Date of Award
Doctor of Philosophy
When using multiple regression models for predictive purposes, it may be desirable to exclude some possible regressor variables in order to reduce both the mean squared error (MSE) of the predictor and the cost incurred in taking observations. Some of the methods that have been proposed for this problem are discussed. The method of Lindley (1968) is detailed and extended to include the case of unknown variance (sigma)('2). A class of informative priors for the vector of unknown regression parameters is included;An evaluation of Lindley's predictor, with respect to bias and MSE, which recognizes that the predictor variables have been selected, is considered. For the case of a predictor based on two observed regressors, the numerical evaluations of the bias and the MSE of the predictor, conditional on selection, are performed for various configurations of data. The conditional bias and MSE of the predictor are numerically compared with the unconditional bias and MSE of the same predictor.
Digital Repository @ Iowa State University, http://lib.dr.iastate.edu/
Peter David Christenson
Christenson, Peter David, "Variable selection in multiple regression " (1982). Retrospective Theses and Dissertations. 8338.