google Scholar

Sriram, N., and Greenwald, A. G. (2009). The brief implicit association test. J. Exp. Psychol. 56, 283–294. doi: 10.1027/1618-3169.56.4.283

PubMed Abstract | CrossRef Full Text | Google Scholar

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

PubMed Abstract | CrossRef Full Text | Google Scholar