Demographic Parity criterion describes?

• A. Predictive Rate Parity
• B. Demographic Parity
• C. Exact Match Metrics
• D. BLEU and ROUGE

Answer: (B)

Demographic parity requires that the model’s positive outcome rate is the same across groups defined by a protected attribute (such as race or gender). In other words, the probability of a positive prediction should be independent of the group, so P(Ŷ=1 | group) is equal across groups. If the overall approval rate is, say, 30%, each group should have approximately a 30% approval rate, regardless of their base rates.

This focuses on equality of outcomes across groups, not on how accurate predictions are for individuals. It’s a specific fairness criterion about independence between the protected attribute and the prediction.

Other options describe different notions of fairness. Predictive Rate Parity (equal positive predictive value across groups) asks that the accuracy of positive predictions be the same across groups, which can conflict with demographic parity when base rates differ. The remaining choices aren’t standard fairness criteria relevant to this concept (they pertain to other tasks like text evaluation or different metrics).