Every hiring step sorts people. A CV screen, a knock-out question, an online test, a structured interview and an AI ranking all decide who moves on. The law cares about two ways that sorting can go wrong. Someone can be treated differently because of who they are, or a rule that looks fair can quietly shut out a whole group. The first is disparate treatment. The second is disparate impact.
This guide explains both in plain English, shows how to run a basic disparate impact analysis on your own funnel, and covers what it means for AI screening and assessments. It focuses on US federal law, where the terms come from, with short notes on New York City, the UK and the EU. It is a plain-English explainer, not legal advice; talk to employment counsel about your own process.
What is disparate impact?
Disparate impact is when an employment practice that is neutral on its face falls more harshly on a protected group, and the employer cannot show the practice is job-related and necessary for the business. Intent does not matter.
The idea comes from the US Supreme Court's 1971 decision in Griggs v. Duke Power Co. The company required a high school diploma and a general intelligence test for certain jobs. Both rules applied to everyone. The Court found them unlawful because neither had been shown to relate to performance in the job, and held that good intent does not save a practice that works as "built-in headwinds" for minority groups.
Congress wrote the rule into Title VII of the Civil Rights Act with the Civil Rights Act of 1991. Under 42 U.S.C. 2000e-2(k), a disparate impact claim works in three steps:
- The candidate (or agency) shows that a particular employment practice causes a disparate impact on the basis of race, colour, religion, sex or national origin.
- The employer must then show the practice is job-related for the position and consistent with business necessity.
- Even if it does, the employer can still be liable if there is a less discriminatory alternative that it refuses to adopt.
Disparate impact also applies under other laws. The Supreme Court held in Smith v. City of Jackson (2005) that the Age Discrimination in Employment Act allows impact claims, with a defence for reasonable factors other than age. The Americans with Disabilities Act bars tests and selection criteria that "screen out or tend to screen out" people with disabilities unless they are job-related and consistent with business necessity.
What is the adverse impact definition?
Adverse impact is the term the federal Uniform Guidelines on Employee Selection Procedures use for the same effect: "a substantially different rate of selection in hiring, promotion, or other employment decision" that works to the disadvantage of a race, sex or ethnic group.
In practice, HR teams and test publishers say "adverse impact" for the measured gap in selection rates, and lawyers say "disparate impact" for the legal claim built on it. The Uniform Guidelines were adopted jointly in 1978 by the EEOC, the Department of Labor, the Department of Justice and the then Civil Service Commission, and they still sit in the Code of Federal Regulations at 29 CFR Part 1607.
Comparing Disparate Impact and Disparate Treatment in Hiring? Try a third option on a live role
Cohesyve takes a job description, finds candidates, reaches out to them and runs an AI job simulation of the real work, then returns a ranked shortlist. Ten candidates free, no card.
What is the difference between disparate impact and disparate treatment?
Disparate treatment is intentional discrimination: a candidate is treated differently because of a protected characteristic. Disparate impact does not need intent: a neutral rule produces unequal outcomes that the employer cannot justify.
| Disparate treatment | Disparate impact | |
|---|---|---|
| What it is | Treating a person differently because of a protected characteristic | A neutral practice that selects a protected group at a much lower rate |
| Intent | Required (motive can be inferred from evidence) | Not required |
| Typical evidence | Comments, inconsistent reasons, how similar candidates were treated | Selection rates by group, statistical tests |
| Usual legal test (US) | McDonnell Douglas burden-shifting: prima facie case, employer's legitimate reason, proof the reason is a pretext | 42 U.S.C. 2000e-2(k): impact shown, then job-relatedness and business necessity, then less discriminatory alternative |
| Employer's defence | A legitimate, non-discriminatory reason for the decision | The practice is job-related and consistent with business necessity |
| Hiring example | A manager rejects candidates over 50 for a "young team" | A physical test unrelated to the job's demands passes far fewer women |
The burden-shifting test for treatment comes from McDonnell Douglas Corp. v. Green (1973). A candidate shows they belong to a protected group, applied and were qualified, were rejected, and the employer kept looking for applicants. The employer then gives a legitimate reason, and the candidate can show that reason was a pretext.
