Constrained Coefficient Component Regression Analysis and Its Application(<Special English Issue>-Information and Operations Management)

概要

論文の詳細を見る
In this paper we propose a new approach to regression analysis. In the proposed approach the basic model is transformed into several independent partial models through orthogonal transformations of dependent variable and regression coefficients. In the transformed model, we treat the coefficient components, which are defined in a set of linear combinations of the regression coefficients, as the parameters instead of the regression coefficients. Because every coefficient component is included in an independent partial model, each can be manipulated independently. Hence, model building no longer must be considered a combination of coefficient components. This not only reduces the number of conceivable candidate models, making model selection easier, but it also leads to an improvement in estimates of parameters and the possibility of finding better models. We also analyze the relationship between the productivity of rice farming and the temperature in Hokkaido as an illustrative application of the proposed method. The estimation results suggest the advantages of the proposed approach.
社団法人日本経営工学会の論文
2006-02-15