Demographic parity, also called statistical parity, is one of the simplest definitions of group fairness for an automated decision system. It requires that the share of people who receive a positive decision, such as a loan approval or a job interview, be the same across protected groups like gender or race. If 40 percent of one group is approved, demographic parity demands that roughly 40 percent of every other group be approved too. The criterion was formalized in the 2011 paper “Fairness Through Awareness” by Dwork and colleagues.
Its appeal is that it is easy to state, easy to measure, and does not require knowing the true outcomes the model is trying to predict; you only need the decisions and the group labels. That makes it attractive to regulators and auditors who want a clear, checkable number.
Its weakness is equally well known. Demographic parity ignores whether the positive decisions are actually warranted. If two groups genuinely differ in the base rate of the thing being predicted, forcing equal approval rates can mean approving less-qualified members of one group or rejecting more-qualified members of another, which can be both inefficient and, in some framings, unfair to individuals. This is why later work introduced alternatives like equalized odds.
Why a business reader should care: demographic parity is often the first fairness metric an organization reaches for because it is intuitive, but it is rarely the only one that matters. Choosing it without understanding what it ignores can create the appearance of fairness while introducing new problems.