Cognitive processes of ingroup favoritism across 20 countries: An eye-tracking investigation of culture, behavior, and cognition

The study addresses how cultural contexts shape cognitive processes underlying in-group favoritism and discrimination in prosocial decision-making. Against a backdrop of global challenges—pandemics, climate crisis, rising xenophobia, and populism—cooperation often falters, especially when in-group bias is activated. Prior research shows robust in-group favoritism across real and minimal groups, but typically focuses on behavioral outcomes and often assumes or perturbs underlying cognitive processes. Moreover, high-resolution cognitive research has struggled with generalizability due to reliance on lab-based apparatuses and WEIRD samples. This project investigates both individual-level cognitive mechanisms and macro-level cultural factors by leveraging scalable webcam-based eye-tracking to capture information search and processing during intergroup prosocial decisions across 20 countries. The core research questions are: (1) Is behavioral in-group favoritism ubiquitous across diverse cultures in minimal group settings? (2) Which country-level factors (material security and cultural dimensions) explain cross-cultural heterogeneity in discrimination? (3) How do decision processes (effort and attention) differ when facing in-group versus out-group members, and are these conditioned by individual social preferences? (4) Who remains visually colorblind to group membership, and do individual- or country-level predictors explain this?

Extensive behavioral evidence documents in-group favoritism across societies and contexts (e.g., Balliet et al., 2014; Romano et al., 2017, 2021). Minimal group paradigms show that even arbitrary categorization (e.g., Klee vs. Kandinsky, coin flips) produces preferential treatment of in-group members (Tajfel, 1970; Billig & Tajfel, 1973). Prior work links prosocial preferences (Social Value Orientation, SVO) to parochial altruism and differential generosity (Chen & Li, 2009; De Dreu, 2010; Aaldering et al., 2013). The Material Security Hypothesis posits that uncertainty and weaker institutions foster tighter, homogenous networks and in-group favoritism (Hogg, 2000; Hruschka et al., 2014). Cultural Dimensions Theory suggests collectivism should heighten in-group bias via reputational concerns (Hofstede, 1984; Yamagishi et al., 1998; Fischer & Derham, 2016), though evidence of national parochialism is broad and heterogeneous. Cognitive and neuroscience studies show group-based differences in attention and neural processing (Van Bavel et al., 2008; Kawakami et al., 2014; Cikara et al., 2017), but process-level generalizability across cultures remains underexplored due to methodological constraints. Eye-tracking provides fine-grained process data (Krajbich et al., 2010, 2011; Schulte-Mecklenbeck et al., 2017), and webcam-based systems enable diverse, global sampling (Papoutsaki et al., 2016; Yang & Krajbich, 2021). Prior work indicates prosocial preferences increase decision effort and can condition in-group biased generosity (Rahal et al., 2020).

Design: Mixed design with within-subject manipulation of group setting (in-group vs. out-group), between-subject country factor (20 countries), and continuous individual SVO. Minimal groups were formed via a color identification task, followed by a reinforcement stage (incentivized reaction-time competition against an out-group member for a monetary bonus) to increase group salience. Participants: N_collected = 1850; N_after exclusions = 1792; mean age = 26.22 (SD = 4.78); 777 female, 30 diverse. Recruitment was online in spring 2023. Ethics approval: Max Planck Ethics Committee (Bonn DecisionLab). Compensation: fixed fee equivalent to £3.5 plus £1.45 variable payoff based on decisions and performance. Countries: 20 regions including Australia, Chile, Germany, Greece, Hungary, India, Indonesia, Italy, Kenya, Mexico, Pakistan, Philippines, Poland, Portugal, Russia, South Africa, Spain, UK, USA, Vietnam; some pooled to regional groups when sample sizes were small (e.g., Southeast Asia; Indian Subcontinent). Tasks and Measures: - SVO Slider Task (Murphy et al., 2011) assessed prosocial orientation. - Main task: 80 trials (58 target items, 2 fillers) of decomposed dictator games. In each trial, participants chose between a selfish option (maximizing own payoff, minimal other’s payoff) and a prosocial option (lower own payoff than selfish choice but higher other’s payoff and higher joint sum). Information displayed in a table: own points, other’s points, and other player’s group membership (diagnostic, in-/out-group coded as 1/0) plus a random number (nondiagnostic). Presentation was counterbalanced; text in small font and spatially spread to reduce peripheral legibility. - Eye-tracking: Webcam-based gaze estimation recorded fixation-based metrics during decision screens. Decision process variables included decision time (log-transformed), fixation counts, number of inspected information pieces, and attention distribution (proportion of attention to own vs. other outcomes). Analyses restricted to trials where group-identifying information was fixated at least once; complementary analyses considered all trials. - Colorblindness assessment: Look-up rates of group membership info vs. random number; predictors included individual SVO, differences in team identification and liking (in-group vs. out-group), and expectations that others look up group info. - Explicit colorblindness preference: At session end, participants indicated whether they wanted to learn the matched player’s group before allocating 100 points in a final incentivized dictator game; regardless of preference, they were randomly matched to in-/out-group or unknown-group counterparts. Country-level predictors: Material Security indicators (government effectiveness from World Bank Governance Indicators; religiosity from World Values Survey; historic disease burden via DALYs; COVID-19 burden via confirmed deaths) and Cultural Dimensions (individualism). Covariates included platform (Prolific vs. other), English as a second language, and trial index. Analysis: Pre-registered hypotheses and meta-analytic approach. Mixed-effects logistic models for prosocial choice; linear mixed-effects for process measures; random-effects meta-analyses across countries for in-group favoritism in behavior and process variables; exploratory multivariate meta-regressions testing moderators. Data and code were preregistered and made available on OSF and GitHub.

