DataTerminology.comNEW
A legal and statistical concept where neutral policies or models disproportionately harm protected groups, regardless of intent.
DESCRIPTION
Disparate impact examines whether outcomes disproportionately affect protected groups even when rules or models appear neutral. In employment testing and algorithmic decisions, the 80% rule (selection rate test) is often used as a screening heuristic. Organizations assess disparate impact across stages—data collection, model training, and deployment—using fairness metrics and impact analyses. When detected, mitigations include revising criteria, auditing features, applying fairness constraints, and documenting trade-offs. It is central to responsible AI practices and compliance regimes.
ALTERNATIVE DEFINITIONS
U.S. Equal Employment Opportunity Commission
Disparate impact occurs when an employer’s neutral policy or practice has a disproportionately adverse effect on a protected group, even though the policy or practice is applied evenly.
Legal framing emphasizing neutrality of policy and protected classes.
Cornell Law School Legal Information Institute
Disparate impact refers to practices that adversely affect one group of people of a protected characteristic more than another, even when applied in a neutral fashion.
Highlights adverse effects independent of discriminatory intent.
NIST SP 1270
Disparate impact describes outcomes where algorithmic or decision-making processes produce different effects across protected classes despite ostensibly neutral criteria.
Connects disparate impact to AI and algorithmic decision-making contexts.
RELATED TERMS
Equalized Odds Equal Opportunity 80 Percent Rule AI Fairness Bias Mitigation Protected Class Impact Assessment Bias & Fairness Demographic Parity
COMPETING TERMS
Disparate Treatment
KNOWLEDGE AREAS
AI Governance Data Privacy & Compliance Domain & Business Context
TARGET AUDIENCE
Data Privacy Officer Data Governance Manager Chief AI Officer AI Product Manager Data Scientist Chief Data Officer
EXTERNAL SOURCES
- Wikipedia - Disparate impact
- EEOC - Federal guidance on employment discrimination
- NIST - Framework for managing bias in AI
- OECD AI - Policy resources on trustworthy AI
TOP QUESTIONS
Why is Disparate Impact relevant in AI systems?
AI can encode historical biases and produce uneven outcomes. Assessing disparate impact ensures fairness and compliance with anti-discrimination laws.
Category: Relevancy
How is Disparate Impact measured in practice?
Common measures include selection rate comparisons (80% rule), demographic parity, and equalized odds. Multiple metrics are used to triangulate fairness.
Category: Requirements
What is the risk of ignoring Disparate Impact in models?
Organizations face legal, ethical, and reputational harm. Biased outcomes can harm individuals and undermine trust in AI.
Category: Risks
What is the biggest misunderstanding about Disparate Impact?
That equal treatment guarantees equal outcomes; neutral policies can still disadvantage protected groups due to underlying data and structural factors.
Category: Misunderstanding
How can teams mitigate Disparate Impact?
Revise criteria, audit features, use fairness constraints, rebalance data, and document trade-offs and impacts.
Category: Management
What is a competing concept to Disparate Impact?
Disparate Treatment involves explicit differential treatment based on protected characteristics and requires proof of intent.
Category: Alternatives
What is a hot topic around Disparate Impact in AI?
Adoption of opt-out preference signals and model cards that disclose fairness testing and impacts for transparency.
Category: Topic
Who ensures Disparate Impact compliance in organizations?
Data Privacy Officers and Governance Managers collaborate with AI Product Managers and Data Scientists to test and remediate.
Category: Management
Where to start learning about Disparate Impact?
Review EEOC guidance, NIST bias management publications, and academic fairness literature to understand legal and technical aspects.
Category: Overview