⚖️ Equal Opportunity (Fairness)

Equal Opportunity (Fairness)#

Requires equal true-positive rates across groups — a relaxation of equalized odds.

Important

✨ AI-generated content. This page was written with the assistance of an AI language model and is provided as a learning aid. Despite careful review, it may still contain mistakes, omissions, or out-of-date information. Whether you are new to the topic, a team lead, or a senior practitioner, treat it as a starting point rather than an authoritative reference: read it critically and independently verify anything you act on (code, commands, figures, and factual claims) against official documentation and primary sources before relying on it.

What it is#

Equal opportunity is the relaxation of equalized odds that equalises only the true-positive rate across groups:

\[P(\hat{Y}=1 \mid Y=1, A=a) = P(\hat{Y}=1 \mid Y=1, A=b) \quad \forall\, a, b.\]

Among the people who should receive a positive outcome, every group has the same chance of being correctly recognised — so it targets false negatives not falling disproportionately on a disadvantaged group.

Where it sits among fairness criteria#

It is a one-sided separation criterion: equalized odds asks for equal TPR and equal FPR; equal opportunity keeps only the TPR condition. It sits between the label-blind demographic parity and the full equalized odds.

Example#

Among genuinely qualified applicants, a loan model approves 80% of men but only 60% of women. Qualified women are recognised less often — equal opportunity is violated, even if overall approval rates happen to match.

The catch#

Because it uses ground-truth labels it is more practical than demographic parity, but it ignores false positives, and when base rates differ across groups achieving it can still reduce accuracy and may clash with predictive parity.

Limitations#

Says nothing about false-positive fairness (a group could be over-approved).
Base-rate differences can force an accuracy trade-off.

In code#

from fairlearn.metrics import true_positive_rate, MetricFrame

tpr_by_group = MetricFrame(metrics=true_positive_rate,
                           y_true=y_true, y_pred=y_pred,
                           sensitive_features=A).by_group

Theme: Fairness & Calibration · All terminology

Hint

Mind map — connected ideas

Equalized Odds (Fairness) · Demographic Parity (Statistical Parity) · Predictive Parity (Calibration)

Hint

More in Fairness & Calibration

Demographic Parity (Statistical Parity) · Equalized Odds (Fairness) · Fairness Guardrails · Fairness parity · Four-Fifths (80%) Rule · Predictive Parity (Calibration) · Selection Rate