Demographic Parity (Statistical Parity)
What you'll learn: You'll understand demographic parity as a fairness criterion that requires equal positive prediction rates across all demographic groups, independent of actual outcomes.
What Is Demographic Parity?
Demographic parity (also called statistical parity) is one of the most straightforward group fairness criteria in machine learning. It demands that a model's positive predictions are distributed equally across all protected groups, regardless of the true labels or base rates in those groups.
Formally, a classifier satisfies demographic parity if:
P(Ŷ = 1 | A = a) = P(Ŷ = 1 | A = b) for all groups a and b
Where Ŷ is the predicted outcome and A is the protected attribute (like race or gender).
When Is This Appropriate?
Demographic parity is appropriate when:
- Equal representation matters most: You want equal access or opportunity regardless of historical patterns (e.g., college admissions, job screening)
- Historical bias exists in labels: The ground truth itself may be discriminatory (e.g., past hiring decisions were biased)
- Perception of fairness is critical: Stakeholders value visible equality in outcomes
When It May NOT Be Appropriate
Demographic parity can be problematic when groups have legitimately different base rates. For example, if a disease genuinely affects one demographic group at higher rates, enforcing equal positive prediction rates would harm predictive accuracy and potentially deny care to those who need it most.
It also ignores individual qualifications—two people with identical features but different group membership must receive predictions that maintain group-level balance, which may feel unfair at the individual level.
Key Takeaway: Demographic parity requires equal positive prediction rates across groups and is most appropriate when equal representation outweighs predictive accuracy or when historical labels themselves encode discrimination.