Department of Statistics, UCLA

We discuss several forms of Nonlinear Principal Component Analysis (NLPCA) that have been proposed over the years: Linear PCA with optimal scaling, aspect analysis of correlations, Guttman’s MSA, Logit and Probit PCA of binary data, and Logistic Homogeneity Analysis. They are compared with Multiple Correspondence Analysis (MCA), which we also consider to be a form of NLPCA.