1985 年 12 巻 18 号 p. 39-51
A method of simultaneous scaling of categories is proposed in order to disclose the structural interdependency of a set of discrete variables, in which categories within one discrete variable are regarded as a set of dummy variables. From this viewpoint, we propose a tentative method of simultaneous scaling by the use of generalized canonical correlation analysis for several sets of variables proposed by Horst (1961). An application of the generalized canonical correlation analysis to discrete variables is demonstrated by making use of the following three different criteria; i.e., (a) Dual scaling, (b) the sum of the correlation coefficients and (c) the sum of the squared correlation coefficients.