Impact of racial categorization on effect estimates: an analysis of HIV stigma

This article was originally published here

Am J Epidemiol. January 5, 2022: kwab289. doi: 10.1093/aje/kwab289. Online ahead of print.


Suboptimal racial categorization potentially introduces bias into epidemiological analysis and interpretation, making it difficult to appropriately measure factors leading to racial health disparities. In an analysis focused on predictors of HIV-related stigma among men who have sex with men living with HIV in San Francisco, we wrestled with the most appropriate ways to categorize people who have stated more than one racial identity, and we sought to explore the implications of different methodological choices in this analysis. We ran three different multivariate linear regression models, each using a different approach to racial categorization: the ‘multiracial’, ‘other’ and ‘hypodescent’ models. We estimate an adjusted risk difference in the mean score for reported frequency of HIV-related stigma on a 4-point scale, adjusting for age, race, gender identity, injection history, housing, mental health issues and viral load. Using a hypodescent model for racial categorization resulted in a shift of the point estimate towards zero for Blacks/African Americans, and improved precision for this group. However, this masked the association of increased stigma and race for multiracial people, compared to their monoracial counterparts. We conclude that methodological decisions related to the racial categorization of participants can significantly affect the results of race-related studies in predictive regression models.

PMID:34999778 | DOI:10.1093/aje/kwab289

Comments are closed.