Which fairness criterion requires identical likelihood of a positive prediction across subpopulations regardless of correctness?

Which fairness criterion requires identical likelihood of a positive prediction across subpopulations regardless of correctness?

• A. Group Fairness
• B. Predictive Rate Parity
• C. Demographic Parity
• D. Individual Fairness

Answer: (C)

This question centers on Demographic Parity. It requires that the model produce positive predictions at the same rate across different subpopulations, regardless of whether those predictions are correct. In practical terms, the likelihood of predicting “positive” should be equal for each group, even if the true outcomes differ between groups. This focuses on the classifier’s output rate by group, not on accuracy or how many true positives there are. Why this is the best fit: Demographic Parity aims for equalized positive prediction rates across groups, which matches the idea of identical likelihood of a positive prediction irrespective of correctness. It does not consider the actual labels or the correctness of predictions, only the rate at which the model says “positive.” For contrast: Predictive Rate Parity (predictive parity) requires equal positive predictive value across groups, which ties to correctness by ensuring that among those predicted positive, the proportion who are truly positive is the same across groups. Individual Fairness focuses on treating similar individuals similarly, not on group-level prediction rates. Group Fairness is a broader umbrella term; Demographic Parity is a specific instantiation under that umbrella.

This question centers on Demographic Parity. It requires that the model produce positive predictions at the same rate across different subpopulations, regardless of whether those predictions are correct. In practical terms, the likelihood of predicting “positive” should be equal for each group, even if the true outcomes differ between groups. This focuses on the classifier’s output rate by group, not on accuracy or how many true positives there are.

Why this is the best fit: Demographic Parity aims for equalized positive prediction rates across groups, which matches the idea of identical likelihood of a positive prediction irrespective of correctness. It does not consider the actual labels or the correctness of predictions, only the rate at which the model says “positive.”

For contrast: Predictive Rate Parity (predictive parity) requires equal positive predictive value across groups, which ties to correctness by ensuring that among those predicted positive, the proportion who are truly positive is the same across groups. Individual Fairness focuses on treating similar individuals similarly, not on group-level prediction rates. Group Fairness is a broader umbrella term; Demographic Parity is a specific instantiation under that umbrella.