Group-disparity metrics for auditing classification systems
Source:R/fairness_metrics.R
fairness_metrics.RdR port of morie.fairness.metrics. Each callable is an
audit measure: given decisions a system made (and, where
available, the realised ground truth) plus a protected attribute,
it quantifies whether outcomes differ across groups. None of these
functions make predictions; they only measure disparity in
predictions that already exist.
Details
Functions
morie_fairness_disparate_impact: the four-fifths rule.morie_fairness_demographic_parity: favourable-rate gap.morie_fairness_equalized_odds: TPR/FPR gaps (needs ground truth).morie_fairness_average_odds_difference: mean TPR+FPR gap.morie_fairness_gini: concentration of a score distribution.morie_fairness_bias_amplification: composite of parity gap and inequality.
Prior art reimplemented independently (no code copied): the COMPAS fairness audit in pbiecek's XAI Stories and IBM's AI Fairness 360 definitions; the predictive-policing disparity framing of the SciencesPo Predictive-policing-Chicago project (Lacherade, Szabo, Krikava & Aeby, 2021) and Barman & Barman, arXiv:2603.18987.