Discrete-time event history analysis using segmented hazards

Event history analysis is a means of explaining variation in the timing of events in individual life histories. This article describes methods for overcoming two difficult problems likely to be encountered in applications of event history analysis to studies of aging and human development. First, in many studies the ages of occurrence of critical life events are recorded in discrete units such as years, but the probability distributions of life events are usually specified in continuous-time form. We show how to estimate models for discrete-time data based on an underlying continuous-time specification. Second, the standard distributions for life events often fail to capture the complex age-dependence seen in actual data. We show how to construct a model using segmented hazards, that is, a composite of different functions for different segments of time. To illustrate these points, we study the age of first intercourse of 11,883 subjects from the National Longitudinal Study of Youth.