Affirmative Action with Multidimensional Identities

Affirmative action policies are widely employed in college admissions, hiring, and other decisions to reduce underrepresentation of disadvantaged groups. Existing policies predominantly treat each identity dimension (e.g., race, gender, caste) as independent. We find that generically such nonintersectional policies cannot eliminate underrepresentation. Under certain conditions, every nonintersectional policy worsens the representativeness of at least one intersectional group. Accounting for interactions between identity dimensions, we construct intersectional policies that achieve a representative outcome. Nonintersectional policies can, however, eliminate underrepresentation along each identity dimension, despite preserving (and potentially disguising) underrepresentation among intersectional groups