Research map / Learning theory and algorithms / Fairness
The Importance of Modeling Data Missingness in Algorithmic Fairness: A Causal Perspective
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Train fairness algorithms on data censored according to past decisions rather than representative data | Fairness algorithms evaluated on Adult and COMPAS · Adult and COMPAS | fairness metric disparity: 0.0725 (-0.00724); 0.0377 (0.00295); -0.1137 (-0.0390); -0.2410 (-0.0705); 0.5566 (0.7785); -0.1486 (0.0188); -0.1327 (0 | Hurt |
| Use DF² instead of the centralized oracle EAGGL | Two-stage decision making · Adult, German, and COMPAS | utility | No clear effect |
| Require local fairness at every stage rather than only global fairness | DF² compared with EAGGL · Adult, German, and COMPAS | utility | No clear effect |
| Use causal-graph recoverability to select estimable conditional distributions instead of requiring the non-recoverable joint feature distribution | Multi-stage fair decision algorithm · Two-stage decision making | precision | No clear effect |
| Estimate group error rates naively from systematically incomplete data | Equalized-odds classifiers · General population under past-decision missingness | group error-rate consistency | Hurt |
| Estimate allocation rates naively from systematically incomplete data | Demographic-parity classifiers · General population under past-decision missingness | allocation-rate consistency | Hurt |
| Use fully automated past decisions based on observed features | Estimating outcome distributions under missingness · Fully automated decision-making | recoverability of P(Y|X) and P(Y|X,Z) | Helped |
| Allow the sensitive attribute to directly affect the outcome | Fully automated decision-making · Fully automated decision-making | consistency of naive P(Y|X) estimation | Hurt |
| Include unobserved human decision features that affect the outcome directly | Estimating outcome distributions under human decisions · Human decision-making | recoverability of P(Y|X,Z) | Hurt |
| Use a machine-aided decision process with unobserved human influences | Estimating outcome distributions under machine-aided decisions · Machine-aided decision-making | recoverability of P(Y|X,Z) | Mixed |
Closest papers by meaning
- Reconciling Predictive and Statistical Parity: A Causal Approach (AAAI 2024 · similarity 0.88)
- FairPFN: A Tabular Foundation Model for Causal Fairness (ICML 2025 · similarity 0.88)
- Interventional Fairness on Partially Known Causal Graphs: A Constrained Optimization Approach (ICLR 2024 · similarity 0.88)
- A Local Method for Satisfying Interventional Fairness with Partially Known Causal Graphs (NeurIPS 2024 · similarity 0.88)
- The Fragility of Fairness: Causal Sensitivity Analysis for Fair Machine Learning (NeurIPS 2024 · similarity 0.87)
- Fairness without Harm: An Influence-Guided Active Sampling Approach (NeurIPS 2024 · similarity 0.87)
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.