Home / Curated / Evaluation Criteria / Demographic parity

Demographic parity

Evaluation Criteria AIMech:DemographicParity DRAFT

Aliases: Statistical parity

A group-level prediction criterion requiring the distribution of a decision or prediction to be independent of a specified protected attribute; for a binary decision, the positive-decision rate is equal across the compared groups.

Identity Notes & Concept Scope The record denotes exact distributional parity, also called statistical parity in the inspected fairness literature. It is not equalized odds, equality of true-positive rates, an approximate disparity tolerance, or a complete fairness judgment. All 442 AIO identities/scoped synonyms, 94 native identities and ignored-inclusive text in both checkouts were searched. Conditional parity and parity for selected subgroups require separately stated conditioning and scope.

Claims & Grounded Evidence

3 Claims

Scope #criterion-scope

Compare the prediction distributions conditional on each protected group without conditioning on the target outcome. In binary classification the condition reduces to equality of P(prediction=1 | A=a) across groups. For the statistical-parity formulation of Dwork and colleagues, exact parity is the zero-tolerance case in which the total-variation distance between group output distributions vanishes.

supports Source: Equality of Opportunity in Supervised Learning · Locator: Section 1, demographic-parity definition and limitations, PDF pp. 1–2
supports Source: Fairness Through Awareness · Locator: Section 3, Definition 3.1, printed p. 7 / PDF p. 8; Section 3.1 examples, printed p. 8 / PDF p. 9

Scope #group-parity-boundary

The statistical-parity formulation compares group output distributions by total variation with a stated tolerance; exact parity is its zero-tolerance case. The accompanying examples show that group parity can coexist with unequal treatment within groups and does not imply parity for all subgroups.

supports Source: Fairness Through Awareness · Locator: Section 3, Definition 3.1, printed p. 7 / PDF p. 8; Section 3.1 examples, printed p. 8 / PDF p. 9

Limitation #causal-equivalence-assumptions

Rosenblatt and Witter derive connections with counterfactual fairness using their Criteria 3 independence assumption. Their conversion from demographic parity introduces a causal model and latent-variable construction of their choice; they explicitly distinguish that setting from checking a predictor against a fixed causal model. These results do not justify treating the two criterion identities as synonyms.

supports Source: Counterfactual Fairness Is Basically Demographic Parity · Locator: Related Work distinction between a fixed and freely constructed causal model, PDF p. 3; Criteria 3 and Theorems 4–5, PDF p. 4

⚠️ Curated Limitations & Negative Results

Matching group output rates does not ensure similar individuals receive similar treatment or that parity holds within every subgroup.
If target outcomes have different prevalences by group, a perfectly accurate predictor can violate demographic parity. This statistical observation does not by itself choose which fairness criterion a domain should use.
Exact independence is distinct from approximate parity within a specified tolerance; an empirical estimate needs uncertainty and population context.
Connections to counterfactual fairness depend on causal-model assumptions and how the predictor/model is constructed. Shared population statistics do not make the two definitions interchangeable for a fixed causal model.

Artificial Intelligence Ontology (AIO) Grounding

Mapped AIO Classes

AIO Target Class	Mapping Relation	Mapping Rationale
`AIO:Classification` Classification	related_match	Demographic parity directly constrains the distribution of classifier decisions across groups. Classification is a task to which this output criterion applies, not an equivalent criterion or a bias type; the mapping does not limit the mathematical independence condition to binary outputs.

Evidence Sources & Primary Literature

3 Sources

Source ID	Publication / Document	Type / Access	Reference & Checksum
`original-paper`	Equality of Opportunity in Supervised Learning Moritz Hardt, Eric Price, Nathan Srebro	primary paper full_text	ID: arXiv:1610.02413 sha256:e92b4f579180…
`statistical-parity-formulation`	Fairness Through Awareness Cynthia Dwork, Moritz Hardt, Toniann Pitassi, Omer Reingold, Richard Zemel	primary paper full_text	ID: arXiv:1104.3913 sha256:0a5afde6e5a2…
`causal-parity-analysis`	Counterfactual Fairness Is Basically Demographic Parity Lucas Rosenblatt, R. Teal Witter	primary paper full_text	ID: arXiv:2208.03843 sha256:ebc09da1ee78…

Curation Provenance

Curator: Codex

Method: ai_assisted

Model: GPT-6

Created: 2026-10-11T04:48:50Z

Updated: 2026-10-11T04:48:50Z