The R lme4 package (59) was used to fit a longitudinal logistic multilevel model, which tested whether baseline brain volume in nondrinking adolescents predicted future initiation of alcohol use. The model included both random intercept and random slope components, and time was coded as both the linear and quadratic age at the date of response. The model tested the interaction between GMV and age (i.e., does baseline ROI volume predict a different likelihood of initiation as participant’s age?). Covariates were z-scored, and they included demographic variables (age, sex, ethnicity, and SES), stress (Childhood Trauma Questionnaire and Stressful Life Events Schedule), Tanner stage, Mood and Feelings Questionnaire scores, family history of depression, age at magnetic resonance imaging scan, and intracranial volume. Second-order interactions between covariates and primary variables (e.g., middle frontal volume × sex, middle superior volume × SES, age × sex, age × SES) were also included (58). Each of 2 ROIs, right superior frontal cortex and right middle frontal cortex, was tested in a separate model. Insula volume was excluded, as it was not significant in DNS longitudinal analyses. The p values were subsequently FDR corrected (4 tests).
