The sampling restrictions are of special importance. The vast majority of disasters occur in low- and middle-income countries (Roy et al. 2011), but our analyses were carried out exclusively in high-income countries. In addition, the samples were restricted to household residents. This means that we excluded people living in displacement camps and other group quarters, which is an especially serious limitation given that displacement was a significant predictor of disaster-related PTSD. While the 378 respondents assessed for randomly selected disasters is sufficient to estimate prevalence of post-disaster PTSD with good precision, the fact that PTSD was an uncommon outcome (n = 16) meant that we lacked statistical power to estimate multivariate predictor coefficients with precision. Indeed, with 13 model coefficients, the 1.2 events-per-variable (EPV) ratio was well below the value recommended to avoid biased OR estimates (Peduzzi et al. 1996). Caution is consequently needed in interpreting our results because of low EPV and the clear evidence of model instability noted in Table 3.
