Abstract:
|
Excessive zeros may cause over-dispersion and invalidate inferences when fitting Poisson regression models. If a zero-modified Poisson regression model is assumed, then the commonly used tests: Score, Wald and likelihood ratio tests, can be applied. Without assuming such a zero-modified model, He et al. (a test of inflated zeros for Poisson regression models, Statistical Methods in Medical Research, in press) recently developed a test simply based on the Poisson model. In this paper, we further study these tests and compare their theoretically asymptotic behavior and empirical performance for small to moderate sample sizes by simulation.
|