28 – Analysis of Count Data
Modeling Correlated Counts with Excess Zeros and Time-Dependent Covariates: A Comparison of ZIP and Hurdle Mixed Models
Trent L. Lalonde
University of Northern Colorado
Count responses often show an excess of zeros under the assumption of a Poisson distribution. Common modeling solutions include the zero-inflated Poisson model and the hurdle mixture model (Hu et el (2011); Mullahy (1986); Lambert (1992)). Recently researchers have begun to consider the modeling options for clustered or correlated count responses with excess zeros (Kassahun et al, preprint). However, it has yet to be considered whether certain models are preferred for correlated count responses with excess zeros in the presence of time-dependent covariates. Time-dependent covariates have been shown to affect parameter estimate bias and efficiency in longitudinal analyses (Pepe and Anderson (1994); Fitzmaurice (1995); Lai and Small (2007)). In this paper a comparison is made between the zero-inflated Poisson and the hurdle model for correlated count data with time-dependent covariates. Consideration is given to parameter estimate bias and hypothesis test results. An example data set is analyzed, using a longitudinal measure of the number of times of drug use as response.