Back to Annual Meeting

Back to Annual Meeting

APHA Scientific Session and Event Listing

Jichuan Wang, PhD¹, Robert G. Carlson, PhD¹, Russel S. Falck, MA¹, and Peichang Shi, MS². (1) Community Health, Wright State University, 3640 Colonel Glenn Hwy, Dayton, OH 45435, 937-775-2084, jichuan.wang@wright.edu, (2) Math & Statistics Dept., Wright State University, 3640 Colonel Glenn Hwy, Wright State University, Dayton, OH 45435

Count data, common measures in substance abuse study surveys, often do not follow the traditional Poisson distribution, but have extra zeros or structural zeros, leading to violation of the assumption of normally distributed standardized residuals. Very often, such count data are treated as ordinal measures or recoded as dichotomous measures, thus cumulative logit or regular logit model are applied for data analysis. For such zero- inflated count data, the aforementioned models do not work, and neither does the Poisson regression or negative binomial model. Appropriate analytical method for such data is the zero-inflated Poisson (ZIP) regression model, in which two equations (with a log link for the mean of the Poisson distribution and a logit link for the extra zeros, respectively) are simultaneously estimated. In this study, the recently developed group-based zero-inflated Poisson (ZIP) model is used to analyze longitudinal count data with extra-zeros (e.g., annually reported number of times in drug abuse treatment over an eight years period among a sample of urban crack users). Similar to the latent growth mixture model, the group-based zero-inflated ZIP model accommodates the clustering in individual outcome changes by identifying a small number of latent groups of growth trajectories; it also handles the extra-zeros in the count data. The SAS PROC TRAJ procedure (Jones et al., 2001) will be used to conduct the ZIP and group-based ZIP models.

Learning Objectives: