Equal selection rates
Demographic parity requires that the rate of positive predictions be the same across protected groups. If a hiring model selects twenty percent of one group, it should select roughly twenty percent of every group. It does not look at the true outcome, only at who is chosen.
Stated simply
The probability of a positive prediction does not depend on group membership. This is an independence criterion: prediction is independent of the protected attribute.
Strengths
- Easy to explain and audit from outcomes alone.
- Directly targets unequal access to a beneficial decision like a loan or interview.
Weaknesses
- It ignores whether the groups truly differ in the outcome. If base rates legitimately differ, forcing equal selection can lower accuracy or harm the qualified.
- It can be gamed by selecting unqualified members of a group just to hit the quota.
Key idea
Demographic parity equalizes the positive prediction rate across groups regardless of true outcomes, making it simple to audit but blind to legitimate base rate differences.