Abstract:
|
In the class of zero-inflated models, the heterogeneity in the population is often referred to zeros generated from two different sources. As a goodness of fit, it is often of interest to evaluate whether the heterogeneity is consistent with the observed data. Typically, we use homogeneity tests to examine whether the mixture probability equals to zero under the assumption of well-specified null model. However, in practice, the mean function and the distribution of the null model may not be correctly specified in the test, which may result in biased statistical inferences. In this paper, the impact of misspecifications on the validity of these homogeneity tests is evaluated through intensive simulations. The simulation results show that the empirical type I error rates can not be maintained under misspecifications, suggesting a more robust test which can address the misspecifications is needed.
|