Testing for a monotone trend in a modulated renewal process

Abstract

In examining point processes which are overdispersed with respect to a Poisson process, there is a problem of discriminating between trends and the appearance in data of sequences of very long intervals. In this case the standard "robust" methods for trend analysis based on log transforms and regression techniques perform very poorly, and the standard exact test for a monotone trend derived for modulated Poisson process is not robust with respect to its distribution theory when the underlying process is non-Poisson. However, experience with data and an examination of the departures from the Poisson distribution theory suggest a modification to the standard test for trend, both for modulated renewal and general point processes. The utility of the modified test statistic is verified by examining several sets of data, and simulation results are given for the distribution of the test statistic for several renewal processes