Research map / Learning theory and algorithms / Fairness
Does mitigating ML's impact disparity require treatment disparity?
Zachary Lipton, Julian McAuley, Alexandra Chouldechova
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use protected-group-specific thresholds instead of a group-blind DLP | Accuracy and impact-parity decision rules · CV and p-% constrained classification | accuracy | Improved |
| Use protected-group-specific thresholds instead of the DLP | Income classifier · Income | p-% rule at constant accuracy: 29% -> 52.9% | Improved |
| Use protected-group-specific thresholds instead of the DLP | Marketing classifier · Marketing | p-% rule at constant accuracy: 102% -> 100.3% | No clear effect |
| Use protected-group-specific thresholds instead of the DLP | Credit classifier · Credit | p-% rule at constant accuracy: 85% -> 100.0% | Improved |
| Use protected-group-specific thresholds instead of the DLP | Employee classifier · Employee | p-% rule at constant accuracy: 69% -> 100.4% | Improved |
| Use protected-group-specific thresholds instead of the DLP | Customer classifier · Customer | p-% rule at constant accuracy: 84% -> 100.2% | Improved |
| Apply the DLP fairness constraint rather than unconstrained classification | Fair constrained classifier · Credit | accuracy: 0.82 -> 0.74 | Hurt |
| Apply the DLP fairness constraint rather than unconstrained classification | Fair constrained classifier · Credit | p-% rule: 88% -> 85% | Hurt |
| Apply the DLP fairness constraint rather than unconstrained classification | Fair constrained classifier · Income, Marketing, Employee, and Customer | p-% rule: Income: 31% -> 29%; Marketing: 82% -> 102%; Employee: 65% -> 69%; Customer: 49% -> 84% | Mixed |
| Use a DLP when protected membership is only partially encoded in other features | DLP classifier · Synthetic hiring example | within-group ranking fairness | Hurt |
| Use a DLP when protected membership is only partially encoded in other features | DLP classifier · CS graduate admissions with synthetic discrimination | do-no-harm for women: 5 of 43 | Hurt |
| Use protected membership directly to set group-specific thresholds | Decision rule under CV or p-% constraints · CV and p-% constrained classification | accuracy | Improved |
Closest papers by meaning
- Understanding challenges to the interpretation of disaggregated evaluations of algorithmic fairness (NeurIPS 2025 · similarity 0.86)
- Reconciling Predictive and Statistical Parity: A Causal Approach (AAAI 2024 · similarity 0.85)
- The Disparate Benefits of Deep Ensembles (ICML 2025 · similarity 0.84)
- Feature Importance Disparities for Data Bias Investigations (ICML 2024 · similarity 0.84)
- The Fragility of Fairness: Causal Sensitivity Analysis for Fair Machine Learning (NeurIPS 2024 · similarity 0.83)
- Fairness-Accuracy Trade-Offs: A Causal Perspective (AAAI 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
