mKBO: Multigroup Kitagawa-Blinder-Oaxaca Decomposition

This paper introduces the multigroup Kitagawa-Blinder-Oaxaca (mKBO) decomposition, a methodological innovation that extends the traditional KBO approach beyond binary group comparisons. While conventional KBO decompositions require potentially arbitrary choices of reference groups that can significantly affect results, the mKBO generalization uses the sample average as a standardized reference point. This approach allows researchers to decompose differences between each group and the sample average into components attributable to differences in characteristics (endowment effect), differences in returns to those characteristics (coefficient effect), and their interaction. We address methodological challenges unique to multigroup settings, including specification issues, situations of biased sampling, and conceptual challenges relating to the use of interactions or effect heterogeneity. We illustrate the method’s utility through an analysis of racial income disparities in the United States, demonstrating how education prevalence and returns create patterns of cumulative advantage for White and Asian Americans while reinforcing disadvantage for other racial groups.