The K-Zig: A Flexible Model for Zero-Inflated Counts

• Activity Number: 462
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Souparno Ghosh*+ and Alan E. Gelfand and James S. Clark and Kai Zhu
• Companies: Duke University and Duke University and Duke University and Duke University
• Keywords: Zero-inflated models ; logit link ; Posterior predictive loss function ; FIA data

Abstract

In many applications involving count data, one comes across data generating processes yielding significantly high number of zeros. Zero-inflated Poisson(ZIP) and zero-inflated negative binomial (ZINB) models are generally used to deal with such situations. However, these traditional models require very large amount of data to estimate the parameters accurately when one encounters extremely high proportion of zeros, say more than 80\%. In other words, when sample size is moderate and/or covariate information is weak, the ZIP or ZINB models are not flexible enough to handle such high proportion of zeros. To redress this problem we propose the k-ZIG model that allows more flexible modeling of zero-inflation and non-zero counts. The model is fitted within a Bayesian framework. The methodology is illustrated with simulated data examples as well as forest seedling data obtained from the Forest Inventory and Analysis National Program.