Prior to main analyses, univariate distributions and bivariate associations among variables were examined. The reciprocal relationships between friends' alcohol use and participants' own alcohol use and heavy drinking were evaluated with autoregressive cross-lagged models conducted in Mplus version 7. To reduce the complexity of the models and presentation of the results, alcohol use and heavy drinking were analyzed separately (the results were identical when these variables were analyzed together). The models included autoregressive paths modeling the continuity in each construct over time, covariances between friends' and own alcohol use within each time point, and cross-lagged paths linking friends' alcohol use and own alcohol use or heavy drinking across adjacent time points (see Figures 1 and 3). All paths were adjusted for participants' sex and racial/ethnic minority status, as well as age, family income and marital status at the time each predicted variable was measured. Because some variables were not normally distributed, the MLR estimator which is robust to violations of normality was used. The role of DRD4 genotype as a moderator of the relationships between friends' and own alcohol use
