group fairness
In plain wordsMeasuring fairness by comparing how a system treats whole groups, such as women and men, on average: their approval rates, their error rates or what their scores mean.
The family of fairness criteria requiring some statistic of a model's decisions to be equal, or nearly so, across groups defined by a protected attributeprotected attribute, n.: A personal trait, such as race, sex, religion, age or disability, that the law says must not be used to treat people unfairly.. Most fall into three classes defined by conditional independence among the prediction R, the attribute A and the true outcome Y: independence (R independent of A, as in demographic paritydemographic parity, n.: A fairness test that asks whether a system says yes (approves the loan, shortlists the applicant) at the same rate for every group, whatever other differences exist between them.), separation (R independent of A given Y, as in equalized oddsequalized odds, n.: A fairness test that asks whether a system makes mistakes at the same rates for every group: wrongly flagging innocent people, and wrongly missing real cases, equally often. and equal opportunityequal opportunity, n.: A fairness test that asks whether people who truly deserve a yes (who would repay the loan or succeed in the job) get one at the same rate in every group.) and sufficiency (Y independent of A given R, as in predictive paritypredictive parity, n.: A fairness test that asks whether a system's verdict means the same thing for every group: of the people it flags as high risk, the same share really are, whatever their group. and calibration). Group criteria are easy to measure and audit and map onto legal concepts such as disparate impactdisparate impact, n.: When a rule or system that treats everyone alike on paper still lands much harder on one group, such as women or older workers; the law can forbid this even if no one intended it., but they protect averages rather than individuals, can be met while particular subgroups fare badly, and conflict with one another when base rates differ (see fairness impossibility theoremfairness impossibility theorem, n.: A mathematical result showing that when groups differ in how often an outcome really occurs, no imperfect prediction system can be fair by every common measure at once.). Subgroup and multicalibration methods extend them to many overlapping, intersectional groups.
Group fairness looked fine for women and for Black applicants separately, but not for Black women.
Origin
Developed in discrimination-aware data mining from 2008 and set against individual fairness after Cynthia Dwork and colleagues' 'Fairness Through Awareness' (2012). The independence–separation–sufficiency taxonomy was popularized by Solon Barocas, Moritz Hardt and Arvind Narayanan's textbook Fairness and Machine Learning, circulated in drafts from 2017.
Related
- individual fairnessindividual fairness, n.: The idea that a fair system should treat similar people similarly: two applicants who are alike in every way that matters should get similar outcomes.
- demographic paritydemographic parity, n.: A fairness test that asks whether a system says yes (approves the loan, shortlists the applicant) at the same rate for every group, whatever other differences exist between them.
- equalized oddsequalized odds, n.: A fairness test that asks whether a system makes mistakes at the same rates for every group: wrongly flagging innocent people, and wrongly missing real cases, equally often.
- predictive paritypredictive parity, n.: A fairness test that asks whether a system's verdict means the same thing for every group: of the people it flags as high risk, the same share really are, whatever their group.
- fairness impossibility theoremfairness impossibility theorem, n.: A mathematical result showing that when groups differ in how often an outcome really occurs, no imperfect prediction system can be fair by every common measure at once.
- algorithmic fairnessalgorithmic fairness, n.: The effort to make automated decisions fair: to say in numbers what fair treatment means, measure whether a system achieves it, and correct the system if it does not.
- individual fairnessindividual fairness, n.: The idea that a fair system should treat similar people similarly: two applicants who are alike in every way that matters should get similar outcomes.
Sources
- Barocas, S., Hardt, M. & Narayanan, A. (2023). Fairness and Machine Learning: Limitations and Opportunities. MIT Press.
- Kearns, M., Neel, S., Roth, A. & Wu, Z. S. (2018). Preventing Fairness Gerrymandering: Auditing and Learning for Subgroup Fairness. Proc. ICML 2018.
Mentioned in
algorithmic fairnessalgorithmic fairness, n.: The effort to make automated decisions fair: to say in numbers what fair treatment means, measure whether a system achieves it, and correct the system if it does not., fairness impossibility theoremfairness impossibility theorem, n.: A mathematical result showing that when groups differ in how often an outcome really occurs, no imperfect prediction system can be fair by every common measure at once., Gender ShadesGender Shades, prop. n.: A 2018 study showing that commercial face-analysis software guessed the gender of darker-skinned women far less accurately than that of lighter-skinned men., protected attributeprotected attribute, n.: A personal trait, such as race, sex, religion, age or disability, that the law says must not be used to treat people unfairly., proxy discriminationproxy discrimination, n.: Discrimination that happens when a system cannot see a trait like race directly but relies on something closely linked to it, such as a postcode, and so treats groups unequally anyway., demographic paritydemographic parity, n.: A fairness test that asks whether a system says yes (approves the loan, shortlists the applicant) at the same rate for every group, whatever other differences exist between them., equal opportunityequal opportunity, n.: A fairness test that asks whether people who truly deserve a yes (who would repay the loan or succeed in the job) get one at the same rate in every group., equalized oddsequalized odds, n.: A fairness test that asks whether a system makes mistakes at the same rates for every group: wrongly flagging innocent people, and wrongly missing real cases, equally often., individual fairnessindividual fairness, n.: The idea that a fair system should treat similar people similarly: two applicants who are alike in every way that matters should get similar outcomes., predictive paritypredictive parity, n.: A fairness test that asks whether a system's verdict means the same thing for every group: of the people it flags as high risk, the same share really are, whatever their group.