Quasi-Independence and Maximum Likelihood Estimation in Incomplete Contingency Tables

Abstract

Many authors have been concerned with contingency tables containing cells which are missing, a priori zero or otherwise specified. This article examines the problem of maximum likelihood estimation for such tables under the “quasi-in-dependence” model. In particular, conditions are provided to ensure the existence of unique nonzero maximum likelihood estimates for the cells of incomplete tables obtained by deleting the missing or a priori zero cells, even when other cells contain zero counts due to sampling variation.