Esther Mertens

| 101 Classmate Influence in Intervention Likewise, the mediators deviant and prosocial reinforcement were analyzed as parallel mediators in one model per outcome measure. Baseline measures of the concerned mediators and outcome were added as covariates. Additionally, since the conditions differed significantly on ethnic background, this variable was added as a covariate. If ethnic background was a significant covariate it was retained in the model, otherwise it was dropped in favor of a more parsimonious model. Second, we tested whether the relation between changes in modeling and reinforcement and students’ perceived classroom peer context was moderated by dyadic mutuality, using multilevel analyses in M plus 8.2. At level 1, the individual level, within-dyad variation between modeling or reinforcement and the outcome was modeled. This relation was allowed to vary between individuals using a random slope. At level 2, the dyad level, dyadic mutuality was examined as predictor of the variation in the mean slope of modeling or reinforcement and the outcome. Deviant and prosocial modeling were analyzed in parallel in one model per outcome, as were deviant and prosocial reinforcement. Condition was added at level 2 as a predictor of the outcome to control for change in students’ perceived classroom peer context explained by the condition in which the students participated. Baseline measures of modeling/reinforcement and the concerned outcome were added as covariates. Due to the estimation of cross-level interactions, no standardized fit indices were available. There is significant moderation when the slope between modeling or reinforcement and the outcome is dependent on the level of dyadic mutuality. In case of a significant moderation, the differing relations for dyads with low, average, and high levels ( M ± 1 SD ) of dyadic mutuality between the concerning independent variable and outcome were graphically displayed. Results Preliminary analyses We examined group differences post intervention on the perceived classroom peer context, modeling, and reinforcement using ANCOVAs, controlling for ethnicity and the corresponding baseline measure (see Table 1). No significant differences between the conditions were found. Nevertheless, we examined mediation by modeling and reinforcement as the absence of an intervention effect (i.e., a direct effect) does not exclude the presence of an indirect effect (O’Rourke & MacKinnon, 2018). Correlations between the variables are reported in the supplementary materials. 5