The two theories collide in one important way. In Ricci v. DeStefano (2009), the city of New Haven threw out promotion exam results because white candidates had outperformed minority candidates, fearing an impact lawsuit. The Supreme Court held that discarding the results was itself disparate treatment. An employer needs a "strong basis in evidence" that it would face impact liability before taking race-conscious action like that. The lesson for recruiters: check a procedure for impact before you rely on it. Reviewing it only once you have seen who passed is the risky order.
How do employers run a disparate impact analysis?
A disparate impact analysis compares the selection rate of each group at each hiring step. The usual first check is the four-fifths rule, followed by a test of statistical significance when the numbers allow.
Step 1: calculate selection rates
The Uniform Guidelines define the selection rate as the proportion of applicants or candidates who are hired, promoted or otherwise selected. For each step, divide the number who passed by the number who took part, for each group. To do this you need applicant data by sex and by race or ethnic group; the Uniform Guidelines ask employers to keep those records.
Step 2: apply the four-fifths rule
Under 29 CFR 1607.4(D), federal agencies generally regard a selection rate for any race, sex or ethnic group that is less than four-fifths (80%) of the rate for the group with the highest rate as evidence of adverse impact.
A hypothetical example for one online assessment:
| Group | Took the test | Passed | Selection rate | Impact ratio |
|---|---|---|---|---|
| Men | 200 | 120 | 60% | 1.00 (highest) |
| Women | 150 | 60 | 40% | 0.67 |
The impact ratio is the group's rate divided by the highest group's rate: 40% divided by 60% is 0.67. That is below 0.80, so the test shows adverse impact under the rule of thumb and needs a closer look.
Step 3: check significance and sample size
The four-fifths rule is a rule of thumb, and the same regulation says so. Smaller gaps can still count as adverse impact if they are significant in both statistical and practical terms. Larger gaps may not count if they rest on small numbers that are not statistically significant. Courts often look at standard deviations: in Hazelwood School District v. United States (1977), the Supreme Court said a difference of more than two or three standard deviations makes chance an unlikely explanation.
Step 4: look at each step, as well as the overall result
The Uniform Guidelines say that if the whole process shows no adverse impact, agencies usually will not examine each component. The Supreme Court has gone further. In Connecticut v. Teal (1982), the Supreme Court held that a fair overall result does not excuse a pass-fail test that screened out Black candidates at an earlier step. Measure every step that removes people: knock-out questions, the CV screen, the assessment and the interview.
Step 5: if you find impact, review the procedure
Finding adverse impact does not make a procedure illegal. It means you need evidence that it is job-related, and you should look for an alternative. The Uniform Guidelines say that where two procedures are substantially equally valid, you should use the one with less adverse impact. Evidence of job-relatedness usually means a validation study; our guides to criterion-related validity and content validity explain the two most common kinds.
What you cannot do in the US is adjust scores or set different cut-off scores by race, colour, religion, sex or national origin. 42 U.S.C. 2000e-2(l) makes that an unlawful employment practice in its own right.
Which hiring practices most often cause disparate impact?
Any step that filters people can cause impact. These are the ones to check first.
| Practice | Why it can cause impact | What to check |
|---|---|---|
| Degree or diploma requirements | The Griggs pattern: a credential the job does not need | Whether people without it do the job well |
| General cognitive tests | A broad test can measure more than the job needs, and the cut-off decides who is screened out; see our guide to cognitive ability tests for hiring | Validation for this role; the cut-off you set |
| Physical tests | Can screen out women or disabled candidates | Whether the standard matches the real job |
| Knock-out questions | Gaps in work history, location or availability rules can fall on one group | Each question's pass rate by group |
| Unstructured interviews | Inconsistent questions invite both treatment and impact claims | Use a structured interview with a shared scorecard |
| AI CV screening and ranking | Learned patterns can track protected characteristics through proxies | Selection rates by group; what the model uses |
What does disparate impact mean for AI screening and assessments?
An AI screen is a selection procedure, so it faces the same disparate impact rules as a paper test. The Uniform Guidelines cover "tests and other selection procedures" used as a basis for any employment decision, and the employer remains responsible when it relies on a vendor's validation evidence.
Four practical points follow.
- Ask the vendor for impact data before you buy. Request selection rates by sex and race or ethnicity for roles like yours, plus the validation evidence for the job. Our guide to enterprise assessment platforms lists the questions to put in the RFP.
- Re-run the analysis on your own applicants. A vendor's numbers come from someone else's candidate pool. Run the four-fifths check on your own data once you have enough candidates, and again whenever the model, the role or the cut-off changes.
