Correspondence analysis applied to grouped cohort data

In this paper we investigate application of generalized correspondence analysis (GCA) to model a two‐way table of rates consisting of a cross‐classification of case numerators and person‐time denominators. The GCA reconstitution formula is proposed as a correlation model for analysing these data. This provides two major advantages over the techniques usually applied. The number of interaction parameters to be estimated is reduced, and, moreover, they have useful interpretation; the singular values and the scores are measures of row‐column modification/interaction effects; in addition the scores allow graphical display of these effects. An asymmetrical plot is proposed for optimal representation of interaction parameters as distances. Finally, two examples are presented to illustrate the method of analysing a two‐way table of covariate by exposure and a two‐way table derived from a cross‐classification of two exposure factors.