Reweighting intersectionality: Statistical and epistemic alignment in intersectional MAIHDA

Intersectionality offers a critical framework for understanding how multiple axes of social identity shape health outcomes. Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA) has emerged as a popular method for operationalizing intersectionality in social epidemiology. By structuring individuals into intersecting social categories (strata), and modeling both within- and between-stratum variation, MAIHDA offers a way to quantify stratum-level heterogeneity while mitigating issues of sparse data through partial pooling. In this paper, we argue that the statistical structure of MAIHDA, and its subsequent interpretation, carries epistemic commitments that are rarely made explicit. First, shrinkage in MAIHDA induces an implicit reweighting of intersectional strata toward a population of equally sized groups. As a result, estimated between-stratum variation reflects a hypothetical target population rather than the empirical population distribution, raising questions about the interpretation of stratum-level effects. Second, we argue that the variance partition coefficient and proportional change in variance are the most informative metrics specifically with regards to intersectionality, and careful interpretation of these is required. In addition, observed heterogeneity is underdetermined by any single social theory, meaning that MAIHDA findings may be consistent with multiple explanatory frameworks beyond identity-based mechanisms, such as intersectionality. We conclude that MAIHDA is best understood as a descriptive tool that identifies stratified heterogeneity. This may provide empirical guidance for when intersectional explanations are relevant, but should remain open to alternative theoretical interpretations. This perspective encourages careful epistemic reflection on the assumptions and inferences made when applying MAIHDA to intersectionality-motivated research questions.