Zero-Inflated Models

This chapter discusses the zero-inflated models as allowing the data to arise from two distributions also: the first is a zero with certain probability and the second is a distribution that includes zero in the support. Zero-inflated models are thought of as being used to model data sets containing excess false zeros. In cases where the zeros may arise from both distributions in a mixture, either from an explicit zero process or from another distribution that contains zero in the support, it is common to use a zero-inflated Poisson model. In addition, studies have extended the basic zero-inflated (ZI)-Poisson model to accommodate explicit spatial structure. The chapter focuses on checking the assumptions for the ZI-Poisson and ZI-negative binomial models to see which is most appropriate for statistical inference. The resulting posterior predictive p-values were 0.03 for the ZI-Poisson model versus 0.73 for the ZI-negative binomial model.