Fig. 3: Defense participation was also impacted by social factors.

a Defense participation was driven by concerns for others. Defenders (n = 40 groups) were less likely to leave once one (38.75%) or two (39.00%) defender(s) lacked the ability to leave, compared to when everyone could leave (71.58%). Assumption-based 95% confidence intervals for the proportion of defenders leaving were calculated using the standard error of the mean. Dots show averages across participants. b Raincloud plot illustrating the contributions to conflict for defenders who stayed exclusively for blocks in which leaving abilities were asymmetric (i.e., when one or two defenders could not leave). Defenders, who could leave, contributed more than the cost of leaving (34.69% on average). However, 35.74% of defenders who stayed voluntarily contributed less to conflict than the leaving cost. Defenders, who could not leave, contributed more than defenders who could leave (42.80% on average). The dotted line shows the cost of leaving (25%). Dots show individual contributions across participants. c Social preferences (measured by the social value orientation angle) predicted defenders’ conflict contributions, leaving propensity, and individual earnings (mediation model based on regressions with bootstrapped confidence intervals, coefficients show standardized path coefficients). Social preferences directly predicted earnings (p = 0.029). Social preferences also predicted conflict leaving (p = 0.034), and conflict leaving predicted earnings (p < 0.001). Furthermore, social preferences predicted conflict contributions (p = 0.001), and conflict contributions predicted earnings (p < 0.001).