A Comparison of Regression and Loading Weights for the Computation of Factor Scores

Abstract

An alternative strategy for computing factor scores was introduced and compared to a popular contemporary scoring procedure. The new strategy involved unit-weighted composites of the standardized items that possessed salient factor score coefficients. Within the context of a sampling model, this strategy was shown to be superior to the common method of computing factor scores by unit-weighting and summing the standardized items with salient factor structure coefficients. Specifically, the new strategy produced factor scores that (a) captured a greater proportion of the true score variance of the factors, (b) were less confounded by true scores from factors other than those they were supposed to be estimates of, and (c) were less correlated with one another when the underlying factor structure was truly orthogonal. The implications of these findings were discussed within the context of two general applications of factor analysis, and practical recommendations were offered.