“Because that’s who they are”: using language abstraction to detect attribution bias in young people’s talk about politics

Abstract

This study examines whether linguistic abstraction can serve as a methodological tool for analyzing social perception, using young people’s discourse about governmental and community actors during a climate crisis as a test case. Drawing on the framework of linguistic intergroup bias and the linguistic category model, the study investigates how implicit evaluations of institutional actors are reflected in the use of abstract versus concrete language. Eleven undergraduate participants took part in a climate hackathon, simulating decision-making and problem-solving in a fictional UK seaside city. During structured, game-based activities designed to prompt negotiation and strategic discussion, participants described the roles and responsibilities of government, community groups, and their own hackathon team. A two-way ANOVA revealed significant differences in abstraction across social groups, with governmental actors described using more abstract language than both community groups and hackathon participants. Although overall sentiment effects were limited, negative evaluations of government were associated with heightened abstraction, suggesting that institutional behavior was constructed in dispositional rather than situational terms. In contrast, community and in-group actors were described more concretely, indicating context-dependent interpretations of their actions. These findings demonstrate the value of analyzing linguistic abstraction for identifying implicit attributional biases in discourse.

Introduction

A central challenge in the study of social perception is determining how individuals evaluate, judge, and relate to other actors. Attribution theory suggests that people routinely draw inferences about the causes of others’ actions, forming judgements about responsibility, intention, competence, and moral character. These inferences shape how people explain behavior, allocate blame or praise, and evaluate social actors across contexts (Martinko and Mackey, 2019).

Beyond individual judgement, attribution processes influence behavior across interpersonal, institutional, and organizational settings, contributing to the development of social structures (Sun et al., 2019). For example, research on sports media shows that when US audiences encounter limited information, such as a crime report in which no racial details are provided, sports fans often infer that the suspect is Black and evaluate the case through stereotype-consistent attributions (Anderson and Raney, 2018). This illustrates how social actors or events are interpreted in terms of assumed motives or dispositions rather than situational constraints. Such interpretations may lead observers to view these actions as intentional or unintentional, stable or situational, appropriate or blameworthy. These attributional judgments, even when shaped by bias, guide emotional reactions, expectations about future conduct, and broader evaluations of social actors, including whether authority is viewed as sincere, criticism as justified, or inaction as neglect. In this way, attribution processes play a central role in sustaining or altering social relationships and power dynamics (Martinko et al., 2007).

Many existing approaches for determining attribution rely on questionnaires, rating scales, and other forms of self-report (Hewett et al., 2019; Campbell and Swift, 2006; Harvey and Martinko, 2009). These instruments have proven valuable, offering structured ways to capture attitudes such as trust, credibility, and perceived legitimacy. However, they also carry well-known limitations. Participants may tailor their responses to appear reasonable, tolerant, or socially conscientious, especially when dealing with politically sensitive issues. Social desirability bias may lead individuals to understate or mask their true attitudes by modifying their responses in line with perceived social norms (Bergen and Labonté, 2020). Moreover, traditional tools may struggle to reach, engage, or accurately represent certain populations (Taras et al., 2010).

To address these limitations, this paper proposes the analysis of linguistic abstraction as a methodological tool that can detect subtle attribution processes underlying people’s perceptions. Linguistic abstraction, whether someone uses generalized, stable characterizations or concrete, situational language (Semin and Fiedler, 1988; Maass et al., 1989) could provide insight into the attributions people make about the behavior or disposition of others. This relationship between language and attribution could provide a foundation for analyzing evaluative patterns embedded in spontaneous descriptions. Importantly, as these patterns emerge implicitly, the method would be less vulnerable to impression management or deliberate self-censorship.

Attribution theory suggests that people routinely draw inferences about the causes of others’ actions. This bias can be reflected in the use of language. When behavior is interpreted as reflecting fixed, dispositional traits, individuals tend to describe it abstractly. When behavior is seen as circumstantial or temporary, they describe it concretely (Martinko and Mackey, 2019). Therefore, by analyzing abstraction, researchers could gain insight into how their participants attribute blame, responsibility, competence, or moral character. For example, should a participant consistently use abstract language to describe certain actions, they may be attributing these actions to enduring characteristics, such as inefficiency, indifference, or authority. Conversely, more concrete descriptions could indicate perceptions of the actors as responding to situational pressures.

This methodological approach could become particularly valuable in domains where perceptions are contested, politically charged, or shaped by unequal power relations. Young people’s assessments of governmental and community actors during a climate crisis, for instance, would be difficult to capture accurately through surveys or current interpretive techniques used in the analysis of interviews. Many young people express frustration with institutional inaction yet may soften their criticisms when asked directly (Ekström, 2016). An analysis of linguistic abstraction could offer a means of uncovering the evaluative patterns embedded in natural descriptions, without requiring participants to explicitly articulate politically sensitive views. In doing so, this method could reveal implicit shifts in judgment.

