Part II of the annotated bibliography covers methodological discussions from an empirical point of view, including reviewing and implementing specific methods.
This article is a systematic review of 16 published studies using latent class analysis, latent profile analysis and clustering algorithms, which the authors identified as methods that group individuals’ characteristics following the intersectionality principles. The review revealed that most articles applying these methods, called by the authors ‘person-centred’, identified four classes/clusters by combining indicators of identities, (social) positions and processes such as stigmatisation and discrimination. Health-related research commonly uses these outcomes for sociodemographic comparisons or as predictors in linear (regression) models. The paper discusses theoretical and methodological rationales behind using these techniques, particularly how well they ‘translate’ the core tenets of intersectionality.
This study compares the performance of two-level multilevel models (MLM) and standard fixed-effects regression estimated via ordinary least squares (OLS) to study intersectionality. As illustrated by the authors, both methods have potential drawbacks. OLS regressions with interaction terms for dummy variables risk identifying statistically significant interactions by chance, especially with smaller sample sizes. On the other hand, MLM models that include dummy variables for each interaction may shrink effects towards the mean due to the uncertainty of their estimates. Additionally, MLM assumes that level-two residuals (the difference between estimated and predicted values) are independent and identically distributed. The performance of MLM and OLS is compared via simulations under different conditions, including those that violate the assumptions for implementing MLM. Simulations showed that MLM outperforms OLS regressions, but shrinkage on MLM may wrongly identify some intersectional effects. The article closes with guidance to enhance the robustness of MLM for the study of intersectional effects for predicting outcomes.
In this article, the authors explore the application of the intersectional MAIHDA (multilevel analysis of individual heterogeneity and discriminatory accuracy) previously introduced by the authors as a strategy to apply MLM models for studying intercategorical inequalities. MAIHDA treats intersecting categorisations as strata, which among other advantages, allows it to better deal with higher dimensionality (combinations of multiple categories). Framed as a response to a critique of the approach, the authors refer to simulations and empirical applications, emphasising MAIHDA’s advantages over single-level approaches. The study complements Bell et al (2019) by providing further explanations on interpreting fixed effects regression coefficients and highlighting the advantages of using multilevel models for analysing high-dimensional interactions
In this book chapter, the author starts by exploring ‘why’ and ‘how’ intersectionality is studied quantitatively using survey data. Dubrow suggests that despite the common limitations of surveys, such as small sample sizes for intersecting categories, hypotheses on cumulative and group-specific disadvantages can be quantitatively tested. Dubrow tested intersectional hypotheses by calculating the international socioeconomic index for combinations of intersecting categories and utilising unpaired t-tests. The simple, yet effective, statistical analysis is a call to action for social scientists, especially those interested in social stratification, to innovate when implementing quantitative methodologies that better reflect the realities of individuals’ lived experiences.
Gross and Goldan highlight the challenges and opportunities presented by applying intersectionality in quantitative studies, contrasting it with qualitative approaches. The authors offer a schematic introduction to OLS, MLM and MAIHDA, including reflections on integrating intersectionality into quantitative research methodologies and evaluating their effectiveness in capturing complex social dynamics. The paper calls for a deeper understanding of intersectionality among quantitative researchers and advocates for improved methodological practices to better address social inequalities.
Mahendran et al explored the application of statistical methods to study intersectionality in health inequalities. Descriptive and causal methods were reviewed, including cross-classification (individuals are assigned to groups based on the combination of two or more variables), decision tree algorithms (non-parametric methods that set decision rules for data segmentation) and regression with interactions and MAIHDA. Using simulations, the authors test how well the prevalence of an intersection-specific dependent variable is estimated by each method. Findings vary depending on sample sizes for simulated scenarios, but broadly point at MAIHDA, followed by conditional inference trees and chi-square automatic interaction detector as the better methods for smaller samples.
In this article, the authors investigate the use of regression models, including OLS with interaction terms, separate interaction regressions for different contexts, and effect decomposition, as well as two-level MLM. Taking a feminist approach to intersectionality, Scott and Siltanen evaluate how context, openness and the complexity of the structuring of inequality are operationalised by each approach. Although the authors acknowledge the inherent limitation of predictive/causal models, they concluded that using the proposed feminist-inspired criteria MLM better aligns with feminist literature on intersectionality.
In this article, Schudde examines the methodological complexities of integrating intersectionality into quantitative research within the field of education. Schudde critiques the reliance on secondary survey data, which often limits researchers’ ability to capture the full spectrum of intersectional identities, and the limitations of grouped results to account for unequal experience. The literature review is focused on varying effects (‘heterogeneous effects’) of identities and context in higher education. The author found that most papers in the field used regression models with main effects and a minority used interaction terms. Another salient method is propensity score models, which estimate the probability of an outcome based on the combination of observed characteristics. More widely, the author identifies significant challenges, including the difficulty of justifying complex models in academic publishing and the limitations of existing datasets.
This chapter explores the multiple ways to understand the link between statistical methods and operationalisations of intersectionality. The author overviews some of the statistical methods recently used to implement intersectionality divided into three groups. The first (‘minimalist’) views the methodological aspects of the study of intersectionality as merely of a technical nature. A step forward is to take a theoretically informed (‘reflective’) approach in which interest moves to the link between theorisations and methods, considering social contexts. A third approach, referred to as ‘radical’, focuses on the innovative use of quantitative methods which better fit specific forms of intersectionality. Ultimately, Spielrings calls for openness in access to relevant datasets and knowledge for progressing the quantitative exploration of intersectional studies.