Preprint #96-8



Modeling Poisson Variables with Positive Spatial Dependence

by

Mark S. Kaiser and Noel Cressie


Abstract

The Poisson auto-model is a natural vehicle for modeling data that consist of small counts and may exhibit dependence, frequently spatial dependence. Unfortunately, it is not possible to model positive dependence with a regular Poisson auto-model. We develop a model that allows positive dependencies in multivariate count data by specifying conditional distributions as Winsorized Poisson probability mass functions. This Winsorized Poisson auto-model may be used to model a set of random variables that individually exhibit Poisson-like behavior, while incorporating either positive or negative dependencies among the variables.


Copies of preprints are available from the author upon request. Use the preprint number (located at the top of the page) and make the request directly to the author, Iowa State University, Department of Statistics, Snedecor Hall, Ames, IA 50011-1210.