By focusing on abstraction rather than explicit opinion, the method could capture attributional processes that are unconscious or automatic, allowing for a more nuanced examination of evaluations that might otherwise appear neutral or ambiguous using traditional methods. Considering the above discussion, the following research question emerges:

Can the analysis of abstraction levels reveal implicit perceptions of different social actors?

Theoretical framework

To understand the attributions of our participants, this study combines a psychological lens, attribution theory (Martinko and Mackey, 2019), with a linguistic one, the Linguistic Category Model (LCM) (Semin and Fiedler, 1988) via Linguistic Intergroup Bias (LIB) (Maass et al., 1989). We hypothesise that youth will use more abstract, negatively valenced language to describe governmental actors compared to their account of community groups and references to the hackathon participants, signaling deeper distrust and perceived systemic inertia. By analyzing these patterns in youth-generated discourse, the study seeks to uncover how language encodes evaluative bias and how such bias reflects broader political and social dynamics of youth engagement with climate change.

Our focus on language provides a valuable window into how the participants cognitively and emotionally position themselves in relation to decision-making processes.

Construal Level Theory (CLT) posits that psychological distance, be it social, temporal, spatial, or hypothetical, leads to more abstract mental representations. Soderberg et al.’s (2015) meta-analysis confirms a robust association between psychological distance and abstraction in language, showing that more abstract discourse reflects greater perceived distance. In our study, then, more abstract descriptions of institutional actors may indicate not only distrust but also a sense of exclusion or disconnection from political decision-making.

To contextualize LIB, it is helpful to first introduce the LCM proposed by Semin and Fiedler (1988), which distinguishes between four levels of linguistic abstraction. At the most concrete level are descriptive action verbs (e.g., punch, kiss), which refer to specific, observable behaviors. These are followed by interpretive action verbs (e.g., fight, care), which introduce inferred meaning. State verbs (e.g., love, anger) describe enduring emotional or psychological states. The most abstract form involves adjectives (e.g., aggressive, affectionate), which generalize across instances and imply dispositional traits (see Table 1).

LIB extends the LCM model to intergroup contexts (Maass et al., 1989). According to this model, people describe positive in-group behaviors in abstract terms (e.g., affectionate), suggesting enduring traits, while negative in-group behaviors are described more concretely (e.g., punch), implying situational explanations. This pattern reverses for out-group members (see Table 2).

In this study, we treat linguistic abstraction as a proxy for perceived internal versus external causality. Abstract language is interpreted as suggesting enduring dispositions, while concrete language indicates transient, context-dependent actions. This framework offers a way to detect implicit evaluative biases in descriptions of in-group and out-group behavior.

Notably, our study expands on previous LIB research (Anolli et al., 2006; Assilaméhou et al., 2013; Assilamehou-Kunz et al., 2020; Chan, 2017; Gorham, 2006; Salès-Wuillemin et al., 2014) by examining abstraction in relation to two out-groups rather than one. In doing so, we explore whether abstraction is similarly employed when describing institutional actors rather than individual group members.

This study also speaks to emerging debates about how citizens resist or reinterpret algorithmic and digital authority in policy settings. Recent scholarship has argued that algorithmic systems increasingly function as what Ling and Yan (2025) term a form of ‘algorithmic meta-capital’, reshaping the narrative policy framework by privileging technocratic framings of governance problems. Should young people’s language reflect dispositional characterizations of governmental actors, this may signal not only distrust of human institutions but also a broader resistance to the impersonal, rule-governed logic that algorithmic governance entails. Understanding such resistance matters for democratic legitimacy: if citizens perceive authority, whether human or algorithmic, as remote and essentialized rather than responsive and contextual, the narrative resources available for policy contestation may be constrained.

These considerations also connect to broader questions of symbolic and narrative power in policy processes. If an abstraction asymmetry emerges, with governmental actors characterized in dispositional, stable terms while community actors are rendered in situational, agentic language, this could be read as a form of narrative positioning. Abstract characterizations of out-group institutions would function to fix their identity within a policy narrative, limiting the symbolic space available for imagining institutional change. In contrast, a concrete framing of in-group and community actors would preserve narrative flexibility, sustaining a sense that their actions are contingent and reformable. This is consistent with narrative policy framework accounts of how villain and hero roles are allocated in policy discourse (Ling and Yan, 2025), and with broader arguments about how linguistic form shapes governance legitimacy. Attending abstraction levels may therefore offer a fine-grained indicator of how policy narratives are structured at the level of everyday discourse, with implications for civic engagement and institutional trust.

Method

Participants

Eleven undergraduate students (7 females, mean age = 20.45 years, SD = 1.21) from three universities in North-West England participated in the Climate Hackathon. Participants were recruited through online advertisements targeting climate enthusiasts and personal invitations.

Five participants (3 females, mean age = 20.20 years, SD = 0.45) were group leaders who attended a two-hour training session a week before the hackathon to familiarize themselves with the event’s structure and help develop scenarios.

Participants were divided into two groups:

• • Group 1: Five participants (3 females; mean age = 20.20 years, SD = 0.45), including three group leaders.

• • Group 2: Six participants. Two participants left after the morning session.

