A Pooling Methodology for Regressions in Prediction

Abstract

WVhen two or more regression equations are the same, it is advantageous to pool the data for making inference about the population regression model of interest. This paper derives the biases and mean square errors of estimation procedures subsequent to preliminary tests for the cases of pooling two or more linear regression lines, and pooling two multiple regressions. Relative efficiencies of the sometimes-pool predictor to the never-pool predictor are obtained, and recommendations of the levels of the preliminary tests are made so that the efficiency of predictions with the finalfitted regression will be at a prescribed level.