Fair prediction with disparate impact: A study of bias in recidivism prediction instruments

Alexandra Chouldechova relates positive predictive value, false-positive rate, false-negative rate and outcome prevalence (2017 manuscript, Eq. 2.6). She pairs the formal analysis with COMPAS data and hypothetical penalty policies.^[1]

Read Fairness Tradeoffs for definitions and a worked example, then the investigation entry for the wider reading sequence. Inherent Trade-Offs in the Fair Determination of Risk Scores considers a different combination of score-level requirements.

Historical context

The short manuscript was submitted October 24, 2016; the separate long manuscript appeared February 28, 2017. Both originals backed up; selected definitions, relation and discussion inspected, with rendered long p.7. Versions are not assumed identical. Full proofs, raw records, code and causal sentencing effects were not audited.^[1]