Equal Opportunity and Equalized Odds
What you'll learn: How to ensure AI systems perform equally well across protected groups by matching true positive rates and error rates.
The Core Idea
While demographic parity (from lesson 1566) requires equal positive prediction rates across groups, equal opportunity and equalized odds focus on equalizing performance metrics — specifically, how accurately the model identifies true positives and handles errors across protected groups.
Equal Opportunity means the model's true positive rate (recall) should be the same for all protected groups. If your hiring AI correctly identifies 80% of qualified candidates in one demographic group, it should achieve ~80% for all groups.
Equalized Odds is stricter: it requires both true positive rates AND false positive rates to match across groups. The model should be equally good at identifying positives and equally careful about false alarms for everyone.
Why This Matters
Imagine a loan approval system:
- Equal Opportunity: Ensures qualified applicants from all backgrounds have equal chances of approval
- Equalized Odds: Also ensures unqualified applicants are rejected at similar rates across groups
This prevents scenarios where a model is "generous" (high approval rate) to one group but makes more mistakes with them, while being "strict" with another group.
The Trade-off
These fairness definitions often conflict with demographic parity. You might achieve equal performance metrics but have different overall approval rates — or vice versa. The choice depends on your application's harm model and which errors matter most.
Key Takeaway: Equal opportunity equalizes true positive rates across groups, while equalized odds equalizes both true positive and false positive rates — ensuring consistent predictive performance regardless of protected attributes.