The heuristic method for detecting clusters of respondents with different factor structure of measured variables

The new heuristic method is proposed for the identification of clusters of cases with different factor structure of measured variables. The classical cluster analysis methods try to minimize within and maximize between-groups distance, implicitly holding the assumption that all entities lie in one common space spanned with measured variables. The proposed algorithm, on the other hand, tries to divide the sample into clusters that are located in different spaces that can differ even in their dimensionality. The basic goal of the proposed method is to detect qualitative differences between clusters, and by that to obtain the new insights into the data.