scientific article; zbMATH DE number 109595

Authors

Publication date 19 April 1993

asymptotic behavior design matrix loss function mean convergence minimum points normal linear model prediction matrix risk function selection of variables

Mathematics Subject Classification ID

Linear regression; mixed models (62J05) Inference from stochastic processes and prediction (62M20)

Review text

Regression analysis is one of the most basic and most powerful tools for almost all branches of sciences. It helps establishing relations between results and variables, preparing the way to causal interpretation and prediction. With present days large data banks the problem is however not so much to find relations between variables but to restrict the set of regression parameters to the essential ones. The paper presents some procedures for the selection of variables and investigates their asymptotic properties in a.s. and in \(r\)-th mean convergence of the loss function \(L\), the risk function \(R\) and its estimate at the minimum points of the choices of regressors \(S\). Theorems are proved for the normal finite as well as infinite linear model. Different cases of similarity or not of the design matrix \(X\) and the prediction matrix \(X^*\) are also considered. The author has developed a logical and coherent system of theorems which for the normal linear model provides some original information on the asymptotic behavior of a selected set of variables in a regression model. This should prove to be of valuable help both for the theoretician and for the practitioner of regression analysis.

Recommendations

Cited in

(13)

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 Q4024508)