Therefore, an intercept only model was estimated (random intercept with no predictors) using the mixed procedure in SAS version 9.3 (SAS Institute Inc., 2011). Interclass correlations (ICCs) (Raudenbush and Bryk 2001) were calculated from clustering effects of family for all study outcomes (teacher-reported externalizing behaviors and the subscales rule breaking and aggression). Results suggested that the amount of variance accounted for by the clustering effect of families on study outcomes was significant for rule breaking (ICC = 0.24, P < 0.001) but not significant for externalizing (ICC = 0.11, P = 0.08) or aggression (ICC = 0.11, P = 0.07). Accordingly, a random coefficients (random intercept with adolescent-level predictors) hierarchical linear model accounting for family clustering was estimated across models to have consistency. Adolescent-level or level-1 predictors included the following: sex, race, GABRA2, peer delinquency, and the interaction term. Moreover, correlated residuals were accounted for using the REPEATED statement.
