We included several sociodemographic covariates in the model, including child’s ethnicity, educational level for mother and partner, and occupational status for mother and partner (as described in Melotti et al., 2011, 2013). In addition, we included indices of parental alcohol consumption and problems, as these are known to relate to both early temperament and adolescent substance use (Chassin et al., 1996; Leonard et al., 2000), and this allowed us greater confidence that any detected associations would not be due to correlated parental alcohol use. Information on parental alcohol use was collected at multiple time points in ALSPAC. A 5-factor solution was derived based on all available information (as detailed in Kendler et al., manuscript submitted for publication), with the factors corresponding to alcohol consumption by mother, alcohol consumption by mother during pregnancy, problem drinking in mother, alcohol consumption by partner, and problem drinking in partner. The resultant factor-derived scores were included as covariates in all structural equation models.
