Multiple regression analyses were used to predict PYDD from drinking onset and SLE. We performed 3 sets of regression analyses separately for men and women: (i) tot-SLE (i.e., all SLE regardless of dependence ratings), (ii) iSLE, and (iii) dSLE. A baseline model estimated the effects of demographic variables (age at interview, current marital status, years of education and household income), and subsequent models retained those that were statistically significant at p ≤ 0.05 for either sex. Next, we tested the main effects of drinking onset and past-year SLE. Finally, we entered the interactions of the drinking onset variables (ONS17 and ONS14) with SLE. Generally, we use change in explained variance (R2) between nested models to index the significance of the added term(s). Given that we were testing a priori hypotheses regarding onset × SLE interaction, we also refer to p-values of the interaction terms in evaluating their significance. Generalized estimating equations adjusted for nonindependence of data from twin pairs and were used to produce correct standard errors for regression coefficients. Analyses were performed with SAS v9.2 (SAS Institute Inc., 2008).
