Course contentsShow
← AI Engineering
Lesson 1566 of 1,88638. Bias, Fairness, and AlignmentFree lesson

Demographic Parity and Statistical Parity

Measuring whether predictions are distributed equally across demographic groups.

Demographic Parity and Statistical Parity

What you'll learn: How to measure whether an AI system's predictions are distributed equally across different demographic groups.

The Core Idea

Demographic Parity (also called Statistical Parity) is a formal fairness metric that asks: "Does my model give positive outcomes at the same rate across all demographic groups?"

In simpler terms: if your AI approves loans, recommends jobs, or flags content, demographic parity means that the percentage of positive decisions should be roughly equal across groups defined by sensitive attributes like race, gender, or age — regardless of other factors.

For example, if your loan approval model accepts 60% of Group A's applications, demographic parity requires it to also accept approximately 60% of Group B's applications.

Why It Matters

This metric directly addresses allocation harms: ensuring no group is systematically denied opportunities. It's intuitive and aligns with legal concepts like "disparate impact" in civil rights law.

However, it's also controversial. Demographic parity ignores whether groups have different base rates of qualification or need. If Group A has a 70% default rate and Group B has a 30% default rate, enforcing demographic parity might mean approving more high-risk loans in Group A or denying qualified applicants in Group B.

The Mathematical Test

P(Ŷ = 1 | A = a) ≈ P(Ŷ = 1 | A = b)

Where Ŷ is the prediction and A is the sensitive attribute (e.g., demographic group).

Key Takeaway: Demographic parity measures fairness by ensuring equal rates of positive predictions across groups, but may conflict with accuracy or other notions of fairness when groups differ in underlying characteristics.