Policy
The Four-Fifths Rule Is a Trigger for Scrutiny, Not the Legal Threshold
Adverse impact is a substantially different rate of selection that works to the disadvantage of a race, sex or ethnic group. The four-fifths rule divides each group's selection rate by the highest group's rate and flags anything under 0.8. It is a rule of thumb for where federal enforcement agencies look, written into 29 CFR 1607.4(D) in 1978, and it is not the legal test. The same paragraph says smaller differences can still be adverse impact where they are significant. At applicant volumes, significance arrives first.
The takeFor a funnel screening thousands, the ratio is the wrong first metric and a fine second one. It was built for a 1970s caseworker with a few hundred applications and a hand calculator, and it still behaves like it. At four figures of volume a disparity a court would take seriously can sit comfortably above 0.8, which means an employer watching only the ratio finds out from the complaint. Compute significance too, and compute it stage by stage. A ratio that clears 0.8 is where the question starts.
Where Olive fits
Open a role and see what the work shows
An Olive report carries no composite and no cut point, so a selection rate is something an employer computes from its own decisions rather than something the report hands over. The six findings come back in words: demonstrated, partly demonstrated, not demonstrated.
Rank your shortlistHow is the ratio actually computed?
Take each group's selection rate, which is the number selected divided by the number who applied. Divide every group's rate by the rate of the group with the highest one. Anything under four-fifths, or 80%, is generally regarded by federal enforcement agencies as evidence of adverse impact 1. That is a ratio of two rates, not a gap in percentage points, and the two are confused constantly.
Worked, with round numbers. Five hundred candidates from one group apply and 50 advance, a selection rate of 10%. Two hundred from another group apply and 12 advance, a rate of 6%. The impact ratio is 6 divided by 10, which is 0.6, and that sits under the threshold. Now notice what the arithmetic never asked: how many people were in each group, whether a difference that size could have arisen by chance, and what the step was measuring in the first place.
Adverse impact attaches to one practice at a time. The guidelines define it as a substantially different rate of selection working to the disadvantage of a race, sex or ethnic group, and define a selection procedure as any measure used as a basis for an employment decision 6. So the resume screen, the assessment, the interview and the offer decision each get their own numbers, and bundling them into one number for the whole funnel is convenient reporting and weak evidence.
Why isn't 0.8 a safe harbor?
Because the agencies that wrote it said so. Their 1979 interpretive questions and answers call the four-fifths rule a rule of thumb, not intended as a legal definition, and a practical means of keeping enforcement attention on serious discrepancies 2. Asked directly whether it meant the guidelines tolerate up to 20% discrimination, they answered no: the rule speaks only to adverse impact and not to the ultimate question of unlawful discrimination 2.
That cuts both ways, which is the part vendors quote selectively. Clearing 0.8 is not a clean bill of health, and missing it is not a finding against you. It is a screen for attention.
The number's provenance is worth knowing before you build a policy on it. A peer-reviewed history traces the earliest mention of the rule to California regulatory guidance in 1972, six years before the federal guidelines, and reports that no official written justification for the value four-fifths could be found. The only account the authors located is a recollection that the test split the difference between a 70% camp and a 90% camp 3. Treat that as what it is, an anecdote the authors label as one, and draw the narrow conclusion: 0.8 is an administrative convention, so a compliance program that treats it as a scientific boundary has mistaken a threshold for a finding. The legal analysis underneath asks more of an employer than the ratio does.
What should a high-volume funnel compute instead?
Both numbers. Keep the ratio, because that is what an agency screen looks like, and add a significance test, because at volume it fires first. The regulation itself says smaller differences may still be adverse impact where they are significant in both statistical and practical terms, and that larger differences may not be adverse impact where they rest on small numbers and are not statistically significant 1. With tens of thousands of applications, small differences stop being noise.
Run it stage by stage. The end-to-end number is the one that hides things. The statute normally requires a complaining party to identify each particular practice causing the impact, with an exception where the elements of the decision-making process are not capable of separation for analysis 4. Two consequences follow. A challenge is aimed at a step, so a step is the level to measure. And where those elements cannot be separated, the statute allows the whole process to be analyzed as one practice, so opacity raises the employer's exposure. How far that clause reaches for an algorithmic tool is unsettled.
The inputs are dull and they are the whole job: applications and selections by stage, by requisition, with dates, and a defensible definition of who counted as an applicant. Most teams discover they cannot produce that history when they need it, either because the resume screen throws away people nobody counted or because the tool holds the data. Running an adverse impact audit when the vendor holds the data is a contract problem before it is a statistics problem.
What happens after a stage fails?
Nothing automatic. A ratio under 0.8 is not a finding of discrimination and triggers no duty to report, notify or stop using the tool. Under the statute a plaintiff must show a particular practice causing the impact, the employer must then demonstrate the practice is job related for the position in question and consistent with business necessity, and the plaintiff can still win by identifying a less discriminatory alternative the employer refuses to adopt 4.
So the response is a decision, not a filing. Change the step, change the threshold, or write down the job-relatedness evidence and the alternatives you considered and why you rejected them. The third option is only worth anything if the writing exists before the dispute, which is the same reason a bias audit is a commissioning decision. Which of the three fits, and what the record has to say, is a question for counsel on your own process.
