2006 年 33 巻 2 号 p. 179-192
A method of generalized constrained redundancy analysis (GCRA) is proposed, which incorporates external information in redundancy analysis (RA). In this method both the criterion variables and the orthogonal projector defined by the predictor variables are first decomposed into several components according to the external information, and RA is applied to the decomposed matrices. By combining the terms in the two decompositions, a variety of existing and new methods of RA are realized including a variety of partial (non-partial, partial, semi-partial and bi-partial) and constrained (unconstrained, semi-constrained and bi-constrained) RA. An example is given to illustrate the method.