Abstract

This paper deals with the oblique component method which has some features of the method of a cluster analysis and a factor analysis. The clustering problem of variables is as follows: for a fixed number of p groups to construct a partition of variables X1,...,xk into disjoint groups G1,....,Gp in such a way that the sum of the largest eigenvalues of correlation matrices for group varibles will be maximal through all possible divisions. Simultanously we get first principal components of variables in groups which we can treat as common factors. So this method enables to do a cluster analysis and a factor analysis for the set of variables. Two algorithms of the clustering and the criterion for the optimum of the number of groups are also presented in the paper. At the end there is an example of the application to the hypothetical correlation matrix. Verf.-Referat