Behavioral in-group favoritism: - Facing an in-group vs. out-group member increased odds of prosocial decisions (OR ≈ 4.57–4.58; 95% CI around 3.44–6.11; P < 0.001). Meta-analysis showed in-group favoritism present in each country studied, with substantial heterogeneity. - Prosocial individuals discriminated more between in- and out-group than selfish individuals (interaction via SVO angle, OR = 1.36; 95% CI [1.32, 1.41]; P < 0.001), though this relationship reversed in some countries. Country-level predictors of discrimination: - Greater societal uncertainty was associated with increased in-group favoritism: lower government effectiveness (World Bank) predicted more discrimination; higher importance of religion predicted more discrimination (opposite to preregistered expectation); lower historic DALYs and higher COVID-19 deaths were related to increased favoritism. - Individualism (Hofstede) positively related to in-group favoritism (opposite to collectivism-based prediction), suggesting individualistic cultures may foster stronger in-group identification needs. Decision processes (effort and attention): - Pooled analyses showed higher decision effort when facing in-group members: increases in log(decision time) (β = 0.04, P = 0.026), fixation count (β = 1.14, P = 0.038), and inspected information pieces (β = 0.08, P = 0.002). Meta-analyses across countries confirmed significant average in-group effects, with large between-country heterogeneity (varying sizes and directions). - No pooled evidence for in-group favoritism in attention distribution to others’ vs. own outcomes; meta-analysis yielded net-zero overall effect despite country-specific effects differing from zero. - More prosocial SVO was consistently associated with increased decision effort: longer decision times (β = 0.10, P < 0.001), more fixations (β = 2.10, P < 0.001), more inspected information (β = 0.16, P < 0.001). The predicted interaction (SVO × group setting) was net-zero in pooled data but significant and directionally varied across countries. Colorblindness (visual attention to group membership): - In most trials (53.49%), participants did not fixate group membership information even once, but they fixated group information more often than the random number (33.28% of trials; Wilcoxon signed-rank V = 1,026,481; P < 0.001; r = 0.39). - Individual predictors of looking up group info: higher SVO (OR = 1.26; P < 0.001), greater in-group vs. out-group identification (OR = 1.08; P = 0.025), greater in-group vs. out-group liking (OR = 1.09; P = 0.017), and stronger expectations that others look up group info (OR = 1.01; P = 0.001). Country-level predictors did not significantly explain colorblindness (Table 4). Explicit preference effects: - 54.79% preferred to learn the matched player’s group. Preference to know predicted higher odds of visual look-up (OR = 1.91; z = 5.45; P < 0.001). - In the final dictator game, those preferring to know significantly favored their in-group, giving ~10% less to out-group than in-group (β = 10.54; t = 2.00; P < 0.001). Those preferring to remain colorblind did not significantly discriminate (β = 3.576; t = 2.08; P = 0.09).

The study demonstrates robust in-group favoritism in prosocial decision-making across 20 countries, even under minimal group conditions. Behavioral discrimination was systematically tied to societal uncertainty (government inefficacy, disease burdens) and cultural individualism, though some effects contradicted preregistered expectations (e.g., religiosity and individualism). Importantly, the research goes beyond behavior to show that cognitive processes preparing decisions—decision effort and information inspection—are modulated by group context and individual preferences, with prosociality increasing effort and the in-group context generally eliciting more processing. However, the direction and magnitude of these effects varied substantially by country, indicating culture-specific cognitive strategies. Colorblindness to group membership was common; nevertheless, individuals with stronger prosocial preferences and in-group identification were more likely to seek diagnostic group information and to discriminate. These findings underscore that cross-cultural differences in both behavior and cognition can shape the construction of intergroup decisions, cautioning against one-size-fits-all theoretical or policy approaches.

This work contributes a large, cross-cultural, webcam eye-tracking dataset linking societal context, individual preferences, and high-resolution decision processes to in-group favoritism. It confirms widespread in-group favoritism and identifies societal uncertainty and individualism as correlates of discrimination, while revealing substantial cross-cultural heterogeneity in cognitive processing. The study highlights the need for theory that models not only choice-level but also process-level predictions and for methods that broaden sample diversity. Policy implications include designing culturally informed interventions to reduce discrimination and tailoring institutional designs to account for cross-cultural variability in how decisions are constructed. Future research should develop process-based models of social cognition, expand multilingual and culturally adapted protocols, and experimentally probe mechanisms via stakes manipulation and contextual framing to disentangle the roles of language, wealth, and norms.

- Causality constraints: Cultural and sociodemographic variables cannot be experimentally manipulated; unobserved heterogeneity may influence results. - Language and instruction limitations: English-only instructions may not trigger native-language norms; speaking English as a second language and migratory history may affect cognition and behavior. - Stakes and wealth: Payoff structures were not tailored to personal wealth or purchasing power, potentially affecting processing in high- vs. low-stakes contexts. - Eye-tracking constraints: Webcam-based eye-tracking, while scalable, may have lower precision than lab systems; analyses often restricted to trials where group information was fixated at least once. - Cross-country pooling and regional grouping: Some countries were pooled due to small samples, potentially obscuring country-specific effects. - Heterogeneity and model generalizability: Large between-country heterogeneity in both behavioral and process effects; preregistered macro-level predictors did not systematically explain process-level variability.