Implementing Anti-discrimination Policies in Statistical Profiling Models
Abstract
How should statistical models used for assigning prices or eligibility be implemented when there is concern about discrimination? In many settings, factors such as race, gender, and age are prohibited. However, the use of variables that correlate with these omitted characteristics (e.g., zip codes, credit scores) is often contentious. We provide a framework to address these issues and propose a method that can eliminate proxy effects while maintaining predictive accuracy relative to an approach that restricts the use of contentious variables outright. We illustrate the value of our proposed method using data from the Worker Profiling and Reemployment Services system. (JEL C53, J15, J65, J71)
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
MC is consistent with methods in Pope and Sydnor and Lindholm et al. and satisfies CPV.
“MC is consistent with the methods provided in Pope and Sydnor (2011) and Lindholm et al. (2022a) and will satisfy CPV.”
From Antidiscrimination Insurance Pricing: Regulations, Fairness Criteria, and Models (2024) · cites this paper as Pope and Sydnor 2011
Pope and Sydnor (2011) and Aseervatham et al. (2016) do not provide a probabilistic justification for the prices used nor address potential bias at portfolio level.
“However, these authors do not provide a probabilistic justification for the prices used nor do they address the critical issue of a potential bias at portfolio level (and associated corrections).”
From Discrimination-Free Insurance Pricing (2022) · cites this paper as Pope and Sydnor (2011)
Cited by
Sorted by citations
Barocas, S., & Selbst, A. D. (2014) · doi:10.2139/ssrn.2477899 · Cited by 3,048 · see citing works
Predictably Unequal? The Effects of Machine Learning on Credit Markets
FUSTER, A., GOLDSMITH‐PINKHAM, P., RAMADORAI, T., & WALTHER, A. (2022) · The Journal of Finance · doi:10.1111/jofi.13090 · Cited by 548 · see citing works
Bias in Bios: A Case Study of Semantic Representation Bias in a High-Stakes Setting
De-Arteaga, M., Romanov, A., Wallach, H., Chayes, J., Borgs, C., Chouldechova, A., Geyik, S., Kenthapadi, K., & Kalai, A. T. (2019) · Proceedings of the Conference on Fairness, Accountability, and Transparency · doi:10.1145/3287560.3287572 · Cited by 337 · see citing works
Layers of bias: A unified approach for understanding problems with risk assessment
Eckhouse, L., Lum, K., Conti-Cook, C., & Ciccolini, J. (2019) · Criminal Justice and Behavior · doi:10.1177/0093854818811379 · Cited by 156 · see citing works
Žliobaitė, I., & Custers, B. (2016) · Artificial Intelligence and Law · doi:10.1007/s10506-016-9182-5 · Cited by 125 · see citing works
Discrimination-Free Insurance Pricing
Lindholm, M., Richman, R., Tsanakas, A., & Wüthrich, M. V. (2022) · ASTIN Bulletin · doi:10.1017/asb.2021.23 · Cited by 60 · see citing works
“The issue of indirect discrimination occurring when using the above mentioned “unawareness” approach, see also Definition 2 below, has been discussed in Pope and Sydnor [18] and Kusner et al.”
BackgroundMethodInfluentialThe Discriminating (Pricing) Actuary
Frees, E. W. (., & Huang, F. (2023) · North American Actuarial Journal · doi:10.1080/10920277.2021.1951296 · Cited by 56 · see citing works
Antidiscrimination Insurance Pricing: Regulations, Fairness Criteria, and Models
Xin, X., & Huang, F. (2024) · North American Actuarial Journal · doi:10.1080/10920277.2023.2190528 · Cited by 43 · see citing works
“…where P (cid:4) ð x P Þ is defined on the same range as P ð x P Þ : Driven by concerns over the proxy effects of X NP on X P , CPV (or the procedure based on CPV) aims to decouple the protected attribute X P from nonprotected attributes X NP ; see Pope and Sydnor (2011) and Lindholm et al. (2022a).”
BackgroundMethodInfluentialThe Fairness of Credit Scoring Models
Hurlin, C., Perignon, C., & Saurin, S. (2021) · SSRN Electronic Journal · doi:10.2139/ssrn.3785882 · Cited by 26 · see citing works
How Do Unisex Rating Regulations Affect Gender Differences in Insurance Premiums?
Aseervatham, V., Lex, C., & Spindler, M. (2016) · The Geneva Papers on Risk and Insurance - Issues and Practice · doi:10.1057/gpp.2015.22 · Cited by 27 · see citing works
Dell'Acqua, F., Kogut, B., & Perkowski, P. (2020) · SSRN Electronic Journal · doi:10.2139/ssrn.3746564 · Cited by 5 · see citing works
Fairness Testing for Algorithmic Pricing
Huang, F., & Hooker, G. (2026) · arXiv · doi:10.48550/arxiv.2605.11614
“Pope and Sydnor (2011) develop a framework for identifying and eliminating proxy discrimination in statistical profiling models, illustrated using employment profiling data.”
- Authors
- Pope, D. G., & Sydnor, J. R.
- Affiliations
- University of Chicago (US)
- University of Wisconsin–Madison (US)
- Venue
- American Economic Journal: Economic Policy
- Publisher
- American Economic Association
- Published
- 2011-08-01
- Volume
- 3
- Issue
- 3
- Pages
- 206-231
- ISSN
- 1945-7731 (ISSN-L)
- 1945-774X
- OpenAlex
- W2085953209
- Cited by
- 87
- References
- 0
- Citations by year
- 2026: 1 · 2025: 5 · 2024: 15 · 2023: 11 · 2022: 8 · 2021: 11 · 2020: 11 · 2019: 6 · 2018: 3 · 2017: 4 · 2016: 2 · 2015: 2 · 2014: 2 · 2013: 2 · 2012: 2
- FWCI
- 6.62
- Type
- journal article
- Language
- en
- Open access
- closed
- Locations
- American Economic Journal Economic Policy (publishedVersion)
- RePEc: Research Papers in Economics (submittedVersion)
- Topics
- Social Sciences › Safety Research › Ethics and Social Impacts of AI
- Social Sciences › Law › Discrimination and Equality Law
- Social Sciences › Sociology and Political Science › Names, Identity, and Discrimination Research
- Keywords
- anti-discrimination policy, statistical profiling, prediction accuracy, algorithmic discrimination, reemployment services, protected characteristics, credit scores, zip codes, eligibility determination
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