  • ◦ Morning session: Three females; mean age = 20.67 years, SD = 1.63, with two group leaders present.

  • ◦ Afternoon session: One female; mean age = 21.00 years, SD = 2.00, with two group leaders present.

Of the original eleven participants, two withdrew before the afternoon session commenced (Group 2), yielding nine participants who contributed data across both sessions; the two who attended only the morning session contributed data from that session only. Neither withdrawing participant produced any eligible comments, and this was also the case for one additional group member. As a result, attrition did not affect the analyzable dataset.

Ethics

Ethical approval for the study was obtained from the University of Salford’s Ethics Committee, adhering to the ethical guidelines outlined by the BPS Ethics Committee (2021). All participants provided written informed consent prior to participation and were fully informed about the study’s purpose, procedures, and their rights. Participants were explicitly informed of their right to withdraw at any time during the Climate Hackathon or afterwards, without any penalty or need to provide a reason. They were assured that their data would be anonymized and treated confidentially throughout the study.

Materials

The Climate Hackathon consisted of two sessions: a 2-h morning session and a 1.5-h afternoon session. Interactions were audio-visually recorded, and transcripts were prepared manually and organized in Microsoft Excel for analysis.

Scenario

The scenario for the Climate Hackathon was developed by the researchers, with input from the five group leaders. Both groups were presented with the following scenario:

“Your city is a seaside city in the UK, in the near future. Rising sea levels—which, scientists say, are due to climate change—are threatening the town with more and more floods each year.

“Due to the flood risk, practically all the older residents of your town have left, but an activist movement led by children and young people have convinced younger residents to stay, and actually attracted newcomers from around the country looking to build a new life in a new community.

Today is the first day of a historically new phenomenon: a community entirely made up of children and young people, aged 30 and below, in charge of the administration of their own city, and with the power to address the issues that have led to the current crisis.

Morning session

Participants were tasked to:

• • Determine the name of the city.

• • Identify what the city is famous for.

• • Outline the city’s major features.

• • Identify the main stakeholders in the city.

• • Develop policies to address at least two of the following issues:

  • ◦ Crime.

  • ◦ A mental health crisis.

  • ◦ Rising house prices.

  • ◦ Police brutality and overuse of stop and search.

  • ◦ Poor public transport.

Afternoon session

Participants were informed that a new central government had assumed political office and would no longer fund the city’s initiatives. In response, they were tasked to develop new strategies to promote climate solutions without central government support.

Procedure

The hackathon followed a structured game-based system to simulate city decision-making and collaborative problem-solving. The process included the following steps:

• 1 Scenario prompt: each team was presented with a scenario prompt that described a specific challenge or situation their city faced. For example, one prompt described dangerously high levels of air pollution in schools and its impact on children’s health.

• 2 Role playing game turns: teams played through a series of turns, with each turn involving multiple steps to develop and evaluate potential actions for addressing the scenario. The steps in each turn included:

  • ◦ Action selection: the team proposed a specific action the city should take in response to the prompt.

  • ◦ Opportunities and challenges identification: the team identified three opportunities, and three challenges associated with the proposed action. Teams could revise their action based on these considerations.

  • ◦ Outcome development: the team defined three potential outcomes of their action—success, partial success, and failure—detailing what each result would entail for their city.

  • ◦ Likelihood assessment: the team evaluated the likelihood of success on a scale from 0 to 100%, aiming for consensus on their assessment. If consensus was not reached:

    • ▪ Team members with dissenting opinions outlined specific conditions under which they would agree.

    • ▪ The team then evaluated the probability of those conditions occurring.

• 3 Probability roll: once a likelihood estimate was agreed upon, the team rolled six-sided dice on a probability chart (1–2 = failure, 3–4 = partial success, and 5–6 = success) to determine the outcome of their action. In cases of conditional agreement, the team performed a two-step roll to account for preliminary conditions (e.g., whether businesses would support the action).

• 4 Outcome application: based on the roll result, the agreed-upon success, partial success, or failure outcome was applied and recorded on a central board.

Example turn

An example turn included the following steps:

• • Prompt: “A research project finds air pollution in city schools is twenty times higher than the safety limit, endangering children’s health.”

• • Action: the team proposed banning private vehicles from the city center and offering free public transport.

• • Opportunities and challenges:

• • Opportunities:

  • ◦ Schoolchildren could organize strikes to demand action.

  • ◦ Local businesses might support the initiative, anticipating increased foot traffic.

• • Challenge:

  • ◦ Insufficient funds to implement the plan without raising taxes.

• • Outcome revision: after considering challenges, the action was revised to include a public transport subscription fee.

• • Outcomes:

  • ◦ Failure: no ban is implemented, and the situation remains unchanged.

  • ◦ Partial success: affordable public transport is introduced, but the vehicle ban is not enacted.

  • ◦ Full success: both the vehicle ban, and public transport initiative are implemented.

• • Likelihood assessment: the team initially estimated a 50% chance of success but lacked consensus. Some members suggested that business support would increase the likelihood to 80%, requiring a two-step roll.

