1996 年 44 巻 1 号 p. 44-54
This paper proposes a method for relating two sets of qualitative variables by extending weighted homogeneity analysis (WHA). WHA is a recently proposed method for partitioning a set of qualitative variables into multiple uni-dimensional groups and scaling in each group. In the extended model, multiple uni-dimensional groups of variables are constructed in each set, and at the same time, they are scaled so that the scales from one set of variables are correlated with the scales from the other set. The result is easily interpretable as compared with canonical correlation analysis which is usually used for two sets of quantitative variables. As a numerical example, the relation between the features of pictures and their impression is explored.