Skip to content
Last Dictionary
Plane
Society, Ethics & EconomyMachine Learning

equalized odds

nounAlso: separation (fairness criterion), error-rate balance, error rate parity

In plain wordsA 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.

  1. A group fairnessgroup fairness, n.: Measuring 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. criterion requiring equal true positive rates and equal false positive rates 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. A; equivalently, the prediction is conditionally independent of A given the true outcome Y. Unlike 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., it allows selection rates to differ when base rates differ, and a perfect classifier satisfies it. Any risk score can be post-processed to meet it by choosing group-specific, sometimes randomized, decision thresholds from each group's ROC curve, usually at some cost in accuracy. It is the condition ProPublica found violated in COMPASCOMPAS, prop. n.: Commercial software used in parts of the US justice system to rate how likely someone is to commit another crime; a 2016 news investigation asking whether it was unfair to Black defendants made it famous.; when base rates differ, an imperfect classifier cannot satisfy it together with 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..

    Equalized odds requires the recidivism tool's false positive rate to be the same for Black and white defendants.

P(Y^=1∣A=a,Y=y)=P(Y^=1∣A=b,Y=y),y∈{0,1}P(\hat{Y}=1 \mid A=a, Y=y) = P(\hat{Y}=1 \mid A=b, Y=y), \quad y \in \{0, 1\}

Origin

Introduced, with 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., by Moritz Hardt, Eric Price and Nathan Srebro in 'Equality of Opportunity in Supervised Learning' (arXiv October 2016; NeurIPS 2016). Closely related error-rate balance conditions appeared the same year in Alexandra Chouldechova's analysis of recidivism prediction.

Related

See also
  • 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.
  • 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.
  • 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.
  • group fairnessgroup fairness, n.: Measuring 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.
  • 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.
  • COMPASCOMPAS, prop. n.: Commercial software used in parts of the US justice system to rate how likely someone is to commit another crime; a 2016 news investigation asking whether it was unfair to Black defendants made it famous.
Contrast with
  • 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.

Sources

  1. Hardt, M., Price, E. & Srebro, N. (2016). Equality of Opportunity in Supervised Learning. Advances in Neural Information Processing Systems 29. arxiv.org/abs/1610.02413

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., COMPASCOMPAS, prop. n.: Commercial software used in parts of the US justice system to rate how likely someone is to commit another crime; a 2016 news investigation asking whether it was unfair to Black defendants made it famous., 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., 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., group fairnessgroup fairness, n.: Measuring 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., 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.