Interpret a group disparity before changing a classifier.

Instruction: In an audited period, group A has 20 false positives among 1,000 actual negatives; group B has 20 among 500 actual negatives. Compare false-positive rates. Explain what else you need before deciding on a mitigation or claiming the system is fair.

Context: Assesses group-rate denominators, uncertainty, label and workflow validity, and limits of a single fairness metric.

Updated

Official answer available

Read the opening below, then unlock the full answer and practical guidance.

The false-positive rate is FP / (FP + TN). A has 20 / 1,000 = 2%; B has 20 / 500 = 4%. Equal false-positive counts therefore do not mean equal false-positive rates. The observed rate is twice as high for B, with a difference of 2 percentage points.

This identifies a...

Upgrade to view official answer

Related Questions