Modeling HIV Risk Reduction Interventions: The Zero-Inflated Negative Binomial Regression Model

• Activity Number: 492
• Type: Topic Contributed
• Date/Time: Wednesday, August 1, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Yinglin Xia*+
• Companies: University of Rochester
• Keywords: HIV risk reduction intervention ; structural zero, overdispersion ; Poisson distribution ; negative binomial (NB) distribution ; zero-inflated Poisson (ZIP) distribution ; zero-inflated negative binomial (ZINB) distribution

Abstract

Zero-modified regression models have become increasingly popular in biomedical and psychosocial research to address heterogeneity in the distribution of a count response of interest arising from a mixture study population. Like the popular negative binomial (NB) model, zero-modified models address limitations of the standard Poisson log-linear regression model. However, unlike NB, it addresses a completely different issue, structural zeros. Although the issues addressed by the two classes of models are conceptually distinct, it is quite a challenging task to tease out such differences from observed data. In this talk, we compare several model fit indices and investigate their performances when used to help distinguish the different violations of the classic Poisson model. We also discuss how these formal goodness of fit indices can be augmented by other more intuitive measures of model fit to improve the specificity of the diagnosis of the problems when analyzing such count data in practice.