Additional considerations with regard to multiple regression and equal weighting

Abstract

The present paper reevaluates the relative merits of multiple regression models and the equal weighting procedure. It is argued that coefficients in multiple regression models could be replaced with equal weights only under restricted circumstances. Previous authors ( Dawes, R. M., & Corrigan, B. Psychological Bulletin, 1974 , 81, 95–106; Einhorn, H. J., & Hogarth, R. M. Organizational Behavior and Human Performance, 1975 , 13, 171–192; Wainer, H. Psychological Bulletin, 1976 , 83, 213–217) who recommended the use of equal weights have made some restrictive assumptions. In particular, they overlooked a large family of conditions, namely, those of a suppressor variable as broadly defined by A. J. Conger ( Educational and Psychological Measurement, 1974 , 34, 35–46), under which it is shown that the multiple regression model is by far superior to the equal weights procedure. In addition, equal weighting has been always ompared to the multiple regression model under the assumption that the coefficients for the latter are derived via least squares estimates. It is argued that other estimation procedures (like Ridge regression) might yield a better performance of the multiple regression model. Rather than abandoning the multiple regression model in favor of the heuristic equal weights procedure, the effort should be directed toward better estimation procedures of the regression coefficients and development of criteria to measure the “goodness” or “desirability” of such procedures.