Abstract
The primary goal in Correspondence analysis (CA) is to transform a contingency table into a graphical display in order to facilitate the interpretation of numerical information. As in other multivariate data analysis techniques, the aim is to explain as much variance as possible by considering the lowest possible number of dimensions. Parallel analysis (PA) is an efficient procedure which is applied to determine how many dimensions should be interpreted in a principal component analysis context. The rationale of PA is that nontrivial dimensions should have larger associated eigenvalues than dimensions derived from random data. I propose a procedure for adapting PA to the context of CA. The procedure is tested in a simulation study, and its usefulness is illustrated with a numerical example.

This publication has 24 references indexed in Scilit: