Disaster research methods: past progress and future directions.

Published results for 225 disaster studies were coded on methodological variables, severity of effects, and event year. Methods varied greatly, but cross-sectional, after-only designs, convenience sampling, and small samples were modal. Samples that were assessed before the disaster, selected for reasons of convenience, or were large tended to show less severe effects than other samples. Developing countries were underrepresented overall, but not in recent years. Certain desirable study characteristics (longitudinal designs, representative samples) have been decreasing in prevalence over time, whereas others (early first assessment) have been increasing. Innovations such as latent trajectory modeling or hierarchical linear modeling might advance the field's ability to capture the complexity of disasters, but the field still needs to attend to the fundamentals of sound epidemiologic research.