Score Statistics for Current Status Data: Comparisons with Likelihood Ratio and Wald Statistics

Abstract

In this paper we introduce three natural “score statistics” for testing the hypothesis that F (t0) takes on a fixed value in the context of nonparametric inference with current status data. These three new test statistics have natural intepretations in terms of certain (weighted) L2 distances, and are also connected to natural “one-sided” scores. We compare these new test statistics with an analogue of the classical Wald statistic and the likelihood ratio statistic introduced in Banerjee and Wellner (2001) for the same testing problem. Under classical “regular” statistical problems the likelihood ratio, score, and Wald statistics all have the same chi-squared limiting distribution under the null hypothesis. In sharp contrast, in this non-regular problem all three statistics have different limiting distributions under the null hypothesis. Thus we begin by establishing the limit distribution theory of the statistics under the null hypothesis, and discuss calculation of the relevant critical points for the test statistics. Once the null distribution theory is known, the immediate question becomes that of power. We establish the limiting behavior of the three types of statistics under local alternatives. We have also compared the power of these five different statistics via a limited Monte-Carlo study. Our conclusions are: (a) the Wald statistic is less powerful than the the likelihood ratio and score statistics; and (b) one of the score statistics may have more power than the likelihood ratio statistic for some alternatives.