Zero-inflated Poisson and negative binomial models for count data with extra zeros
Wei Fan, M.S. Candidate, Department of Statistics, Colorado State University.
Tuesday, November 11 2008
11:00 a.m. 223 Weber
The Poisson regression model has been used extensively for the analysis of count data. In practice, however, count data often has a far more zeros than expected for the Poisson distribution. When this phenomenon arises the variability of the counts greatly exceeds the mean under the Poisson assumption, resulting in substantial bias for the parameter estimates. In this project, we consider the problem of modeling count data with extra zeros. Zero-inflated models and hurdle models provide useful improvements for such data. We briefly review negative binomial model, zero-inflated Poisson model, zero-inflated negative binomial model, Poisson hurdle model, and negative binomial hurdle model, and compare these models with standard Poisson model. Three examples are used to illustrate the advantages of zero-inflated models and hurdle models in fitting such data.
Phil Chapman: Adviser
Haonan Wang: Committee Member
Annettte Bachand (Environmental & Radiological Health Sciences)