• • Probability roll: the team rolled first for business support and then for the final action, applying the corresponding outcome.

This structured system encouraged collaborative decision-making, negotiation, and strategic thinking, while the game mechanics ensured an engaging and participatory experience for all involved.

Data analysis

Transcripts were manually analyzed for LIB. When participants described the Climate Hackathon group, the community they were serving, or the central government, verbs, adjectives, and nouns were highlighted. Verbs were categorized using Schmid et al. (2017) into descriptive action verbs (DAVs), interpretative action verbs (IAVs), and state verbs (SVs). Each verb, adjective, and noun was assigned a numerical value based on its abstraction level in the LCM (DAV = 1, IAV = 2, SV = 3, ADJ = 4, Noun = 5) (Graf et al., 2013, 563; Semin and Fiedler, 1988).

All coding was conducted by a single trained coder following a predefined coding protocol derived from the LCM. The coder was familiar with the LCM framework and its application in previous LIB research. The coding task consisted of classifying lexical items into established linguistic categories (descriptive action verbs, interpretive action verbs, state verbs, adjectives, and nouns) based on explicit linguistic and grammatical criteria rather than interpretive judgements about speaker intention. Because these categories are theoretically defined and operationalized with clear decision rules, the level of coder subjectivity is relatively limited compared with interpretive discourse coding. In cases of ambiguity, terms were coded according to the established LCM category definitions and their grammatical function within the utterance to maintain consistency. This rule-based classification procedure enables the coding process to be replicated using the same criteria. To assess the reliability of this procedure, a randomly selected 10% subset of the data was recoded by the same rater, yielding an intrarater reliability score of 98.2%, indicating a high level of agreement and supporting the consistency of the coding scheme.

Terms were distributed across eight participants (range = 1–28 per participant), indicating that abstraction scores reflect contributions from multiple individuals rather than a single source (see Table 3).

Sentiment analysis, using the VADER library, was conducted on the terms using in Jupyter Notebook (version 6.5.4). Terms with neutral sentiment were excluded, leaving only those with positive or negative sentiment.

The analysis produced the following variables:

• • IV1: social group:

  • ◦ Level I: climate hackathon group.

  • ◦ Level II: community being served by the climate hackathon group.

  • ◦ Level III: central government.

• • IV2: sentiment:

  • ◦ Level I: positive.

  • ◦ Level II: negative.

• • DV: abstraction level.

Results

A two-way ANOVA was conducted using SPSS (version 29.0.1.0) to examine the effects of Social Group (Government, Community, Hackathon Group) and Sentiment (Positive, Negative) on levels of Abstraction in language use. To enhance the robustness of the analysis given violations of normality and modest group sizes, bootstrapping (5,000 samples) was applied to all parameter estimates and pairwise comparisons.

Descriptive statistics

For Abstraction Level, the mean score was 2.72 (SD = 1.36), with scores ranging from 1.00 to 5.00.

Tests of normality

Shapiro–Wilk tests were conducted to assess the normality of the distributions. The results indicated that the Abstraction Level (W(121) = 0.86, p < 0.001) significantly deviated from normality. Despite significant deviations from normality in the Shapiro–Wilk tests, the large sample size (N = 121) suggests that the Central Limit Theorem mitigates concerns regarding the violation of normality, and the ANOVA results are considered robust to these deviations (Boos and Brownie, 1995).

Although abstraction scores are ordinal-like, parametric tests such as ANOVA are widely considered robust to violations of normality and strict interval assumptions, particularly when sample sizes are adequate and bootstrapping is applied. This position is consistent with Carifio and Perla’s (2007) argument that aggregated ordinal linguistic measures can be treated as approximately interval for the purposes of statistical modelling.

ANOVA results

There was a significant main effect of Social Group, F(2, 115) = 8.87, p < 0.001, partial η² = 0.13. Estimated marginal means (EMMs) indicated that abstraction levels were highest in the Government group (EMM = 3.83, SE = 0.31, 95% CI [3.23, 4.44]), followed by the Community group (EMM = 2.74, SE = 0.18, 95% CI [2.39, 3.10]), and lowest in the Hackathon Group (EMM = 2.33, SE = 0.18, 95% CI [1.98, 2.69]) (see Figure 1).

Figure 1

There was no significant main effect of Sentiment, F(1, 115) = 0.45, p = 0.506, partial η² = 0.004. Mean abstraction scores were similar between the Negative sentiment condition (EMM = 3.06, SE = 0.17, 95% CI [2.72, 3.40]) and the Positive sentiment condition (EMM = 2.88, SE = 0.20, 95% CI [2.48, 3.28]).

The interaction between Social Group and Sentiment was also not statistically significant, F(2, 115) = 0.73, p = 0.483, partial η² = 0.01.

The overall model accounted for approximately 16.3% of the variance in abstraction scores, R² = 0.163 (adjusted R² = 0.127).

