Three sets of Cox proportional hazards (PH) regression analyses were conducted to predict onset of alcohol, cigarette, and cannabis use (i.e., first use). Cox PH regression analyses are commonly used with time-to-event data to account for the possibility that some participants who have not yet experienced the event of interest (here, use of alcohol, cigarettes, or cannabis) may do so in the future. The advantage of this approach is that data up until the time of censorship (most recent interview) is used in the calculation of hazard ratios (HRs). In addition, tests of the PH assumption that risk remains constant over time can reveal the extent to which risk associated with CSA varies across the period of risk, i.e., whether CSA is a predictor specifically of early initiation of substance use. CSA was modeled as a time-varying covariate by creating a ‘person year’ data set using SAS (37). Data were constructed such that each line of data represented a single year of life for each case. CSA was coded as absent in each year prior to age of onset and