- Know what the model looks at. A system trained on past hires can learn proxies for protected characteristics. Rankings you cannot explain are hard to defend as job-related.
- Keep an accommodation path. Under the ADA, tests must be given so the results reflect the skill being measured, rather than a candidate's impaired sensory, manual or speaking skills, unless those are what the test measures. A timed, voice-only or video-only format needs an alternative on request.
Some jurisdictions now make this analysis a legal duty. New York City's Local Law 144, enforced since 5 July 2023, requires an independent bias audit of an automated employment decision tool within a year before use. The audit must calculate selection or scoring rates and impact ratios across sex, race/ethnicity and intersectional categories, and employers must publish a summary and give candidates 10 business days' notice. The law does not prescribe what to do with the results; other anti-discrimination laws still apply. In the EU, the AI Act's obligations for recruitment AI apply from 2 December 2027. Our guide to AI hiring laws in 2026 covers each jurisdiction in detail.
Assessments built from the real work of the job can make job-relatedness easier to show, because the link between task and role is visible. That is the approach Cohesyve takes. You give it a job description, and it sources candidates, handles outreach, puts them through AI job simulations of the real work and returns a ranked shortlist for your team to review. Plans are a subscription from $199/month billed annually, and you can start with 10 free candidates. Whichever tool you use, ours included, run the same selection-rate check on your own candidates.
Has disparate impact changed under the 2025 executive order?
Federal enforcement priorities changed in 2025, and the statute stayed the same. Executive Order 14281, signed on 23 April 2025, directs federal agencies to deprioritise enforcement of laws that include disparate impact liability, names 42 U.S.C. 2000e-2 among them, and orders a review of pending matters, including those under Title VII, that rely on disparate impact.
The order does not repeal Title VII's disparate impact section. The EEOC's 2023 technical assistance on adverse impact and AI is no longer on its website, but the Uniform Guidelines remain in the Code of Federal Regulations, and New York City's audit rules still require impact ratios. For a hiring team, the practical position is the same as before: measure selection rates by group, keep the records, and be ready to show that each step is related to the job.
What about disparate impact in the UK?
UK law uses different terms for the same two ideas. Under the Equality Act 2010, direct discrimination means treating someone less favourably because of a protected characteristic. Indirect discrimination means a rule or arrangement that applies to everyone but puts people with a protected characteristic at an unfair disadvantage. Indirect discrimination can be lawful if the rule can be justified, so the same habits apply: measure outcomes by group and be able to explain why each step is needed for the job.
A short checklist for hiring teams
- List every step that removes candidates, including automated ones.
- Collect voluntary demographic data in a way that keeps it away from decision-makers.
- Calculate selection rates and impact ratios at each step, every quarter or after each high-volume campaign.
- Where a ratio falls below 0.80, or a gap is statistically significant, review the step's job-relatedness and look for an alternative with less impact.
- Never adjust scores or set separate cut-offs by group, and never discard results because of who passed.
- Score interviews against the same criteria for everyone, using an interview scorecard.
- Keep the records and the reasoning for each change.
Common questions
What is the disparate impact definition in simple terms? A hiring rule that looks neutral but screens out a protected group at a much higher rate, without a job-related reason. The employer does not need to have meant it.
Is adverse impact the same as disparate impact? They describe the same effect. "Adverse impact" is the Uniform Guidelines' term for the measured gap in selection rates; "disparate impact" is usually the name of the legal claim.
Is the four-fifths rule the law? It is a rule of thumb set out in federal regulations, used by agencies to flag possible adverse impact. Courts also look at statistical significance and sample size, so passing or failing the 80% line is not the end of the analysis.
Does disparate impact apply to AI hiring tools? Yes. An AI screen is a selection procedure, and the employer is responsible for its effect even when a vendor built it. New York City also requires a bias audit, done within the year before use, for automated employment decision tools.
Can we lower the pass mark for one group to fix adverse impact? No. US federal law bans adjusting scores or using different cut-off scores by race, colour, religion, sex or national origin. Change the procedure for everyone instead.
Can disparate treatment and disparate impact happen at the same time? Yes. A process can have a biased decision-maker and an unjustified test in it at once, and a candidate can bring both claims.
What is a disparate impact analysis? A comparison of selection rates for each group at each hiring step, usually starting with the four-fifths rule and followed by significance testing when the numbers are large enough.