AD is available from parent self-report only, parental alcoholism was modeled as a time-invariant predictor. Parental separation was modeled as time-varying to ensure onset before or at the same time as initiation of substance involvement. This was achieved using person-year data, with each line of data representing a single year of life for each twin. Intact families were right-censored at twins’ age at last interview if younger than age 18, and in the case of parental death during childhood, right-censored at twins’ age when their parent(s) died. Dummy codes were computed from person-year data to distinguish alcoholic separated, alcoholic intact and nonalcoholic separated families, with nonalcoholic intact families comprising the reference group. Control variables with available ages of onset (CD, SAD, MDD, suicidality, and physical and sexual abuse) were examined as time-varying predictors in adjusted models. To examine potential violation of the proportional hazards assumption, such as might be the case if the effects of parental alcoholism and/or parental separation on substance involvement differ across age periods, the Grambsch and Therneau test of Schoenfeld residuals (Grambsch and Therneau 1994) was employed, with age-interactions modeled to correct observed violations (Cleves et al., 2004).
