589 – New Methods of Modeling Count Data and Its Impact on the Future Analysis of Health Data
Mean and Variance Modeling Using Extended Poisson Process Models
David Smith
Truven Health Analytics
A family of models based on the extended Poisson process that can flexibly handle both under and over dispersion compared to the Poisson and negative binomial distributions will be described. Many sets of count data display such under or over dispersion and, although there are a number of distributional models available that can handle over dispersion, there is a lack of models that can handle under dispersion. Models with mean and variance related to covariates can also be constructed within this family using a generalized linear model formulation; estimation of parameters being by maximum likelihood. An R package for fitting such models will be described, and its use to analyze health outcomes and other types of health related data illustrated.