Direct comparison or indirect comparison via a series of counterfactual decompositions?

Abstract

We illustrate the point with an empirical analysis of assortative mating in the US, namely, that the outcome of comparing two distant groups can be sensitive to whether comparing the groups directly, or indirectly via a series of counterfactual decompositions involving the groups' comparisons to some intermediate groups. We argue that the latter approach is typically more fit for its purpose.