Statistical analyses were conducted using SUDAAN software (Research Triangle Institute, 2001) to adjust variance estimates for design characteristics. Multivariate logistic regression models assessed the odds of incident AUD and AUD criteria among individuals at risk, comparing AFD <15 and 15-17 with 18 or older. Covariates were entered sequentially: sociodemographic factors first, followed by adult risk factors and then by childhood risk factors. At each stage, factors not significant at the p<.05 level were removed before proceeding to the next stage. Duration of exposure was included at all stages to avoid possible bias resulting from differential exposure to risk. Tests for interactions with age and gender were added to the fully reduced models, as were consumption and its interactions with AFD. As an additional sensitivity analysis, reduced models for incidence of AUD and specific AUD criteria were estimated in a low-risk population of drinkers for whom early initiation of drinking was unlikely to have been a manifestation of pre-existing psychopathology or genetic risk. This low-risk population consisted of drinkers who did not report any family history of substance use or mental disorder and who were negative for all personality disorders and childhood risk factors (n=1,867).
