{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/does-mitigating-mls-impact-disparity-require","title":"Does mitigating ML's impact disparity require treatment disparity?","arxiv_id":"1711.07076","date":"2017-11-19","proceeding":"NeurIPS 2018 12","authors":["Zachary C. Lipton","Alexandra Chouldechova","Julian McAuley"],"abstract":"Following related work in law and policy, two notions of disparity have come\nto shape the study of fairness in algorithmic decision-making. Algorithms\nexhibit treatment disparity if they formally treat members of protected\nsubgroups differently; algorithms exhibit impact disparity when outcomes differ\nacross subgroups, even if the correlation arises unintentionally. Naturally, we\ncan achieve impact parity through purposeful treatment disparity. In one thread\nof technical work, papers aim to reconcile the two forms of parity proposing\ndisparate learning processes (DLPs). Here, the learning algorithm can see group\nmembership during training but produce a classifier that is group-blind at test\ntime. In this paper, we show theoretically that: (i) When other features\ncorrelate to group membership, DLPs will (indirectly) implement treatment\ndisparity, undermining the policy desiderata they are designed to address; (ii)\nWhen group membership is partly revealed by other features, DLPs induce\nwithin-class discrimination; and (iii) In general, DLPs provide a suboptimal\ntrade-off between accuracy and impact parity. Based on our technical analysis,\nwe argue that transparent treatment disparity is preferable to occluded methods\nfor achieving impact parity. Experimental results on several real-world\ndatasets highlight the practical consequences of applying DLPs vs. per-group\nthresholds.","url_abs":"http://arxiv.org/abs/1711.07076v3","url_pdf":"http://arxiv.org/pdf/1711.07076v3.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"does-mitigating-mls-impact-disparity-require","repo_url":"https://github.com/autogluon/autogluon-fair","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"decision-making","task_name":"Decision Making"},{"task_slug":"fairness","task_name":"Fairness"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1711.07076","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}