The statistical advantages of MAIHDA for estimating intersectional inequalities

Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA) is a multilevel regression modeling approach rooted in intersectionality theory. It is used to examine inequalities across intersections of multiple social identities (e.g., gender, ethnicity, social class). Proponents argue that MAIHDA provides more accurate predictions of population means for these groups than the conventional approach of calculating the simple arithmetic mean for each intersectional group—either directly or via a linear regression model that includes all possible interactions between the social identities. In this study, we aim to evaluate and demonstrate the validity of this claim to the community of quantitative intersectionality researchers. We do so by analytically comparing the variance and correlation of simple and MAIHDA-predicted means with population means. Additionally, we assess the bias, variance, and mean squared error in predicting the mean of a given intersection. Our findings show that MAIHDA-based means outperform simple means, particularly when using the MAIHDA means that decompose intersectional means into additive and non-additive effects of social identities. However, the relative advantage of MAIHDA means depends on the nature of the intersectional inequalities and the sizes of the intersectional groups being studied. MAIHDA’s benefits are most pronounced when inequalities are subtle or when data on certain intersections, such as those for marginalized groups, are sparse. These conditions are common in practice, highlighting the practical significance of our findings.