Why Not to Use Difference Scores in Research on Affective Polarization: A Critique and a Tutorial for Using Polyvariate Regression to Study Congruence as an Outcome

Affective polarization, people’s emotional attachment to their own party and dislike of rival parties, has been a growing area of interest in recent decades. However, studies on affective polarization have tended to use difference score measures to capture affective polarization. Specifically, it is common to use the difference between people’s rating of their own party, and their ratings of a rival party (or parties). It has long been known that such methods are problematic for measuring any variable and the extensive conceptual, statistical, and theoretical problems with using difference scores are here reviewed at length. Then, a tutorial for polyvariate regression is provided. Polyvariate regression is an analytic approach that allows the researcher to test how a predictor is associated with affective polarization without using difference scores. This is then demonstrated by testing the association of people’s stances on government defense spending with feeling thermometer ratings of the Republican and Democratic parties in the United States (U.S.), both to help researchers in this area to use it in their work and to demonstrate further how using difference scores may obscure possibly interesting findings. The same association is tested with a difference score measure as the outcome variable. The outcomes of both analyses are compared, and show the superiority of using polyvariate regression over difference score measures. The application of polyvariate regression in multiparty contexts and under other conditions is discussed, as are limitations of polyvariate regression. Polynomial regression is also recommended as an approach to investigating affective polarization when polarization is the predictor rather than the outcome.