Analysis of extra-omultinomial data derived from extra-Poisson variables conditional on their total

Abstract

SUMMARY A model for extra-binomial variation is derived from conditioning two extra-Poisson variables on their total. Approximate moments are found, and from these a quasilikelihood is constructed for estimation and performing analyses of deviance. Comparisons are made with other models for extra-binomial variation. The idea is extended to a model for extra-multinomial variables.