Our primary analyses consisted of comparing alcohol use across time between the level of response and prevention program groups. Because participants were allowed to miss individual modules/questionnaires without being dropped from the study, many individuals had missing data at one or more time points; thus we employed a linear mixed model design rather than the repeated-measures ANOVA used by Schuckit et al. (2012). Analyses were conducted in SPSS version 21 using the GENLINMIXED command, with an unstructured covariance matrix to account for the unequal spacing between the repeated measurement occasions, and a robust sandwich estimator to improve estimation of the standard errors. These models estimated the effects of LR (Low; High), Program (LRB; SOTA), and their interaction across time to predict change in alcohol use across the prevention program and follow-up assessments (Weeks 4, 8, and 22). Sex, ethnicity, and baseline alcohol use (from the Baseline/Fall survey) were included as covariates to account for residual confounding after participant matching. To follow up these analyses, we also included other factors in the models, including percentage of post-module quiz questions answered correctly,
