Peer Experiences in Short-Term Residential Treatment: Individual and Group-Moderated Prediction of Behavioral Responses to Peers and Adults

Because our focus is primarily on fixed effects, we present only a brief summary of random effects. There was significant intercept variation over groups for each dependent variable (general response ICCs were 0.09–0.16 with tests of intercept variability χ ²(1) ≥ 4.21, p’s ≤ 0.04; interactant-specific response ICCs were 0.12–0.29 with χ ²(1) ≥ 6.24, p’s ≤ 0.012), supporting the decision to use multilevel modeling with random intercepts. We used significant cross-level interactions as evidence of random slopes (Snijders and Bosker 1999) and found them for aggressive, prosocial and withdrawn responses, in general and/or interactant-specific analyses (see Tables 3 and 4). To facilitate comparisons of effects across behaviors, random slopes were fit for all responses and all models included a cross-level interaction term. For all general responses, this model resulted in significant improvement over the unconditional means model using deviance tests, χ ²(8) ranging from 22.34 to 59.88 with p’s ≤ 0.004. For aggressive and prosocial responses to adults and withdrawn responses to peers this model also resulted in significant improvement over the null model, χ ²(8) ranging from 23.22 to 43.99 with p’s ≤ 0.003, suggesting that we had more success explaining the variability over groups for these interactant-specific responses than we did for aggressive and prosocial responses to peers, and withdrawn responses to adults.