Hello

,

It would be very helpful for me to know if one has to dichotomize (0,1) nominal variables before running MCA in SPSS 13. Thank you!

I think I found the answer in the paper "Composite Indicator of Multidimensional Poverty" of Louis-Marie Asselin, pp. 12-13,

http://pep-net.homeip.net/NEW-PEP/Group/PMMA/pmma-train/files/Multi-Dim-Pov-Doc 1.pdf
"MCA (Multiple Correspondence Analysis), also called Homogeneity Analysis, is applied to categorical or nominal data, where each variable zj is assumed to have kj distinct categories.

From the distance analysis viewpoint, the crucial part of the optimal transformation is performed beforehand: each of the m variables zj is replaced by an n×kj orthogonal binary matrix Gj … As in canonical distance analysis, a Mahalanobis metric is implicit … In this special case, since the columns of Gj sum to 1, the Mahalanobis metric is equivalent to the χ2 metric" [Meulman (1992), pp. 550-551]. Thus, here, the M sets of variables correspond to the m original categorical variables, once broken down each in their mj constituent categories becoming as many dichotomous "variables". Thus, the problem of "same" measurement units for the variables involved in the computation disappears: all variables are binary. In fact, from the computational point of view, MCA is a CA analysis applied to the 0/1 matrix generated from the categorical indicators, what is precisely named the "indicator matrix". "