Main effects of the ADH1B variant and peer drinking were examined in univariate and multivariate models of age of first intoxication and first DSM-5 symptom in the sample of adolescent ever-drinkers (N=1,550, called univariate model set and multivariate model set, Table 2). All models presented in the tables employed STRATA statements for gender and ethnicity to adjust for differences in baseline hazards in these groups. The interplay between the ADH1B variant and peer drinking was assessed by adding product interaction terms to models of drinking milestones (called interaction model set, Table 2). This final proportional hazards model was λ(t)=λ(t)exp(β1*(rs1229984) + β2*(peer_drinking) + β3*(rs1229984*peer_drinking)). The possibility of a gene-environment correlation between ADH1B rs1229984 and peer drinking was also assessed because genetic factors influence selection of peers who drink (Fowler et al., 2007) and inadequate control of this correlation could produce false interactions. Using logistic regression, the outcome peer drinking was modeled with the variables of the ADH1B variant, gender, and ethnicity.
