ABSTRACT

The advantage of a Cox model is that the researcher can determine the influence of covariates without having to make further assumptions about the form of the hazard-rate path. A disadvantage, however, is that part of the regression model remains unknown and unspecified. This does not present a problem as long as the researcher does not have specific hypotheses regarding the hazard-rate path or when changes in event risks over the duration are unknown or occur so unsystematically that they may not be approximated by a parametric model. However, whenever fundamental hypotheses are to be tested or the researcher has clear ideas concerning the duration dependency, one should—as far as possible—rely on parametric models.