An Application of Stagewise and Stepwise Regression Procedures to a Problem of Estimating Precipitation in California

Abstract

This paper illustrates a method of blending stagewise and stepwise regression procedures for selecting predictors and deriving equations. Only a few minutes of large-scale computer time were required to screen almost 4500 potential predictors and identify the leading contenders for use in the prediction equations. The equations developed for specifying and predicting California precipitation produced estimates of the amount of precipitation which were distinctly superior to estimates based on persistence or recurrence.