INFORMATION THEORETIC CONTINGENCY TABLE ANALYSIS OF DORN STUDY OF SMOKING AND MORTALITY

Procedures for analyzing multiple cross-classifications, referred to as multidimensional contingency tables, lead to well-defined calculations for which well-tested computer programs are available. The log-linear models which underlie them are less restrictive than those of the multiple logistic in the sense that linearity on the logit scale is a special case, to be accepted only if warranted by the data, but easily relaxed if not. This class of models is introduced, and one version of them is applied to the analysis of the effect on mortality from coronary heart disease of a number of smoking variables.