kidnapped, held captive, raped or molested) and nonassaultive traumas (life-threatening accident, disaster, witnessing someone seriously injured or killed, and unexpectedly finding a dead body). The earliest reported age of occurrence of an internalizing or externalizing disorder or traumatic event was selected to represent age at onset of the respective composite variable. These control variables were modeled as time-varying (i.e., coded “0” before onset and “1” during year of onset and thereafter), so that only disorders or events that preceded or occurred at the same time as the outcome contributed to risk. Demographic control variables were yearly family income (low [<$30,000], middle [referent], high [≥$75,000]), sex, ethnicity (African American and other versus European American), case family (versus comparison), and birth cohort. Age in years is implicit in the models where interactions with age were necessary to satisfy the proportional hazards assumption, described below in statistical methods.
