Univariate and multivariable discrete-time survival analyses (with person-year as the unit of analysis) (Jenkins, 1995) were implemented using SUDAAN version 9.1, (Research Triangle Institute, Research Triangle Park). The models aimed at assessing the association between socio-demographic, psychiatric comorbidity and substance use-related covariates and the hazards of transition to substance dependence. The person-year variable was defined as the number of years from substance use onset to age of dependence onset or age at Wave 2 interview (for censored cases). Education, marital status, and presence of DSM-IV mood, anxiety, and SUDs other than the one under examination were included as time-dependent covariates. Stepwise model selection procedures were used to identify independent correlates based on likelihood-ratio test. Taylor series linearization methods implemented in SUDAAN were used to estimate standard errors and significance, and to accommodate for the complex survey design. All estimates were obtained using Wave 2 weights.
