Universally applicable guidance for using race as a statistical variable

Establish universally applicable guidelines for determining when and how race should be used as a variable in statistical analyses and data-science models, while addressing the tensions between predictive usefulness, fairness, the reification of harmful social constructs, and the illumination of racial injustice.

Background

The paper presents race as a particularly complex variable in statistics and data science. Using race in a model may provide useful predictive information or help reveal institutional inequities, but it may also introduce harmful bias, reinforce racial hierarchies, or reify socially constructed categories. The author describes classroom discussions that expose students to these competing considerations and explicitly states that the paper does not resolve them or provide universal guidance. This leaves the development of broadly applicable principles for handling race as a statistical variable as an unresolved problem.

References

While we do not resolve these tensions or come to prescribed guidelines for how to universally handle race as a variable, we hope these discussions contribute to students’ formation as thoughtful data scientists who are aware enough to pause, think through, and seek input on the implications of their decisions in any given context.

— Becoming Good Stewards of Information: A framework for integrating ethical, civic, and professional formation throughout the statistics and data science curriculum  (2608.23352 - Fitzgerald, 24 Aug 2026) in Section 4.4, Data Science