Knowledge MaxxingDemographic parity
Demographic parity is a fairness criterion requiring a model's positive prediction rate to be roughly equal across different demographic groups.
Prerequisite: Proxy variable
What matters
This criterion focuses purely on equal outcome rates, which can conflict with other fairness measures that instead focus on equal accuracy or equal error rates between groups.
Example
A hiring model satisfies demographic parity if it recommends candidates from two different groups at similar rates overall, regardless of each group's actual qualification distribution.
Common misconception
Demographic parity does not guarantee a model is accurate or unbiased in every sense; it specifically measures equal outcome rates, which is only one particular definition of fairness.
Recall question
What specifically does demographic parity measure, compared to accuracy?