Post-hoc analysis

Post-hoc comparisons using the LSD test revealed that the Government group (EMM = 3.83, SE = 0.31, 95% CI [3.23, 4.44]) used significantly more abstract language than both the Community group (EMM = 2.74, SE = 0.18, 95% CI [2.39, 3.10], p = 0.003) and the Hackathon group (EMM = 2.33, SE = 0.18, 95% CI [1.98, 2.69], p < 0.001). Although the LSD procedure can inflate Type I error when many comparisons are conducted, it was retained here because the comparisons were theoretically motivated and limited in number. The difference between the Community and Hackathon groups was not statistically significant (p = 0.111).¹

In the Negative Sentiment condition, Government messages (EMM = 4.09, SE = 0.38, 95% CI [3.34, 4.85]) were significantly more abstract than those from the Community group (EMM = 2.86, SE = 0.27, 95% CI [2.33, 3.40], p = 0.010) and the Hackathon group (EMM = 2.22, SE = 0.22, 95% CI [1.78, 2.66], p < 0.001). The difference between Community and Hackathon messages approached significance (p = 0.069).

In the Positive Sentiment condition, Government messages (EMM = 3.57, SE = 0.48, 95% CI [2.62, 4.52]) again showed higher abstraction than Hackathon messages (EMM = 2.45, SE = 0.28, 95% CI [1.89, 3.01], p = 0.046), but the difference between Government and Community messages (EMM = 2.62, SE = 0.24, 95% CI [2.16, 3.09]) was not statistically significant (p = 0.077).

Summary of results

The two-way ANOVA tests revealed that Social Group significantly affected abstraction levels, with the Government group using more abstract language compared to the Community and Hackathon groups. Sentiment did not have a significant impact on abstraction, and there was no interaction between Social Group and Sentiment. Post-hoc analyses confirmed that this effect was most pronounced in the Negative Sentiment condition. Overall, these findings indicate that Social Group plays a more substantial role in determining language abstraction than sentiment.

Discussion

This study aimed to evaluate whether linguistic abstraction can serve as a reliable methodological tool for studying social perception, using young people’s discussions of governmental and community actors during a climate crisis as a test case. By analyzing the level of abstraction in participants’ descriptions, the study identified attributional patterns that would otherwise remain implicit or obscured through traditional self-report methods.

Summary of key findings

The findings demonstrate that subtle shifts in linguistic abstraction provide insight into whether individuals view a group’s actions as isolated incidents or as reflections of stable, dispositional characteristics. Participants used significantly more abstract language when describing government actors than when describing community groups or their own hackathon group. In contrast, descriptions of community groups and the hackathon participants were more concrete, indicating that their behaviors were seen as context-dependent rather than reflective of inherent traits (Maass et al., 1989; Semin and Fiedler, 1988).

Although the overall effect of sentiment was not significant, post-hoc comparisons revealed that negative sentiment in particular heightened abstraction when directed at governmental actors. This suggests that negative evaluations of governmental institutions were construed as indicative of deep-seated or persistent qualities, such as incompetence or indifference, rather than temporary failings. Conversely, negative sentiment directed at the hackathon group did not produce more abstract language, implying that in-group behavior continued to be explained situationally. This reflects classic intergroup attribution biases, wherein out-groups are more likely to be essentialized (Pettigrew, 1979).

Connecting these findings to emerging debates on digital and algorithmic governance deepens their theoretical significance. The abstraction asymmetry we document, governmental actors rendered dispositionally, community actors rendered situationally, is structurally analogous to what Ling and Yan (2025) describe as the narrative policy framework’s allocation of villain and hero roles in algorithmic policy contexts. When citizens characterize institutional actors in stable, trait-like terms, they effectively remove those actors from the space of contingent political negotiation, treating governmental failure not as a correctable policy choice but as an expression of durable institutional character. This framing has implications for governance legitimacy: if governmental institutions come to be narrated primarily in abstract, dispositional terms, the political imagination of change is impaired. Conversely, the concrete framing of community actors implies that their behavior remains open to negotiation and reform, a narrative resource for civic agency. Understanding these linguistic dynamics is thus relevant not only for social psychologists but also for policy scholars interested in how narrative power shapes the possibility space of democratic governance.

The stereotype content model (Fiske et al., 2002) offers a complementary lens for interpreting the abstraction asymmetry. Government actors may be perceived as high in competence but low in warmth, a combination associated with envy and moral distancing rather than active hostility. The dispositional abstractness with which participants characterized governmental actors may thus reflect not generalized negativity but a specific evaluative structure in which competence is acknowledged while moral responsiveness is questioned. This interpretation is consistent with broader accounts of institutional distrust among young people (Bowman, 2021), and future research could profitably explore whether abstraction patterns vary as a function of perceived warmth and competence independently.

Interpreting the findings through attribution theory and LIB

These results reinforce the utility of abstraction analysis for detecting how people attribute certain traits, such as responsibility, blame, or competence. The heightened abstraction associated with governmental actors aligns with attribution theory’s prediction that abstract language signals dispositional inference. Likewise, the asymmetry between in-group (hackathon participants) and out-group (government) descriptions reflects the patterns outlined in the LIB (Maass et al., 1989). The present study, therefore, extends LIB research by demonstrating its relevance in dynamic discourse about institutions, not merely interpersonal or intergroup settings.

