On statistical testing and mean parameter estimation for zero-modification in count data regression.

Abstract

For the problem of testing for zero-modification in Poisson regression, a simple and intuitive test can be constructed by computing directly confidence intervals for the number of 0's under the Poisson assumption. This requires the ability of estimating the mean function accurately even if the data are in fact zero--inflated or deflated. A novel hybrid estimator is introduced for this purpose, which is of interest beyond the scope of the motivating test problem.