Test Statistics for Dispersion Parameter in Poisson Regression and Generalized Poisson Regression Models

Abstract

Two symmetrical distributed test statistics, called Zm and Z0_New are proposed and their goodness-of-fit tests are compared with other available five test statistics: Wald-t, Score test, Z μ, ZY, and Z0, for overdispersion in Poisson regression model versus generalized Poisson model. Five thousand data sets in each condition of overdispersion parameters and sample sizes are simulated to perform the assessment of the models’ fits using those statistics, concerning the coverage probability and power of tests. Results show that the Zm test performs closely as good a Zμ and ZYtests but it tend to be better than the others when the sample size is large. Even if the Z0_New test has the largest power; however, in consideration for coverage probability and power of tests, the Zm test probably be more reliable. The Zm test statistic is interesting not only in its simplest form, with the reasonable coverage probability and power but also in its robust property of using median that needs fewer assumptions for its parent distribution.