To test our hypotheses, we fitted four stepped models. Model 1 (unconditional growth model) included a random intercept for individuals plus a random term for age. This permitted an assessment of variation between individuals in the level and rate of change in alcohol use. In Model 2, family environment variables were added (mother/father emotional closeness, family conflict, and parent disapproval of alcohol use). This enabled an evaluation of how much of the growth in alcohol use over adolescence was attributable to changes in family environment. In Model 3 peer alcohol use was added. An evaluation of changes in family environment variables with peer alcohol use in the model was used to test for mediation effects (Hypothesis 2). In Model 4, controls for sensation seeking, family structure and family SES were added.
