3 added perceived substance use of friends. Examination of the tetrachoric correlations between covariates indicated no evidence of collinearity in the regression models. Dummy variables representing missing maternal data (for mothers who did not complete an interview) and missing sibling data (for offspring with no siblings) were included in all regressions with maternal or sibling data to adjust for potential non-response bias; in all cases these were non-significant. The Huber White robust variance estimator was used in all regressions to correct for the non-independence of family data. The STATA statistical program was used for all tests (StataCorp, 2009).
