Testing the proportional hazards assumption in the presence of unmeasured heterogeneity

This paper develops a discrete‐time hazard model which accounts for unmeasured hetergeneity while allowing the coefficients of the regressors to vary over time. Sufficient conditions for nonparametric identifiability of the unmeasured heterogeneity distribution are derived. Testing for time‐varying coefficients in this model is, under suitable conditions, equivalent to testing the proportionality assumption of the underlying continuous‐time hazard model. Some Monte Carlo evidence is presented regarding the small‐sample properties of this test. As an illustration, these tests are applied to the joblessness durations of displaced workers.