The second and third series of multinomial logistic regressions substitute disaster and unexpected death of someone close, in this order, for accidents, as the comparison event. N’s in tables and in the text are unweighted. Weighted percentages and RRRs from multinomial logistic regressions were estimated using Stata 11.0 (StataCorp LP, USA). The ‘mlogit’ command was used for the multinomial logistic regressions. Taylor series linearization was used to take into account the complex survey design. Models are adjusted for sex, given the consistent finding of a sex difference in the conditional risk of PTSD. Evidence of the association of PTSD with other sociodemographic variables, such as indicators of SES and race/ethnicity, is inconsistent (and when found, the associations are far weaker). Estimating these models with education and race as additional covariates did not alter the results depicted below.
