Home » Additional Features, JEDI Corner

The JEDI Corner: Statistics for Equity: Capturing, Not Masking, Intersectional Dynamics in Data

1 November 2021 2,023 views No Comment
Stephanie H. Cook, Suzanne Thornton, Samantha E. Robinson, James J. Cochran, and Godwin Yung

The JEDI Corner is a regular component of Amstat News in which statisticians write about and educate our community about JEDI-related matters. If you have an idea or article for the column, email JEDI Outreach Group member Cathy Furlong.

Statistics is the language of data. Just like any other language, statistics requires structure and rules for interpreting messages from data. However, statistics poses additional communications challenges, as it depends on one’s written language, which is always rife with nuances. Hence, the effectiveness of statistics for communicating concepts from data (e.g., selecting, understanding, and interpreting statistical models) also depends on one’s written language acuity.

Another concept that continues to present communication challenges in the world of statistics is intersectionality, which refers to the overlapping nature of multiple oppressions based on one’s identity (e.g., gender, race/ethnicity, sexual orientation, age). Conceptions of intersectionality posit that intersecting identities and processes (e.g., systems of sexism such as laws that limit the individual’s ability to make decisions about their bodies) are both vitally important to understanding inequity, according to Greta Bauer and Ayden Scheim in “Advancing Quantitative Intersectionality Research Methods: Intracategorical and Intercategorical Approaches to Shared and Differential Constructs,” published in Social Science & Medicine.

To properly measure and analyze concepts related to intersectionality, the educated statistician must understand the benefits and limitations of historical and current approaches while contributing to the development and use of validated new statistical methodologies that can more accurately capture the embedded, multi-level nature of intersectionality.

Early approaches to measuring intersectionality, mostly in sociological and health-related research, used the additive approach, which combines social categories (e.g., gender and race) by adding new terms to statistical models. This early approach signaled a statistical appreciation for the need to measure and assess diverse features of the human experience.

The additive approach inherently assumes that occupying multiple social statutes at once (e.g., being both a female and a sexual minority) “add” to increase poor health, for instance. This is in direct contrast to social theorists who posit that intersectionality is multi-level and positional and, decidedly, not additive.

For instance, Kimberlé Crenshaw, in her 1990 Stanford Law Review article, “Mapping the Margins: Intersectionality, Identity Politics, and Violence Against Women of Color,” cites the example of a Black woman who was unable to successfully defend her case on employer discrimination based on her identity as both a Black American and a woman because the legal protections in her favor treated discrimination based on race and gender in separate manners; that is, the effect of being a Black American differs by gender. From this perspective, additive experiences do not adequately speak to the positionality of one who occupies multiple social statuses.

Of course, there are exceptions and cases in which statisticians are examining the accumulation of an experience on an outcome (e.g., the additive effect of racial and sexual orientation–based discrimination). But when statisticians are specifically attempting to measure, model, and understand intersectional social processes themselves, they must appreciate that an additive approach is not appropriate.

Today, many statisticians—and researchers more broadly—use the multiplicative approach to understand intersectionality. In this approach, interaction terms between different groups (e.g., Black and woman) are used to understand the differences in an outcome compared to another group (e.g., white and male). However, scholars such as Clare Evans, who wrote the Social Science & Medicine article titled, “A Multilevel Approach to Modeling Health Inequalities at the Intersection of Multiple Social Identities,” as well as Bauer and Scheim have pointed out the theoretical and empirical limitations of such a method.

First, this method does not allow us to statistically account for the intrinsic intersections of social disadvantages and privileges that may interact within a given group. For instance, white females may experience some measurable privilege based on their whiteness while also experiencing disadvantage due to being female. In another example, a gay and white man may experience sexual orientation discrimination based on his sexual minority status, but also have privileges afforded to him for being white and a male. The widely used multiplicative statistical approach to evaluating intersectionality cannot capture these nuanced, but critical, aspects of the effect of intersectionality on an outcome.

Second, this approach can suffer from small sample size bias, interpretability problems, and model fit issues.

Current approaches to measuring and modeling intersectionality in statistical analyses are promising but, unfortunately, underused. One of the more widely used methods of modeling the complex nature of intersectionality is multi-level modeling. Scholars such as Tom Snijders, Roel J. Bosker, Hae Yeon Choo, and Myra Ferree have used multilevel modeling to examine how social structures interact with individual identities to produce inequities. This powerful tool gives us the ability to describe variability both within particular social processes and between intersecting social processes.

In one example, Evans and colleagues employed a multi-level modeling approach to understand the intersections of gender, race/ethnicity, income, education, and age. The authors were able to identify 384 intersecting interactions and compare these to the main effect of each variable on body mass income using the National Epidemiologic Survey on Alcohol and Related Conditions. Through this inquiry, the authors were able to explore the interactions of intersections between and within categories of gender, race/ethnicity, income, education, and age. In other words, multi-level approaches allow us to explore the variance between and within intersectional categories rather than treating intersectionality as a fixed effect.

Though multilevel modeling holds much promise in the study of intersectionality research, it has drawbacks. One significant drawback is the reliance on large sample sizes and a small number of categories (and levels within the categories). Hence, to understand the complex nature of intersectionality, statisticians must use validated existing approaches while continuing to develop new approaches that address some of the known limitations.

Another drawback is that, despite the existence of these multilevel modeling techniques, many researchers continue to rely on techniques they are more familiar with (e.g., the multiplicative or additive approaches), which ultimately limits our ability to understand the ways in which multiple intersecting statuses and processes overlap to produce varying degrees of inequity.

This overview of statistical approaches to study intersectionality is by no means exhaustive. There are related modeling approaches that support the study of intersectional processes at multiple levels, including moderated mediation and latent class analysis.

It is clear, however, that the statistics field already plays an important role in communicating intricate concepts such as Intersectionality. Effective communication begins with education. Students, academics, and professionals in the inherently collaborative discipline of statistics must understand the meaning of intersectionality before we can appropriately measure and analyze it. Such knowledge is what enables a statistician to understand the difference between measuring multiple statuses and social processes (e.g., additive models) and intersecting statuses and social processes (e.g., multiplicative models).

Once educated on intersectionality, the statistician must become familiar with different approaches for measuring and modeling intersectionality. Tutorials, workshops, and presentations within both academic and other professional settings will help the statistician inform practitioners about the many tools available. Professional organizations such as the ASA’s Justice, Equity, Diversity, and Inclusion (JEDI) Outreach Group provide a space for this dialog by focusing explicitly on intersecting identities and systems in relation to our discipline.

Statistics can help make the invisible, visible. But we first must remove any and all stigmas associated with discussing power, privilege, and marginalization in statistical inquiry. These stigmatizations limit our ability to communicate about intersectional identities and social processes with one another and the broader scientific community, especially in relation to methods pertaining to intersectional identities and social processes. Once conversations around equity and inclusion, for instance, become endemic, it will become easier to have open and direct dialogs about intersectionality. Advances in education and research from a statistical point of view can contribute to a larger discussion on intersectionality for the ultimate goal of improving equity.

Further Reading

1 Star2 Stars3 Stars4 Stars5 Stars (No Ratings Yet)
Loading...

Comments are closed.