The findings also resonate with Construal Level Theory (Soderberg et al., 2015). The higher abstract characterizations of the governmental actors may reflect psychological distance, both social and institutional. The participants in this context appear to perceive central government as remote, unresponsive, and detached from their lived concerns, whereas community groups and peers are construed at a closer psychological distance, prompting more concrete descriptions.

Youth, institutions, and the meaning of abstraction

The abstraction patterns observed here also align with wider accounts of contemporary youth political engagement. Young people often express frustration with institutional actors who seem unresponsive to climate concerns. The maxim “Nothing about us without us,” originally developed within disability rights movements (United Nations, 2004), has been increasingly adopted by youth activists to articulate demands for agency and inclusion (Diffey et al., 2022). Their growing visibility in protests, digital activism, and policy advocacy (Müller-Bachmann et al., 2023) contrasts sharply with their under-representation in formal decision-making structures. The abstraction patterns detected in this study, particularly the attribution of abstract, dispositional qualities to governmental actors, suggested that the participants not only distrusted these institutions but also perceived them as structurally inflexible or incapable of meaningful reform.

Methodological contributions

Although the number of participants was small (N = 11), the study demonstrates that abstraction analysis offers a promising methodological approach for examining implicit attribution processes and evaluative biases across different social groups. By focusing on language rather than explicit opinion, the method can reveal underlying perceptions that may not be consciously articulated, particularly in politically sensitive contexts where impression management is common. The findings therefore support the broader argument that linguistic abstraction can serve as a valuable tool for researchers investigating trust, credibility, and social perception.

Limitations

One limitation is the study’s sample size and the limited geographical range of the participants. The sample consisted of only eleven undergraduate students, with a disproportionate number of females and a narrow age range (Mean_Age = 20.45 years). Additionally, the participants were all from three universities in North-West England. This small, homogenous sample limits the generalizability of these findings to broader populations. The lack of diversity in both the participant pool and geographical location means that the results may not accurately reflect how different groups, based on age, gender, educational status, or region, might use language or feel toward differing institutions in similar scenarios. A larger and more diverse sample would increase the external validity of the findings.

Directions for future research

Future research should address several areas to overcome the limitations of this study. Furthermore, additional factors should be considered to develop a more sophisticated methodology. First, the treatment of abstraction levels as nominal rather than ordinal data needs to be reconsidered. Future studies could improve the categorization of abstraction levels by adopting a more nuanced measurement approach, such as ordinal scales, which would better capture the varying degrees of abstraction within each category. This might involve a deeper investigation into how different levels of abstraction are perceived and operate across diverse contexts, allowing for more precise analyses of language use. Additionally, researchers could explore alternative models that do not assume equal distances between abstraction levels, providing a more accurate representation of the abstraction continuum.

An example of this approach is found in Collins and Boyd (2025) who introduced autoLIB, which combines sentiment analysis with LCM–based abstraction coding to estimate intergroup bias patterns in text. Rather than replacing theoretical constructs, autoLIB operationalizes LIB by automatically identifying desirable and undesirable behaviors and calculating their relative abstraction levels. Building on this foundation, future research could refine how abstraction is represented, for example, by adopting ordinal scales that more accurately reflect gradations between linguistic categories. Similarly, models that relax the assumption of equal distance between abstraction levels may offer more sensitive detection of subtle bias patterns across diverse communicative contexts. Together, these developments would extend the precision and applicability of automated LIB analysis beyond the current implementation.

Second, given the limited sample size and geographical range of this study, future research should aim for a larger and more diverse participant pool. Expanding the sample to include participants from a variety of demographic backgrounds, such as age, gender, educational level, and geographic location, would enhance the generalizability of the findings. Conducting studies with more diverse populations could provide insights into how different groups use language differently or interpret terms across various contexts, improving the external validity of the results. Such research could be conducted through expanded hackathon-based settings, which provide access to engaged, goal-oriented participants and naturally occurring group discourse, or through alternative methodological formats commonly used to measure social perception, such as interviews, focus groups, surveys, or experimental tasks. Employing a combination of these approaches would allow future studies to assess whether the abstraction patterns observed here are specific to participatory, activist-oriented contexts or generalize across different social and institutional settings.

To address the potential for response style bias, future studies should consider implementing indirect measures or implicit tests in the collection of linguistic data, in order to reduce social desirability bias. Techniques such as implicit association tests (IAT) or anonymous surveys could help capture more genuine responses on sensitive topics, minimizing the tendency of participants to tailor their responses to what they perceive as socially acceptable. Additionally, future research could investigate the relationship between social desirability bias and language use in controlled environments to explore its impact on language patterns more thoroughly.

