scientific article; zbMATH DE number 3996918

From MaRDI portal
Publication:4723072





Two methods for selecting variables in multiple regression are proposed. The first is used if the objective of the analysis is to predict a future response, the second method is used if the objective is to control the response at a preassigned value. In the first part of the paper a reparametrization of the regression model is introduced in order to separate the influence on the response variable due to any subset of the complete set of regressor variables from the influence of the regressors that are left out. In the second part the procedure for model selection presented in the authors' previous paper [J. R. Stat. Soc., Ser. B 46, 296-303 (1984; Zbl 0566.62004)] is applied to obtain the best subset of regressors for predictive purposes. In the third part a different criterion, based on a quadratic loss function, is introduced for the case of control. The relationship between the two selection methods is discussed and two numerical examples are presented.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4723072)