Data from longitudinal studies have a number of advantages for gerontological research, but the effects of attrition and the increased complexity of data structures with multiple observations may pose some problems for statistical analysis. Proportional hazards models allow for the examination of event histories using all observations of a dependent variable, and these models can incorporate time-dependent covariates to increase their explanatory power. The assumptions and applications of event history analysis using proportional hazards models are described, and the analysis of mortality data from the Ontario Longitudinal Study of Aging provides a relevant example. Extensions of proportional hazards models and commercially available software are also discussed.
