Interpoint Distance Comparisons in Correspondence Analysis

Abstract

Correspondence anal/sis is a metric technique for finding a spatial representation of data that has particular applicability to the analysis of cross tabulations (or contingency tables). The authors focus on some of the geometric underpinnings of the procedure and the kinds of statements that can be made about interobject similarity based on a distance model. In particular, a rationale is developed, along with a scaling of coordinates, that enables one to compare both between-set and within-set squared distances. This scaling differs from the conventional scaling currently used in correspondence analysis, where only one set of distances usually are compared.