WHY NO ADDITIVE HAZARDS MODELS

The intuitively attractive additive hazards model is compared with pro portional hazards and accelerated failure time models. The lack of ide ntifiability limits the use of the model and prevents the application of regression versions using covariates. Fortunately, data analysis ba sed on nonhomogeneous Poisson processes or on proportional hazards is likely to yield most of the information available in the data, even th ough they: 1) do not necessarily represent the underlying process, and 2) even seem unlikely in certain situations. In particular, proportio nal hazards modeling appears very robust and requires few assumptions.