After assessing univariate and bivariate statistics for the full sample, we estimated logistic regression models for the association between PTEs and lifetime prevalence of drug use among adolescents. First, we estimated the association between any PTE and the five drugs of interest, controlling for baseline demographic variables. We then controlled for parent substance misuse in these models. We repeated these steps using a categorical variable for number of PTEs reported (0, 1, 2, 3 or more) and for specific categories of PTEs. Finally, we assessed statistical interaction between PTEs and parental substance misuse, adolescent distress disorder, and adolescent behavior disorder. For any significant interaction, regression models were stratified by the moderator of interest. Sensitivity analyses assessed whether effect estimates changed when controlling for all PTE types in the same model simultaneously.
