GROUP DURATION ANALYSIS OF THE PROPORTIONAL HAZARD MODEL - MINIMUM CHI-SQUARED ESTIMATORS AND SPECIFICATION TESTS

This article develops a semiparametric, minimum chi-squared estimation method of the proportional hazard model for the case when durations a re grouped and covariates are categorical. The proposed estimator is e asy to compute, yet asymptotically as efficient as the maximum likelih ood estimator. This article also suggests simple specification tests f or the proportional hazard model. If proportionality holds, then two s ets of minimum chi-squared estimators, one from a further grouped data and the other from the original grouped data, will converge to the sa me quantity; otherwise, they will not. Therefore, a test of the equali ty of these two sets of estimators will offer a test for proportionali ty. Monte Carlo simulations demonstrate the performance of these estim ators and specification tests. In addition, two real data applications illustrate the implementation of the suggested methods and the contex ts in which these methods are useful.