1. HIERARCHICAL MODELS FOR SIGNIFICANCE TESTS IN MULTIVARIATE CONTINGENCY TABLES: AN EXEGESIS OF GOODMAN'S RECENT PAPERS.
- Author
-
Davis, James A.
- Subjects
MATHEMATICAL models ,MULTILEVEL models ,LINEAR statistical models ,MATHEMATICAL statistics ,SOCIOLOGISTS - Abstract
The article provides an explanation of sociologist Leo Goodman's paper hierarchical models for significance tests in multivariate contingency tables. In several papers Goodman presents and elaborates a system for the analysis of contingency tables that promises to be extremely useful to sociologists. The papers are however, extremely compressed, heavily complex with the cumbersome notation of contingency analysis and are not easily accessible to the student and average research worker. The aim of this article is to explain the logic and procedures of the system in terms the reader may find more comfortable. The Goodman system consists of two parts: a scheme for making significance tests by means of hierarchical models and an extensive discussion of a set of techniques known as log linear models. The two parts are logically and practically distinct. Among the important uses of hierarchical models are the following: tests for the significance of partial correlations, tests for interactions, where the control variable has as many categories as one pleases, tests for higher-order interactions, succinct statements of what is and what is not going on in a contingency table. Since none of these tools have been easily available to the average sociologist, the Goodman system is well worth learning especially because it provides considerable insight into the properties of cross-classifications and the logic of significance tests.
- Published
- 1973
- Full Text
- View/download PDF