It is difficult to interpret a correlation when one variable is a difference score. We show in this paper how the same correlation can arise from many different patterns, with each pattern implying a different interpretation. We give examples that arise in research on dyads (e.g., studying the relation between husband and wife salary on marial satisfaction). We provide recommendations for how to test research questions involving differences and discrepancies.

Griffin, D., Murray, S., & Gonzalez, R. (1999). Difference score correlations in relationship research: A conceptual primer. Personal Relationships, 6, 505-518.

doi:10.1111/j.1475-6811.1999.tb00206.x PDF

Abstract

The practice of computing correlations between “difference” or “discrepancy” scores and an outcome variable is common in many areas of social science. Relationship researchers most commonly use difference scores to index the (dis)similarity of members of two-person relationships. Using an intuitive, graphical approach-and avoiding formulas and pointing fingers-we illustrate problems with using difference score correlations in relationship research, suggest ways to ensure that difference score correlations are maximally informative, and briefly review alternatives to difference score correlations in studying similarity, accuracy, and related constructs.