Selection of categorical variables using correspondence and <em>procrustes</em> analysis

Abstract

Correspondence analysis is a technique of multivaried analysis (particularly, a method of factorial analysis to categorical variables) that allows us to obtain a graphic representation through the distribution of the scores from the categories of lines and/or columns in a system of coordinates. Krzanowski (1987) presented a methodology that combines the analysis of principal components with procrustes analysis to determine how much the new subset of variables represents the structure of the original data. The aim of this study is to propose a procedure that applies to the procrustes analysis combined with the correspondence analysis to find a rank of importance to the columns (variables) of a contingency table. That procedure has been applied according to an example by Krzanowski (1993). Some conclusions are presented about the behavior of the procedure.