Demographic Parity and Statistical Parity
What you'll learn: How to measure whether positive predictions are distributed equally across different demographic groups.
The Core Idea
Demographic parity (also called statistical parity) is one of the most intuitive fairness metrics. It asks a simple question: Do different groups receive positive outcomes at the same rate?
Formally, a model satisfies demographic parity if the probability of receiving a positive prediction is the same across all protected groups. For example, if your loan approval model accepts 60% of applications from Group A, it should also accept approximately 60% from Group B.
Mathematically, for groups A and B:
P(Ŷ = 1 | A = a) = P(Ŷ = 1 | A = b)
Where Ŷ is the prediction and A is the protected attribute (like race or gender).
Why This Matters
Imagine a hiring algorithm that recommends 40% of male candidates but only 20% of female candidates for interviews. Even if individual predictions seem justified, this disparity signals potential systemic bias. Demographic parity catches this pattern by examining outcome distributions rather than individual decisions.
The Practical Test
To evaluate demographic parity:
- Calculate the positive prediction rate for each demographic group
- Compare these rates across groups
- The closer they are, the more the model satisfies demographic parity
A common threshold is the "80% rule" — no group should receive positive outcomes at less than 80% of the rate of the highest group.
Key Takeaway: Demographic parity ensures that positive predictions are distributed proportionally across groups, regardless of ground truth labels — it focuses purely on outcome equality, not prediction accuracy.