ABSTRACT
Correspondence analysis provides visualizations of associations in a two-way contingency table in a small number of dimensions. Multiple correspondence analysis extends this technique to n-way tables. Other graphical methods, including mosaic matrices and biplots, provide complementary views of loglinear models for two-way and n-way contingency tables, but correspondence analysis methods are particularly useful for a simple visual analysis. Correspondence analysis has a very large, multi-national literature and was rediscovered several times in different fields and different countries. The method, in slightly different forms, is also discussed under the names dual scaling, optimal scaling, reciprocal averaging, homogeneity analysis, and canonical analysis of categorical data. Mathematically, correspondence analysis is related to the biplot, to canonical correlation, and to principal component analysis. Correspondence analysis shows only row and column categories as points in the two dimensions that account for the greatest proportion of deviation from independence.
