Change Scores—Necessarily Anathema?

Abstract

Psychologists have been warned repeatedly in recent years of the hazards of change scores. Unfortunately, these warnings seem to have created the belief among many researchers that the use of change scores is universally misleading and therefore should be avoided at all costs. However, the use of change scores is perfectly legitimate in certain circumstances and at times may even be preferable to other methods of analysis. The ANOVA of change scores is acceptable in randomized pretest-posttest designs, where it is equivalent to a repeated measures approach to the data. In addition, the unreliability of difference scores poses no problem here. Despite this, the analysis of covariance is generally preferred, because it is more powerful. However, in at least two situations the analysis of change scores is preferable to the analysis of covariance.