Accounting For Intersectional Social Identities

Abstract

Intersectionality theory garners increased attention from researchers interested in understanding the many ways in which oppression impacts lived experiences. In any given present and evolving context, oppression leads to advantages for some social positions and disadvantages for others (Collins & Bilge, 2016; Crenshaw, 1989). Quantitative researchers have attempted to adapt statistical modeling methods to reflect intersectional identities as a proxy for oppression and advantage in their models (Bauer et al., 2021; Schudde, 2018). This dissertation expanded on existing knowledge about the statistical limitations of three methods of modeling intersectional analyses on a continuous outcome variable: 1) Interaction, 2) Categorical, and 3) MAIDHA (multilevel analysis of individual heterogeneity and individual accuracy). Using a Monte Carlo simulation, four demographic data characteristics were manipulated to explore the three models under different scenarios which manipulated: a) the number of demographic categories (and thus intersections); b) the proportion of the sample represented by each demographic group; c) the within-intersectional-group variance in the outcome variable of interest; d) overall sample size. Each scenario and model were replicated 1000 times; results summarized performance of the intersection estimates and effect detection using the outcomes: bias, accuracy, power, type 1 error, and confidence interval coverage. The fundamental questions that guided this dissertation were: 1. What are the statistical advantages and disadvantages of each model under different demographic data characteristics? 2. In what ways does each model perform differently from one another under each demographic data characteristic condition? The findings of this dissertation contribute to intersectional quantitative research methods by providing greater insight into how each model performs under more complex data scenarios.