Survival analyses were performed in STATA version 11.1 (StataCorp, 2009), with the Huber-White robust variance estimator used to compute standard errors and confidence intervals adjusted for non-independence of twin-family data. Cox proportional hazards (PH) regression was conducted to predict timing of alcohol involvement (separately for first alcohol use, first AD symptom, and AD diagnosis) from father and co-twin substance dependence and parental separation. We also conducted tests of interactions between parental separation and father substance dependence. Consistent with earlier work using an Australian sample (Waldron et al., 2014a), father and co-twin AD and DD risk groups were initially modeled separately, with post-hoc tests for equality. Parental separation was modeled as a time-varying covariate to ensure that only its occurrence prior to or during the same year as the transition of interest (initiation, first symptom, or AD diagnosis) was counted toward risk for transitioning. Offspring from intact families were right-censored on this variable at age at interview if younger than 18 years: these individuals were not assessed throughout childhood (defined as birth through age 18), and thus contribute to prediction through
