Application of Random Effect Zero-Inflated Negative Binomial (RE-ZINB) Model to Substance Abuse Studies

This study demonstrates application of the recently developed Random Effect Zero-Inflated Negative Binomial (RE-ZINB) Model to substance abuse studies. Under many circumstances count data contain more zeros than would be expected in the Poisson distribution. Count data with extra zeros in hierarchically structured data can be modeled by Random Effect Zero-Inflated Poisson (RE-ZIP) Model. However, when count data have over-dispersion and are hierarchically structured, random effect ZINB (RE-ZINB) models should be applied to accommodate for extra zeros, over-dispersion, as well as ICC, simultaneously. An example of application of the RE-ZINB is demonstrated using a sample of 430 crack cocaine users recruited from the Dayton, Ohio, area between July 1996 and August 1997. The outcome measure under study is number of days of crack-cocaine use in the past 30 days over an 8-year observation period. SAS Proc NLMIXED procedure is applied for model estimation.