Parametric models, duration dependence, and time-varying data revisited

Box-Steffensmeier and Jones (1997) provide a good overview of the class of models known as "event history" or "duration" models, but omit some important points about the relationship between duration dependence and time-varying data in parametric models of duration. The key reason to use duration analysis methods is to allow for the possibility of duration dependence in one's data. However, while assumptions about the form of duration dependence appear to be purely methodological, duration dependence can have important substantive implications for theories of politics. There is often an interaction in estimation between duration dependence and covariates. In particular, duration dependence can be revealed or eliminated when particular time-varying variables are included in or removed from a model. This point is especially important because time-varying covariates are more central and common to important research questions than is sometimes realized. Overall, these points lead to the conclusion that findings about duration dependence go hand-in-hand with proper specification and that model specification both in terms of functional form and included variables must be carefully considered before duration methods are applied.