Regression analysis as an alternative to difference scores

Abstract

Listen

Translate article

For nearly 40 years, it has been asserted (see, e.g. Cronbach, 1958, 1992; Cronbach & Furby, 1970; Cronbach & Gleser, 1953; Edwards, 1994; Edwards & Cooper, 1990; Johns, 1981; Wall & Payne, 1973; Werts & Linn, 1970) that difference scores suffer from various methodological problems (about which more anon). In their position statement, Tisak and Smith argue that some of these problems have been overstated and suggest alternative procedures (e.g., the expanded difference equation) intended to overcome certain problems while maintaining the use of difference scores. Although certain points made by Tisak and Smith have merit, they minimize or overlook several important problems with difference scores, and their recommended procedures fail to overcome these problems. I will elaborate my position according to the two primary issues addressed by Tisak and Smith, the reliability and validity of difference scores. I will then note shortcomings with the Tisak and Smith procedure and contend that the regression procedure described by Edwards (1994) mitigates or avoids arguable problems with difference scores, but permits comprehensive tests of conceptual models that difference scores are intended to represent.