Future work should also directly investigate the relationship between abstraction patterns and broader policy narrative dynamics, including how abstraction varies across different governance contexts (e.g., local versus national government), issue areas (e.g., climate versus welfare policy), and modes of political engagement (e.g., protest, deliberation, electoral participation). Connecting LIB-based methods to the narrative policy framework and to accounts of algorithmic governance (Ling and Yan, 2025, 2026) would help situate linguistic attribution research within a broader theoretical landscape, with implications for both democratic theory and policy communication.

Conclusion

This study examined whether linguistic abstraction can function as a methodological tool for analyzing social perception, using young people’s discourse about governmental and community actors during a simulated climate crisis as a test case. Rather than relying on self-report measures, the study focused on how levels of abstraction in participants’ language revealed implicit attributional patterns. The findings show that government actors were described using significantly more abstract language than community or hackathon groups, indicating that governmental behavior was more often construed as dispositional and stable. In contrast, community and hackathon actors were described more concretely, suggesting that their actions were interpreted as context-dependent rather than reflective of enduring traits. This asymmetry reflects classic intergroup attribution biases, whereby out-groups are more likely to be essentialized while in-groups are explained situationally. The findings also speak to broader debates about narrative power, governance legitimacy, and the symbolic dimensions of institutional trust, connecting fine-grained linguistic evidence to questions of political imagination and civic agency. Together, these findings demonstrate the value of analyzing linguistic abstraction for identifying implicit judgments.

References

• 1

  AndersonL. C.RaneyA. A. (2018). Exploring the relationship between sports fandom and the black criminal stereotype. Commun. Sport6, 263–282. doi: 10.1177/2167479517713152

  • CrossRef
  • Google Scholar

• 2

  AnolliL.ZurloniV.RivaG. (2006). Linguistic intergroup Bias in political communication. J. Gen. Psychol.133, 237–255. doi: 10.3200/GENP.133.3.237-255,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 3

  AssilaméhouY.LepastourelN.TestéB. (2013). How the linguistic intergroup Bias affects group perception: effects of language abstraction on generalization to the group. J. Soc. Psychol.153, 98–108. doi: 10.1080/00224545.2012.711380,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 4

  Assilamehou-KunzY.PostmesT.TesteB. (2020). A normative perspective on the linguistic intergroup bias: how intragroup approval of ingroup members who use the linguistic intergroup bias perpetuates explicit intergroup bias. Eur. J. Soc. Psychol.50, 81–96. doi: 10.1002/ejsp.2616

  • CrossRef
  • Google Scholar

• 5

  BergenN.LabontéR. (2020). “Everything is perfect, and we have no problems”: detecting and limiting social desirability Bias in qualitative research. Qual. Health Res.30, 783–792. doi: 10.1177/1049732319889354,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 6

  BoosD. D.BrownieC. (1995). ANOVA and rank tests when the number of treatments is large. Stat. Probab. Lett.23, 183–191. doi: 10.1016/0167-7152(94)00112-L

  • CrossRef
  • Google Scholar

• 7

  BowmanB. (2021). Missing an opportunity? The limited civic imagination of votes at 16. Parliament. Aff.74, 581–596. doi: 10.1093/pa/gsab016

  • CrossRef
  • Google Scholar

• 8

  BPS Ethics Committee. (2021). "BPS Code of Human Research Ethics." British Psychological Society. (Accessed April 3, 2024). Available online at: https://www.bps.org.uk/guideline/bps-code-human-research-ethics-0

  • Google Scholar

• 9

  CampbellC. R.SwiftC. O. (2006). Attributional comparisons across biases and leader-member exchange status. J. Manag. Issues18:393.

  • Google Scholar

• 10

  CarifioJ.PerlaR. J. (2007). Ten common misunderstandings, misconceptions, persistent myths and urban legends about Likert scales and Likert response formats and their antidotes. J. Soc. Sci.3, 106–116. doi: 10.3844/jssp.2007.106.116

  • CrossRef
  • Google Scholar

• 11

  ChanM. (2017). Social identity and the linguistic intergroup bias: exploring the role of ethnic identification in the context of intergroup relations between Hong Kong and mainland China. J. Lang. Soc. Psychol.36, 473–483. doi: 10.1177/0261927X17695112

  • CrossRef
  • Google Scholar

• 12

  CollinsK. A.BoydR. L. (2025). Automating the detection of linguistic intergroup Bias through computerized language analysis. J. Lang. Soc. Psychol.44, 343–366. doi: 10.1177/0261927X251318887,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 13

  DiffeyJ.WrightS.UchenduJ. O.MasithiS.OludeA.JumaD. O.et al. (2022). "not about us without us"—the feelings and hopes of climate-concerned young people around the world. Int. Rev. Psychiatry34, 499–509. doi: 10.1080/09540261.2022.2126297,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 14

  EkströmM. (2016). Young people's everyday political talk: a social achievement of democratic engagement. J. Youth Stud.19, 1–19. doi: 10.1080/13676261.2015.1048207

  • CrossRef
  • Google Scholar

• 15

  FiskeS. T.CuddyA. J. C.GlickP.XuJ. (2002). A model of (often mixed) stereotype content: competence and warmth respectively follow from perceived status and competition. J. Pers. Soc. Psychol.82, 878–902. doi: 10.1037/0022-3514.82.6.878

