Rotation in Multiple Correspondence Analysis: a planar rotation iterative procedure

Multiple Correspondence Analysis (MCA) is a well-known multivariate method for statistical description of categorical data (see for instance Greenacre and Blasius, 2006). Similarly to what is done in Principal Component Analysis (PCA) and Factor Analysis, the MCA solution can be rotated to increase the components simplicity. The idea behind a rotation is to find subsets of variables which coincide more clearly with the rotated components. This implies that maximizing components simplicity can help in factor interpretation and in variables clustering. In PCA, the probably most famous rotation criterion is the varimax one introduced by Kaiser (1958). Besides, Kiers (1991) proposed a rotation criterion in his method named PCAMIX developed for the analysis of both numerical and categorical data, and including PCA and MCA as special cases. In case of only categorical data, this criterion is a varimax-based one relying on the correlation ratio between the categorical variables and the MCA numerical components. The optimization of this criterion is then reached by the algorithm of De Leeuw and Pruzansky (1978). In this paper, we give the analytic expression of the optimal angle of planar rotation for this criterion. If more than two principal components are to be retained, similarly to what is done by Kaiser (1958) for PCA, this planar solution is computed in a practical algorithm applying successive pairwise planar rotations for optimizing the rotation criterion. A simulation study is used to illustrate the analytic expression of the angle for planar rotation. The proposed procedure is also applied on a real data set to show the possible benefits of using rotation in MCA.