Intersectionality in Fairness
You'll learn why fairness analysis must account for overlapping identity dimensions, not just single attributes in isolation.
When we measure fairness, it's tempting to check one protected class at a time: Does our model treat men and women equally? Does it treat different racial groups fairly? But real people don't experience the world through just one identity dimension—they live at the intersection of multiple identities.
Intersectionality recognizes that a Black woman's experience isn't just "being Black" plus "being a woman"—it's a unique combined experience. A hiring algorithm might show no bias when you examine gender alone and no bias when you examine race alone, yet still systematically disadvantage Black women specifically.
Why Single-Attribute Analysis Fails
Imagine a loan approval model. Your fairness audit shows:
- Women and men: similar approval rates ✓
- Black and white applicants: similar approval rates ✓
You conclude the model is fair. But when you check Black women specifically, you discover their approval rate is significantly lower than Black men, white women, and white men. Simpson's paradox can hide intersectional bias when you only examine attributes separately.
Practical Implications
- Subgroup analysis: Always disaggregate your fairness metrics across combinations of protected attributes (gender × race, age × disability status, etc.)
- Sample size challenges: Intersectional groups may be small, making statistical analysis harder
- Exponential complexity: With multiple attributes, the number of subgroups grows quickly
- Contextual understanding: Some intersections carry unique historical disadvantages that single-attribute analysis misses entirely
Key Takeaway: Fairness metrics calculated for single attributes can mask severe discrimination against people who belong to multiple protected groups simultaneously; always analyze intersectional subgroups to uncover hidden disparate impact.