Decide whether to use demographic predictors case-by-case based on context: risk of discriminatory decision-making, who holds final choice, and stakes¶

Strategy · All strategies
Evidence · no claims cited

Description¶

The article offers a contextual decision framework rather than a blanket rule. Using demographic data as predictors is riskier when it can drive discriminatory decisions, such as admissions selection or violence-risk prediction with punitive interventions. It "may be less harmful if the person being impacted ultimately still has the final choice about what happens, or if the stakes involved in the decision are low", as in activity recommendations or college application suggestions where the student still chooses. Decision-makers should weigh the intervention type, who controls the outcome, and the cost of a poor prediction.

Design Implications¶

Context¶

Requirements¶

• An analysis of how the model's predictions will be used and by whom, and what interventions follow

Constraints¶

• Even low-stakes cases carry some risk, e.g. activity ranking by declared sex may be undesirable for students with gender-atypical interests

Target Learners¶

• Students in systems using predictive analytics, from K-12 through higher education and admissions

Target Learning Goals¶

• Fair deployment of predictive analytics in educational decision-making

Related Strategies¶

• Demographics For Validation Not Prediction
• Use early cheating-risk predictions only for low-stakes support with human review, never for disciplinary decisions

Examples¶

-

Key Sources¶

• Baker, R. S., Esbenshade, L., Vitale, J., & Karumbaiah, S. (2023). Using Demographic Data as Predictor Variables: a Questionable Choice. Journal of Educational Data Mining, 15(2). https://jedm.educationaldatamining.org/index.php/JEDM/article/view/619. doi:10.35542/osf.io/y4wvj