Kolmogorov-type tests of goodness-of-fit against stochastic ordering

Abstract

This article addresses the problem of testing the null hypothesis H0 that a random sample of size n is from a distribution with the completely specified continuous cumulative distribution function Fn(x). Kolmogorov-type tests for H0 are based on the statistics C+ n = Sup[Fn(x)−F0(x)] and C− n=Sup[F0(x)−Fn(x)], where Fn(x) is an empirical distribution function. Let F(x) be the true cumulative distribution function, and consider the ordered alternative H1: F(x)≥F0(x) for all x and with strict inequality for some x. Although it is natural to reject H0 and accept H1 if C + n is large, this article shows that a test that is superior in some ways rejects F0 and accepts H1 if Cmdash n is small. Properties of the two tests are compared based on theoretical results and simulated results.