How to Run a Four-Fifths Adverse Impact Analysis on Your Own Applicant Flow

A four-fifths rule analysis tells a US employer whether one step in its hiring process selects a protected group at a substantially lower rate than another. It needs a spreadsheet, your applicant flow data, and an afternoon. The method comes from 29 CFR 1607.4(D).

Most employers never run it, or run it once across the whole funnel and file the result. Both leave a hiring gate rejecting people at a rate nobody measured.

Key Takeaways

• Divide each group's selection rate by the highest group's rate. Below 0.80 is generally treated as evidence of adverse impact.
• A ratio below 0.80 does not prove discrimination. It shifts the question to whether the procedure is job related and consistent with business necessity.
• Test every gate separately. Connecticut v. Teal (1982) held that a favorable overall result is no defense to a claim about an earlier barrier.
• A ratio above 0.80 is not a clearance. Smaller gaps count where they are statistically and practically significant.
• Liability sits with the employer that uses the tool, not the vendor that built it.

What the four-fifths rule actually says

The rule sits in section 4(D) of the Uniform Guidelines on Employee Selection Procedures (29 CFR Part 1607), adopted in 1978. Part 1607 remains codified, although the executive branch's enforcement position on disparate impact changed in 2026. A selection rate for any race, sex or ethnic group that is less than four-fifths of the rate for the highest group will generally be regarded by the federal enforcement agencies as evidence of adverse impact.

Two qualifiers in the same paragraph matter as much as the threshold. Smaller differences may still constitute adverse impact where they are statistically and practically significant; larger differences may not, where they rest on small numbers. The figure screens, it does not decide legality.

A selection procedure is any measure used as a basis for an employment decision: a resume screen, a structured interview, a cutoff score, a physical ability standard, an algorithmic ranking tool. So is a hiring manager's unstructured judgment.

The five-step calculation

The arithmetic is simple. The discipline is in the setup.

• Define the gate. Pick one step and one job family. Count everyone who entered that step, not everyone who entered the process.
• Count applicants and selections per group. Selected means advanced past that gate.
• Calculate the selection rate. Selections divided by applicants, per group.
• Identify the highest rate. It is not automatically the majority group, and you do not get to pick the comparison group that flatters the result.
• Divide. Each group's rate over the highest rate gives the impact ratio. Below 0.80 is your flag.

Record the raw counts alongside the ratios, or the result cannot be re-run a year later when a charge lands.

A worked example: one assessment gate, four groups

The table below shows one skills assessment gate for a warehouse associate opening, using illustrative numbers. The highest selection rate is 32.2 percent, so every group is measured against it.

• Asian - Applicants at gate: 90; Passed: 29; Selection rate: 32.2%; Impact ratio: 1.00 (highest rate)
• White - Applicants at gate: 400; Passed: 120; Selection rate: 30.0%; Impact ratio: 0.93
• Hispanic - Applicants at gate: 180; Passed: 48; Selection rate: 26.7%; Impact ratio: 0.83
• Black - Applicants at gate: 250; Passed: 55; Selection rate: 22.0%; Impact ratio: 0.68

One group falls below the threshold. That implies nothing about intent. The employer now has to show the assessment is job related and consistent with business necessity, and consider less exclusionary alternatives.

Which groups you have to include

Run the calculation separately for every group in your applicant pool, then repeat it for sex. The Guidelines call for records by sex and by race and ethnic group using the EEO-1 categories, with determinations at least annually for each group constituting 2 percent of the relevant labor area force or of the applicable workforce.

Two percent of a 920-applicant pool is roughly 18 people. Keep records for smaller groups too: last year's small group is this year's reportable one.

The common failure is stopping at the first clean result. A process can pass on sex and fail on race, or pass on both while failing a combination nobody cross-tabulated. Each basis is a separate test.

Run the analysis at every gate, not just the end

Pooling the whole funnel into one number is the most common way this analysis produces a false clean bill of health. Section 4(C) describes a bottom-line approach under which enforcement agencies will not usually expect component-level evaluation when the total process shows no adverse impact. That discretion is narrow.

In Connecticut v. Teal, 457 U.S. 440 (1982), the Supreme Court held that a nondiscriminatory bottom line is neither a bar to a prima facie case nor a defense where an identifiable pass-fail barrier excluded individuals along the way. There, 22.9 percent of Black candidates were promoted against 13.5 percent of white candidates, and the employer still lost on the earlier written examination.

The table below shows a failing gate disappearing into a passing total.

• Resume screen - Rate, white applicants: 50%; Rate, Black applicants: 48%; Ratio (lower ÷ higher): 0.96; Reading: No flag
• Skills assessment - Rate, white applicants: 30%; Rate, Black applicants: 22%; Ratio (lower ÷ higher): 0.73; Reading: Flag, investigate
• Interview - Rate, white applicants: 40%; Rate, Black applicants: 50%; Ratio (lower ÷ higher): 0.80; Reading: At the threshold
• Offer - Rate, white applicants: 70%; Rate, Black applicants: 72%; Ratio (lower ÷ higher): 0.97; Reading: No flag
• End to end - Rate, white applicants: 4.2%; Rate, Black applicants: 3.8%; Ratio (lower ÷ higher): 0.91; Reading: Looks clean, is not

Read only the bottom row and the process looks fine. The rows above it locate the assessment gate.

