Raw Score and Factor Score Multiple Regression: An Evaluative Comparison

Abstract

An empirical investigation and theoretical examination was made of the issues arising out of the use of factor scores in multiple regression from the perspective of exploratory analysis. Using data collected on 189 subjects, two raw score regressions were compared to and contrasted with a single factor score regression. Issues examined included: the relative predictive power and generalizability of each solution, the effect of factor score regression on the degrees of freedom in the analysis, and for each solution the relative ease of interpretation and loss of predictors. It is argued that in cases of consistently large commonalities across a large predictor set factor score regression is the more appropriate solution. In cases of consistently low commonalities across a small predictor set a raw score solution appears the more appropriate. With various sizes of commonalities, semi-partial correlations, and zero-order correlations, "research specific" requirements of the exploratory analysis will often predetermine the priority to be given one or other regression solution.