A Heuristic Method for Estimating the Relative Weight of Predictor Variables in Multiple Regression

TLDR

Determining the relative weight of predictors in multiple regression is challenging because intercorrelations obscure each variable’s contribution, and although several measures exist, they become impractical when more than about five predictors are involved. The authors propose a computationally efficient heuristic that estimates predictor relative weights for any number of variables and provide guidance on its appropriate use and limitations. The heuristic yields relative weight estimates that closely match those from more complex, established methods.

Abstract

The relative weight of predictor variables in multiple regression is difficult to determine because of non-zero predictor intercorrelations. Relative weight (also called relative importance by some researchers) is defined here as the proportionate contribution each predictor makes to R2, considering both its unique contribution and its contribution when combined with other variables. Although there are no unambiguous measures of relative weight when variables are correlated, some measures have been shown to provide meaningful results (Budescu, 1993; Lindeman, Merenda, & Gold, 1980). These measures are very difficult to implement, however, when the number of predictors is greater than about five. A method is proposed that is computationally efficient with any number of predictors, and is shown to produce results that are very similar to those produced by more complex methods. Recommendations are made for when this procedure may be applied and in what situations it is not appropriate.