Where these analyses go wrong

Small samples produce noise, not findings

With fewer than about 30 applicants in a group at a gate, two or three outcomes swing the ratio wildly and a dramatic failure may be noise. Pool the gate across a longer window or comparable requisitions.

A ratio above 0.80 is not an all-clear

Statistical significance often carries more weight than the four-fifths screen, and a gap clearing 0.80 can still matter when the numbers are large. With 5,000 applicants, a few percentage points can be highly significant. Treat 0.80 as the point where you look harder.

A failure is a trigger, not a verdict

Adverse impact is not discrimination. Under Title VII a procedure with adverse impact is lawful if it is job related and consistent with business necessity, and no equally effective, less exclusionary alternative was passed over. Without validation evidence for that gate, you are defending a number you never justified.

Your vendor's audit is not your audit

Employer liability does not transfer to a vendor. In its May 2023 technical assistance on software, algorithms and artificial intelligence in selection procedures, the US Equal Employment Opportunity Commission stated that an employer administering a selection procedure may be responsible under Title VII even where an outside vendor developed the test. A vendor audit runs on the vendor's reference population; yours runs on the people who applied to your jobs.

Turning this into a standing process

Build this as a recurring report: gate-level selection counts by race, sex and ethnic group, refreshed at least annually and before any new screening tool goes live. Capture the same fields for every instrument in the funnel, including automated reference checks and interview scorecards.

Two design choices make the data usable. Standardize gate definitions across requisitions, and hold demographic data apart from hiring decisions. If you use structured instruments such as a skill tests library or AI skill assessments, version the scoring rules and cutoffs, because a cutoff change resets the analysis.

When a gate fails, work in order: confirm the data, check sample size, test significance, review the validation evidence, then consider alternatives such as adjusting the cutoff. Document each step. Further guidance sits in Righteo's HR and pre-employment resources.

The 2026 federal enforcement shift

Executive Order 14281 directed agencies to step back from disparate-impact liability, and on June 9, 2026 the Justice Department's Office of Legal Counsel issued a memorandum opinion concluding that the EEOC's disparate-impact guidelines, UGESP included, are unconstitutional and inconsistent with Title VII. Under that framework, aptitude tests, knowledge tests and background checks are presumptively job related. The EEOC has aligned its enforcement posture.

What did not change matters more for planning. Part 1607 is still codified. An OLC opinion is executive-branch legal advice: it does not bind courts or amend the statute, and disparate-impact liability sits in 42 U.S.C. 2000e-2(k). Private plaintiffs keep an independent right of action, fourteen state attorneys general have affirmed disparate-impact enforcement, and California, New York, New Jersey and Illinois have codified it. Exposure has relocated, not disappeared. Ratios and validation evidence remain the practical defense, and they are now also the evidence base for the intentional-discrimination claims the EEOC says it is prioritizing.

This article is general information for US employers and is not legal advice. Confirm your obligations with employment counsel in the states where you hire.

Related Resources

Avoiding Adverse Impact IQ Testing Legal Risks

How to Audit Recruitment AI Algorithms for Bias

Uniform Guidelines on Employee Selection Procedures (29 CFR Part 1607)

US Equal Employment Opportunity Commission

AI skill assessments

skill tests library

automated reference checks

Frequently Asked Questions

What is the four-fifths rule in hiring?

It is the screening test in 29 CFR 1607.4(D). Calculate each group's selection rate at a hiring step and divide by the highest group's rate. A result under 0.80 is generally regarded as evidence of adverse impact.

Does failing the four-fifths rule mean my hiring process is illegal?

No. The procedure has to be justified as job related and consistent with business necessity. Plenty of procedures with adverse impact are lawful; the problem is having no validation evidence to offer.

How often should we run an adverse impact analysis?

The Guidelines contemplate determinations at least annually for each group meeting the 2 percent threshold. Run it again whenever you add a screening tool, change a cutoff or redesign a job.

Can we rely on our assessment vendor's bias audit?

Treat it as a purchasing input, not compliance. It reflects the vendor's reference population, and the EEOC has been explicit that an employer can be liable for a tool built by someone else. Run the calculation on your own data.

Does the four-fifths rule still matter in 2026?

Yes. Part 1607 remains codified, and the arithmetic still tells you which gate is rejecting people. Federal enforcement priorities have narrowed; private plaintiffs and a growing list of state statutes have not. The same records also answer the intentional-discrimination claims the EEOC is prioritizing.

Run It Gate by Gate, on a Schedule

The four-fifths calculation is easy; the value is in running it gate by gate, on your own data, on a schedule. Start with the step that rejects the most people. If a gate falls below 0.80, pull the validation evidence before changing anything, and if none exists, that absence is the first problem to solve.