Using Panel Data to Estimate the Effects of Events

Although event history analysis provides a highly developed body of methods for studying the causes of events, there is little consensus on the best ways for studying the consequences of events. This article develops some methods for using multiwave panel data to estimate the effects of either naturally occurring events or planned interventions. It does this by synthesizing the literature on interrupted time series with econometric treatments of pooled time-series, cross-section data. The emphasis is on fixed-effects models and estimators because of their capacity to control for all stable differences across individuals, whether or not those differences are correlated with measured variables. In contrast to earlier treatments of the problem, the models allow for timevarying covariates and for events that can occur at different time periods for different individuals. For continuous dependent variables, the basic estimators are easily obtained with standard OLS regression programs. For dichotomous outcomes, logit models can be estimated by the method of conditional likelihood.