Research map / Learning theory and algorithms / Fairness
Cookie Consent Has Disparate Impact on Estimation Accuracy
Erik Miehling, Rahul Nair, Elizabeth M. Daly, Karthikeyan Natesan Ramamurthy, Robert Nelson Redmond
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 |
|---|---|---|---|
| withhold cookie consent instead of consenting, with homogeneous consent rates (q=0.5 both cohorts) | confidence-weighted online matrix factorization recommender · Simulated cohort estimation (Fig. 2a) | cohort log loss | Hurt |
| withhold cookie consent in the low-consent cohort under heterogeneous rates (0.25 vs 0.75) | recommender agent, low-consent cohort theta · Simulated cohort estimation (Fig. 2b) | cohort log loss | Improved |
| withhold cookie consent in the high-consent cohort under heterogeneous rates (0.25 vs 0.75) | recommender agent, high-consent cohort theta' · Simulated cohort estimation (Fig. 2b) | cohort log loss | Hurt |
| increase the consent-rate gap between cohorts | non-consent vs consent error difference · Simulated cohort estimation (Figs. 4-5) | cohort log loss difference | Improved |
| small consent gap (0.2 vs 0.4) with very informative prior | recommender agent · Simulated cohort estimation (Fig. 3a) | cohort log loss | No clear effect |
| raise absolute consent rates while keeping the same gap (0.7 vs 0.9) (ablation) | recommender agent · Simulated cohort estimation (Fig. 3c) | cohort log loss for low-consent cohort non-consent group | Improved |
| vary batch size Tb in {1, 2, 10} (ablation) | recommender retraining schedule · Appendix D sensitivity | cohort error | No clear effect |
| larger batch size Tb > 1 (ablation) | recommender retraining schedule · Appendix D sensitivity | number of interactions facing disparate errors | Hurt |
| add offline response set of size 1 or 5 per user (ablation) | recommender prior · Appendix D sensitivity | cohort error evolution: |L0| = 0 -> 1 -> 5 | No clear effect |
| make topic affinity means of cohorts more similar ((0.3,0.7) to (0.4,0.6)) (ablation) | recommender agent · Appendix D sensitivity | responses needed to reach same cohort error: (0.3, 0.7) -> (0.4, 0.6) | Hurt |
| identical affinity statistics across cohorts (0.5, 0.5) (ablation) | recommender agent · Appendix D sensitivity | cohort uncertainty resolution: (0.3, 0.7) -> (0.5, 0.5) | Hurt |
| fully informative prior (0.5/0 diagonal) versus partially informative (ablation) | recommender interim beliefs · Appendix D sensitivity | cohort error of consent group | Improved |
Closest papers by meaning
- FairJob: A Real-World Dataset for Fairness in Online Systems (NeurIPS 2024 · similarity 0.84)
- On Online Experimentation without Device Identifiers (ICML 2024 · similarity 0.83)
- Leveraging Opposite Gender Interaction Ratio as a Path towards Fairness in Online Dating Recommendations Based on User Sexual Orientation (AAAI 2024 · similarity 0.83)
- More RLHF, More Trust? On The Impact of Preference Alignment On Trustworthiness (ICLR 2025 · similarity 0.82)
- Distributional Preference Learning: Understanding and Accounting for Hidden Context in RLHF (ICLR 2024 · similarity 0.82)
- PrivacyLens: Evaluating Privacy Norm Awareness of Language Models in Action (NeurIPS 2024 · similarity 0.82)
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.
