TESTING THE PROPORTIONAL HAZARDS ASSUMPTION IN THE PRESENCE OF UNMEASURED HETEROGENEITY

This paper develops a discrete-time hazard model which accounts for un measured hetergeneity while allowing the coefficients of the regressor s to vary over time. Sufficient conditions for nonparametric identifia bility of the unmeasured heterogeneity distribution are derived. Testi ng for time-varying coefficients in this model is, under suitable cond itions, equivalent to testing the proportionality assumption of the un derlying continuous-time hazard model. Some Monte Carlo evidence is pr esented regarding the small-sample properties of this test. As an illu stration, these tests are applied to the joblessness durations of disp laced workers.