An Introduction to Multivariate Contingency Table Analysis

Much of the data that political scientists deal with is essentially qualitative in nature, or at best ordinal. Nonetheless, we usually analyze our data as though it met the criteria of an interval scale. We proceed in this way not because we mistakenly believe our data to be interval but because the tools available for the analysis of interval data have been, at least in the past, much more powerful than the tools we have had for use with nonmetric data. With the development of new methods for analyzing nominal level data this gap has been substantially narrowed; the analyst now has a number of techniques for multivariate contingency table analysis from which to choose. This paper is intended to provide an introduction to those techniques, focusing on the linear model approach described by Grizzle, Starmer, and Koch (GSK). The paper includes the following sections: a review of simple contingency table analysis, a discussion of how the analysis of simple tables may be extended to complex tables using the GSK approach, and a brief comparison of the GSK approach to loglinear techniques such as that described by Goodman.