Definition
Demographic Parity (also known as Statistical Parity or Group Fairness) is a fairness metric in machine learning that requires a model’s predictions to be independent of a sensitive attribute, such as race, gender, or age. It ensures that positive outcomes (e.g., hiring, loan approvals) are distributed equally across different demographic groups.
What is Demographic Parity?
Demographic parity occurs when the probability of receiving a positive outcome is the same across all demographic groups. In other words, a model satisfies demographic parity if membership in a sensitive group, such as race or gender, does not affect the likelihood of a favourable prediction. It is also referred to as statistical parity, group fairness, or independence.
For example, in a university admissions model, demographic parity would mean that male and female applicants have the same chance of being admitted. While achieving demographic parity can help prevent allocation harms, it does not account for individual qualifications, so it should be used alongside other fairness metrics to provide a more complete assessment of a model’s fairness.
Applications
Demographic parity is often applied in:
- Hiring algorithms: Ensuring equal interview rates for male and female applicants.
- Loan approvals: Equalising approval rates across demographic groups.
- University admissions: Maintaining proportional representation of applicants from different groups.
-
It can also be used as a training constraint to guide models toward fairness.
Advantages
Demographic parity is a fairness metric designed to ensure proportional representation across groups in outcomes. It is widely used because it is relatively straightforward to implement and interpret. One of its main advantages is that it helps identify allocation harms, where opportunities or resources are distributed unequally between groups. Some of its other advantages include:
- Promotes fair representation, ensuring that outcomes reflect the composition of the applicant or candidate pool.
- Helps detect allocation harms, making disparities visible and actionable.
- Simple to apply and understand, which makes it a useful first step in evaluating fairness.
Disadvantages
Despite its advantages, demographic parity also has notable limitations. By focusing solely on equal outcomes between groups, it does not consider individual qualifications or merits. This can lead to situations where less qualified individuals are selected from one group to meet parity targets. Additional challenges include:
- Potential trade-off with accuracy, as enforcing parity may reduce predictive performance for some groups.
- Underlying disparities in the data, which can make parity impractical or misleading if target labels differ systematically between groups.
- Small sample sizes in granular analyses, making statistical conclusions uncertain.
- Risk of over- or under-correction, potentially creating unintended disparities or perceptions of "reverse discrimination".
Key takeaways
- Demographic parity ensures equal opportunity across groups but does not guarantee that individual outcomes are fair or merit-based.
- It is best used alongside other fairness metrics and careful analysis of the dataset.
- Ethical and responsible AI requires understanding both the strengths and limitations of fairness metrics.