In my work, I am continually trying to find new ways to research marginalized communities in ways that are inclusive and reflective of current best practices. In particular, I have often struggled with quantitative research, which often necessitates grouping people into boxes and operationalizing rigid definitions of identity. Similarly, I have struggled in accepting conventional norms in quantitative research that teach analytical methods that compare marginalized communities as a conglomerate to their privileged counterparts (e.g., people of color vs. White, LGBQ vs. Straight, transgender vs. cisgender). Recently, I learned about effect coding as a way of mitigating some of my concerns (thank you NSSE/FSSE/CUTE Research Scientist Allison BrckaLorenz!).
Put simply, effect coding allows researchers to compare groups of data to the average (mean) of the sample, as opposed to another group of data (Mayhew & Simonoff, 2015). This contrasts with indicator (dummy) coding, which uses a single group as a reference group. This is particularly useful when comparing groups of participants, separated by identity. For example, in indicator coding, a researcher might use white students as the standard by which all other racial groups are then compared with. However, this can be problematic in that it privileges the narrative or experiences of a single group of people. Through effect coding, rather than comparing racially marginalized groups to white people, all selected racial groups are compared to the average score between them.
To be sure, effect coding is not perfect. For instance, if a sample of 5,000 students has 4,000 white students and 1,000 students of color, the average of the sample would still lean towards the scores for white students, and therefore might still privilege white people in the analysis. Nonetheless, by removing some of the inherent power dynamics from the start, effect coding can serve as a more inclusive method of data analysis.
In my own research studying LGBTQ communities, this is especially helpful so as to not perpetuate the privileging of cisgender and straight voices in my analyses. Generally, in my work, I am less interested in comparing LGBTQ students with non-LGBTQ students and more interested in studying within-group differences. And while effect coding certainly feels more appropriate for me to use, I also recognize that bisexual students make up an increasingly large majority of the LGBTQ community and so any comparison to the average would skew towards the scores of bisexual students. Nonetheless, this feels more appropriate than the typical alternative: drawing comparisons toward gay and lesbian students as the reference group.
As another example, Allison BrckaLorenz, Ella Chamis (a FSSE/CUTE Research Project Associate), and I conducted a study, which used data from the College + University Teaching Environment (CUTE) Survey in order to examine the affective components of a faculty environment for queer faculty, faculty of color, and queer faculty of color (BrckaLorenz et al., 2023). In this project, rather than compare the coefficients for white faculty with scores for faculty of color and the coefficients for straight faculty with scores for queer faculty, we effect coded the demographic variables so that each identity could be compared to the average score for faculty in the model. In our view, this was the most inclusive way to ensure that each racial and sexual identity was treated as equally as possible in our statistical model, without affording more power or privilege to a single identity.
At the end of the day, effect coding is one tool in the toolbox of analytical methods. While effect coding is not perfect, nor is it the solution to every research question, it can offer individuals a way to make critical decisions about the statistical models they wish to create in order to study minoritized populations. And in quantitative research, a little care and intentionality can go a long way.
References
BrckaLorenz, A., Chamis, E., & Feldman, S. (2023, April). Faculty feelings matter: Environmental experiences of queer faculty of color [paper presentation]. Annual Meeting of the AERA (American Educational Research Association), Chicago, IL.
Mayhew, M. J., & Simonoff, J. S. (2015). Non-white, no more: Effect coding as an alternative to dummy coding with implications for higher education researchers. Journal of College Student Development, 56(2), 170–175.