[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85386-en":3,"doc-seo-85386-105":30,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},85386,1099513958607,"Jiven","https://ap-avatar.wpscdn.com/avatar/100002390cf8733938c?x-image-process=image/resize,m_fixed,w_180,h_180&k=1778829742770036399",8,"Research & Report","Interventions Against Machine-Assisted Statistical Discrimination","Study statistical discrimination caused by verifiable beliefs produced by machine learning rather than by unverifiable human thoughts. With verifiable beliefs, interventions can be designed as belief-contingent mechanisms that go beyond belief-free equal treatment rules. Analyze a belief-contingent intervention called common identity and show it can outperform widely used alternatives in mitigating statistical discrimination. Emphasis falls on settings where training data contain feature, sample, or label biases typical of machine-assisted decision problems.","arXiv :2310 .04585v5 [ econ .TH] 11 Jul 2026  \nINTERVENTIONS AGAINST MACHINEASSISTED STATISTICAL DISCRIMINATION  \nJOHN Y. ZHU 1  \nJuly 14, 2026  \nAbstract  \nI study statistical discrimination driven by verifiable beliefs, such as those generated by machine learning, rather than by humans. When beliefs are verifiable, interventions against statistical discrimination can move beyond simple belieffree designs, like affirmative action and blinding, to more sophisticated beliefcontingent ones. I analyze a belief-contingent intervention, common identity, and show that it can be more effective at combating statistical discrimination than popular alternatives – particularly when the training dataset exhibits the kinds of statistical biases that often plague machine-assisted decision problems.  \nJEL Codes: D86, J71, L51  \nKeywords: statistical discrimination, beliefs, machine learning, AI, intervention, affirmative action, blinding, feature bias, optimal transport, sample bias, label bias.  \n1 University of Kansas, [johnzhuyiran@ku.edu. I thank Li Hao](johnzhuyiran@ku.edu. I thank Li Hao), Hyunseob Kim, Kris Nimark, and Jennifer Ifft for reading through the manuscript and providing detailed feedback. I also thank Xiaosheng Mu, Fedor Sandomirskiy, Nathan Yoder, and seminar and conference audiences at Kansas State University, the NSF/CEME Decentralization Conference on Mechanism Design with AI and Distributed Ledgers, Princeton, the Kansas Workshop in Economic Theory, the North American Summer Meeting of the Econometric Society, the Midwest Economic Theory Conference, the University of Queensland, the Designing for Redistribution Conference, Pittsburgh-Carnegie Mellon, the University of Western Ontario, and the University of British Columbia for helpful comments and discussions. I am grateful to Crystal Lenz for excellent editorial assistance. All remaining errors are my own.  \n1 Introduction  \nIn many settings, a decision maker (DM) must act on an individual without observing a relevant quality of that individual. For example, employers and colleges make offers without observing who has high ability. Banks lend without observing who will default. Judges grant bail without observing who will commit a violent crime, if released. When an individual’s quality is unobserved, the DM must infer said quality from “features” of the individual that are observed, such as test scores, income, or zip code.  \nSuppose the DM believes one group of individuals, say, B, has worse average quality than another group, A. Then, in comparing two individuals with identical features, one from each group, the DM will form a worse belief about the group B individual’s quality. A vicious cycle of statistical discrimination can now emerge: The DM’s worse beliefs about quality in group B cause the DM to apply a tougher decision policy to the group, such as a higher test score cutoff for acceptance. Group B individuals then have less incentive to invest in quality. This results in a worse distribution of quality throughout the group, which ultimately fulfills the DM’s worse beliefs about the group as a whole. Consequently, group B gets stuck in a worse equilibrium with the DM than group A.  \nTo combat statistical discrimination, policymakers often mandate some form of equal treatment for both groups. For example, affirmative action requires that the DM accept from both groups at the same rate, while group-blinding bans the DM from conditioning acceptance on group membership. While the specific form of equal treatment may differ across interventions, the intent is the same – to give both groups the same incentive to invest in quality. The idea is that equal incentives lead to equally qualified groups. When group B becomes just as qualified as group A, the DM no longer desires to treat group B any worse than group A. The intervention then ceases to be a binding constraint on the DM and can be lifted. In this way, equilibria between the two groups and an intervention-co","cbCaiv6GGRN4NcoB","https://ap.wps.com/l/cbCaiv6GGRN4NcoB","pdf",443648,3,1,47,"English","en",105,"# Abstract\n# Introduction\n## Statistical discrimination and vicious cycles\n## Equal treatment interventions: affirmative action and blinding\n## Why belief-contingent interventions become feasible with machine learning\n# Belief-contingent intervention: common identity","[{\"question\":\"What drives statistical discrimination in this paper?\",\"answer\":\"Statistical discrimination is driven by beliefs about unobserved individual quality, especially beliefs generated by machine learning outputs.\"},{\"question\":\"Why do equal treatment approaches like affirmative action and blinding often fail?\",\"answer\":\"They do not reliably guarantee equal incentives, because equal acceptance rates or equal score cutoffs can still lead to incentives that differ across groups when beliefs or additional features matter.\"},{\"question\":\"What is the key idea behind belief-contingent interventions such as common identity?\",\"answer\":\"The intervention conditions on the decision maker’s beliefs about quality, which becomes possible when beliefs are verifiable through machine learning systems.\"}]",1784203070,118,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"interventions-against-machine-assisted-statistical-discrimination","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":20},"https://docshare.wps.com/document/research-report/",{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/interventions-against-machine-assisted-statistical-discrimination/85386/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-24","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What drives statistical discrimination in this paper?","Question",{"text":75,"@type":76},"Statistical discrimination is driven by beliefs about unobserved individual quality, especially beliefs generated by machine learning outputs.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"Why do equal treatment approaches like affirmative action and blinding often fail?",{"text":80,"@type":76},"They do not reliably guarantee equal incentives, because equal acceptance rates or equal score cutoffs can still lead to incentives that differ across groups when beliefs or additional features matter.",{"name":82,"@type":73,"acceptedAnswer":83},"What is the key idea behind belief-contingent interventions such as common identity?",{"text":84,"@type":76},"The intervention conditions on the decision maker’s beliefs about quality, which becomes possible when beliefs are verifiable through machine learning systems.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]