A less discriminatory alternative is usually concrete. It means the same stage run a different way: a different instrument, a lower cut, a structured version of a step that used to be a conversation, or the same tool used as one input instead of a gate. If a cheaper alternative was on the table and you can show why it did not serve the job, that is a record worth having before anyone asks for it.
One more thing has changed about who asks. Executive Order 14281, signed April 23, 2025, states a policy of eliminating disparate-impact liability to the maximum degree possible and directs all agencies to deprioritize enforcement of statutes and regulations to the extent they include it, naming Title VII's own section 5. That is an instruction to agencies about their own enforcement priorities. The statute is unchanged, private plaintiffs can still bring these cases, state and local law is untouched, and a quieter federal regulator changes the odds of an investigation without changing the standard a court applies.
And the ratio is not the only route to a federal case. In EEOC v. iTutorGroup the agency sued over tutor application software programmed to reject female applicants aged 55 or older and male applicants aged 60 or older; the companies settled in 2023 for $365,000 under a consent decree that admitted no liability 7. That was a written cutoff rather than a learned one, which is the honest way to describe it, and it is the clearest evidence that automated screening draws federal attention under statutes the four-fifths rule was never written for.
Common questions
Is the four-fifths rule part of Title VII?
No. It is one paragraph in a 1978 regulation, the Uniform Guidelines on Employee Selection Procedures, adopted jointly by four federal agencies and unamended since 1981. Title VII is the statute, and the operative disparate-impact test sits at 42 U.S.C. 2000e-2(k). The guidelines tell agencies where to look. The statute decides what a court asks. A vendor certifying a tool against the four-fifths ratio has certified it against neither.
Does a group difference in test scores prove adverse impact?
No. Adverse impact is about selection rates, which depend on how the output gets used: the cut point, the selection ratio, the applicant pool, and whether the step is a gate or one input among several. A method with a known subgroup difference can produce no impact at one threshold and clear impact at another. Measure the rate at which people move through the step, not the spread of the numbers going into it.
Should the ratio be computed end to end or stage by stage?
Both, and the stage view is the one that matters. A plaintiff normally has to identify the particular practice causing the impact, so each stage can be challenged on its own. A funnel that looks clean overall can carry one step that is not, and a passing bottom line is not a defense to it. Keep counts of applicants and selections at every stage, with dates, so the analysis can run at the level a claim is actually brought.
How small a sample is too small for the ratio to mean anything?
There is no fixed floor, and the regulation says as much: larger differences may not be adverse impact where the numbers are small and the difference is not statistically significant. In practice a handful of selections can flip a ratio across 0.8 on a single hire. Run the significance test alongside the ratio at every volume. At low counts it protects you from a false alarm, and at high counts it catches what the ratio misses.
Does the four-fifths rule apply to age or disability claims?
The Uniform Guidelines reach only selection procedures under Title VII, which is race, color, religion, sex and national origin. Age and disability sit outside that boundary, under the ADEA and the ADA, and the ADA's individualized framework does not reduce to a group selection rate at all. Age is also where automated screening has already produced federal enforcement: the EEOC's 2023 iTutorGroup settlement resolved claims that application software auto-rejected older applicants. The absence of a ratio rule there is not the absence of exposure.
References
- 1. 29 CFR 1607.4 - Information on impact (Uniform Guidelines on Employee Selection Procedures, 1978) law.cornell.edu Supports the four-fifths ratio as generally regarded evidence of adverse impact, and the same paragraph's caveats about small and significant differences.
- 2. Questions and Answers to Clarify and Provide a Common Interpretation of the Uniform Guidelines on Employee Selection Procedures (Q.11, Q.19) eeoc.gov Supports that the drafting agencies called the rule a rule of thumb, not a legal definition, and refused the 20% reading.
- 3. The four-fifths rule is not disparate impact: A woeful tale of epistemic trespassing in algorithmic fairness facctconference.org Supports the 1972 California origin, the absence of a written justification for the value, and the 70/90 compromise recollection.
- 4. 42 U.S.C. 2000e-2(k) - Burden of proof in disparate impact cases uscode.house.gov Supports the burden-shifting sequence, the particular-practice requirement, and the not-capable-of-separation exception.
- 5. Executive Order 14281 of April 23, 2025 - Restoring Equality of Opportunity and Meritocracy govinfo.gov Supports the direction to agencies to deprioritize disparate-impact enforcement, cited as executive policy rather than a change to the statute.
- 6. 29 CFR 1607.16 - Definitions (Uniform Guidelines on Employee Selection Procedures) law.cornell.edu Supports the definition of adverse impact as a substantially different rate of selection working to the disadvantage of a race, sex or ethnic group, and of a selection procedure as any measure used as a basis for an employment decision.
- 7. iTutorGroup to Pay $365,000 to Settle EEOC Discriminatory Hiring Suit eeoc.gov Supports the age example: tutor application software programmed to reject applicants above a date-of-birth cutoff, resolved by consent decree with no admission of liability.
7 sources, numbered by first appearance. How Olive sources claims
General guidance, not legal advice. Hiring rules differ by state and country and change often; check anything here against your own counsel before you act on it.
Olive assesses how a person works with AI. It does not detect AI-written documents, and it never produces a score, a ranking, or a match percentage for a person. Candidates read the same report the employer reads.