LAWRENCE F. FEICK* The author demonstrates the usefulness of some association models for ordered cross-classifications. The models are applied to problems common in marketing re- search, including analyzing the association in tables, determining criteria for col- lapsing categories in tables, assigning category scores to cross-classified ordered variables, and deriving correlation coefficient estimates for cross-classifications. These methods are shown to be particularly useful when the variables are partially or- dered, for example in cases in which the variables contain "don't know" responses. Analyzing Marketing Research Data with Association Models Fundamental to much of marketing research is an ex- ploration of the nature of bivariate association. Even many of the most elegant multivariate techniques used by mar- keting researchers are based on an analysis of the as- sociation between pairs of variables. The analysis of covariance structures using LISREL, for example, can be seen as an analysis of a matrix of zero-order corre- lations. Perhaps the most common exploration of the relation- ship between two variables in marketing involves the use of bivariate cross-tabulations. Particularly in applied set- tings, the bivariate cross-tabulation is used to analyze discrete (or discretized) variables at all levels of mea- surement. The use of bivariate contingency table anal- ysis persists despite a growing number of expository treatments of discrete multivariate analysis in marketing (e.g., DeSarbo and Hildebrand 1980; Green 1978; Green, Carmone, and Wachspress 1977; Perreault and Barks- dale 1980). The reliance on bivariate cross-tabulations probably has continued for several reasons. -They provide a means of data display and analysis that *Lawrence F. Feick is Assistant Professor of Business Administra- tion, Graduate School of Business, University of Pittsburgh. In gathering and analyzing the data presented in the article, the author acknowledges the support of USDA/SEA competitive grant 5901-0410-9-0299-0 to Rex H. Warland and Robert 0. Herrmann. A draft of the article was prepared while the author was on Faculty Re- search Grant from the Graduate School of Business at the University of Pittsburgh. The author thanks Clifford Clogg, William Dillon, Ter- ence Shimp, Rex Warland, and anonymous JMR reviewers for their helpful comments. 376 is clearly interpretable even to the less statistically in- clined researcher or manager. -A series of bivariate tabulations provides insights into complex marketing phenomena which might be lost in a single multivariate analysis. -The clarity of interpretation affords a more readily con- structed link between market research and market ac- tion. -Consideration of bivariate cross-tabulations may lessen the problems of sparse cell values which can plague the interpretation of discrete multivariate analyses. We attempt to show the usefulness of models devel- oped by Goodman (1979) for the analysis of bivariate association in contingency tables. Though these models were devised for the analysis of association in doubly ordered contingency tables, they can be applied fruit- fully to singly ordered tables and tables in which one or both of the variables are partially ordered. Because of the prevalence of ordinal data in marketing research, these models have the potential for rather wide applicability. Several alternative measures are available for assess- ing the extent of association in a contingency table. An introduction to some of these measures is provided in Chapter 8 of Mueller, Schuessler, and Costner (1977). Underlying the models presented in this article is the premise that the decision to use a single measure of the association between the variables is, in itself, a testable hypothesis. We demonstrate that though a single mea- sure of association may sometimes be appropriate, there are situations in which a single measure cannot ade- quately characterize the association in the table unless certain assumptions are made. In the latter situation, the researcher in some cases will want to view the associ- Journal of Marketing Research Vol. XXI (November 1984), 376-86