  • CrossRef
  • Google Scholar

• 16

  GorhamB. W. (2006). News media's relationship with stereotyping: the linguistic intergroup bias in response to crime news. J. Commun.56, 289–308. doi: 10.1111/j.1460-2466.2006.00020.x

  • CrossRef
  • Google Scholar

• 17

  GrafS.BilewiczM.FinellE.GeschkeD. (2013). Nouns cut slices: effects of linguistic forms on intergroup bias. J. Lang. Soc. Psychol.32, 62–83. doi: 10.1177/0261927X12463209

  • CrossRef
  • Google Scholar

• 18

  HarveyP.MartinkoM. J. (2009). An empirical examination of the role of attributions in psychological entitlement and its outcomes. J. Organ. Behav.30, 459–476. doi: 10.1002/job.549

  • CrossRef
  • Google Scholar

• 19

  HewettR.ShantzA.MundyJ. (2019). Information, beliefs, and motivation: the antecedents to human resource attributions. J. Organ. Behav.40, 570–586. doi: 10.1002/job.2353

  • CrossRef
  • Google Scholar

• 20

  LingX.YanS. (2025). Algorithmic meta-capital and the narrative policy framework. Policy Stud. J.53, 1108–1122. doi: 10.1111/psj.70019

  • CrossRef
  • Google Scholar

• 21

  LingX.YanS. (2026). Understanding algorithmic aversion: a bold rejection of digital dominion. Commun. ACM69, 27–30. doi: 10.1145/3758088

  • CrossRef
  • Google Scholar

• 22

  MaassA.SalviD.ArcuriL.SeminG. (1989). Language use in intergroup contexts: the linguistic intergroup Bias. J. Pers. Soc. Psychol.57, 981–993. doi: 10.1037/0022-3514.57.6.981,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 23

  MartinkoM. J.HarveyP.DouglasS. C. (2007). The role, function, and contribution of attribution theory to leadership: a review. Leadersh. Q.18, 561–585. doi: 10.1016/j.leaqua.2007.09.004

  • CrossRef
  • Google Scholar

• 24

  MartinkoM. J.MackeyJ. D. (2019). Attribution theory: an introduction to the special issue. J. Organ. Behav.40, 523–527. doi: 10.1002/job.2397

  • CrossRef
  • Google Scholar

• 25

  Müller-BachmannE.ChorvátI.MefalopulosA. (2023). Heading for a better world: micropolitical activism of young people seeking social change. J. Youth Stud.26, 387–405. doi: 10.1080/13676261.2022.2053669

  • CrossRef
  • Google Scholar

• 26

  PettigrewT. F. (1979). The ultimate attribution error: extending Allport's cognitive analysis of prejudice. Personal. Soc. Psychol. Bull.5, 461–476. doi: 10.1177/014616727900500407

  • CrossRef
  • Google Scholar

• 27

  Salès-WuilleminE.MasseL.UrdapilletaI.PullinW.KohlerC.GueraudS. (2014). Linguistic intergroup bias at school: an exploratory study of black and white children in France and their implicit attitudes toward one another. Int. J. Intercult. Relat.42, 93–103. doi: 10.1016/j.ijintrel.2014.06.002

  • CrossRef
  • Google Scholar

• 28

  SchmidJ.FiedlerK.SeminG.EnglichB. (2017). Measuring implicit causality: The linguistic category model.

  • Google Scholar

• 29

  SeminG. R.FiedlerK. (1988). The cognitive functions of linguistic categories in describing persons: social cognition and language. J. Pers. Soc. Psychol.54, 558–568. doi: 10.1037/0022-3514.54.4.558

  • CrossRef
  • Google Scholar

• 30

  SoderbergC. K.CallahanS. P.KochersbergerA. O.AmitE.LedgerwoodA. (2015). The effects of psychological distance on abstraction: two Meta-analyses. Psychol. Bull.141, 525–548. doi: 10.1037/bul0000005,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 31

  SunJ.LidenR. C.OuyangL. (2019). Are servant leaders appreciated? An investigation of how relational attributions influence employee feelings of gratitude and prosocial behaviors. J. Organ. Behav.40, 528–540. doi: 10.1002/job.2354

  • CrossRef
  • Google Scholar

• 32

  TarasV.SteelP.KirkmanB. L. (2010). Negative practice-value correlations in the GLOBE data: unexpected findings, questionnaire limitations and research directions. J. Int. Bus. Stud.41, 1330–1338. doi: 10.1057/jibs.2010.30

  • CrossRef
  • Google Scholar

• 33

  United Nations. (2004). ""Nothing about Us, Without Us" ". UN. (Accessed April 22, 2025). Available online at: https://www.un.org/esa/socdev/enable/iddp2004.htm#:~:text=The%20motto%20%E2%80%9CNothing%20About%20Us,and%20with%20persons%20with%20disabilities

  • Google Scholar