An analysis of categorical repeated measurements in the presence of an external factor

Abstract

SUMMARY This paper provides a log-linear method for the analysis of contingency tables where the population of individuals is divided into distinct subpopulations and where repeated measurements are made on each individual. The proposed model for the full joint probabilities is quasi-symmetry within each subpopulation; the model is derived from one of two within-individual models. The method by which the hypotheses that flow from the model are tested is an extension of the test of marginal symmetry, achieved by testing complete symmetry conditional on quasi-symmetry.