On the mean residual life regression model

Abstract

In this paper, goodness-of-fit testing for the mean residual life regression model recently proposed by Maguluri and Zhang (1994) is studied. The test statistic is derived from a model-based process which is asymptotically Gaussian. Since the asymptotic covariance structure appears to be very complicated and depends on the underlying distribution of the data, the limiting null distribution of the test statistic is analytically intractable. Two resampling approaches, namely the so-called random symmetrization and the well-known bootstrap method, are used to approximate critical values of the test. Given any set of data coming from the null hypothesis, the conditional distributions generated from both methods are asymptotically equal to the limiting distribution of the test statistic. In addition, the proposed random symmetrization method can detect local alternatives which approach the null model at the rate n-1/2. A simulation study indicates that the distributional approximation through random symmetrization outperforms the bootstrap for small sample sizes. Moreover the former is computationally more efficient. © 2002 Elsevier Science B.V. All rights reserved.