Adverse Impact in Hiring: The Four-Fifths Rule and How to Calculate It (2026)

Nobody in the room objected to the requirement. It had been on the job description for years, it was applied to all 412 applicants identically, and not one person involved intended to disadvantage anybody. Then someone ran the numbers on who made it past the resume screen and found one group passing at 42 percent and another at 21 percent. That is adverse impact, and the striking thing about it is that everything above this sentence can be true at the same time. Intent is not the test. The outcome is.

What is adverse impact?

Adverse impact is what happens when a practice that is neutral on its face produces substantially unequal selection rates between groups. The requirement, test or screening rule is applied to everyone in the same way. People from different groups pass it at meaningfully different rates. That gap is the adverse impact, regardless of what anyone intended.

The reason this concept exists at all is that most modern hiring discrimination does not look like discrimination from the inside. It looks like a requirement that sounded reasonable when someone added it, has been on the template ever since, and correlates with something other than the ability to do the job. Nobody has to be acting in bad faith for the outcome to be a problem, which is precisely why the check is statistical rather than intentional.

Adverse impact, disparate impact and disparate treatment

Three terms, two concepts. It is worth being precise, because they get used loosely and they point at genuinely different failures.

Disparate treatment is intentional. Someone was handled differently because of who they are: a question asked only of certain candidates, a standard applied more strictly to one person. It is proved with evidence about how people were treated.

Adverse impact requires no intent at all. One rule, applied uniformly, produces unequal results. It is proved with statistics about selection rates.

Disparate impact is, in practice, the same thing as adverse impact under a different name. Adverse impact is the term you meet in HR reporting and selection-testing work; disparate impact is the legal theory of liability that sits on top of it. If someone asks you the difference, the honest answer is that it is mostly vocabulary, and the analysis you run does not change.

The practical consequence is that a hiring process can be scrupulously free of disparate treatment, with every candidate asked the same questions by well-trained interviewers, and still be producing adverse impact at a stage nobody is looking at.

The four-fifths rule, and how to calculate adverse impact

The standard screening test comes from the federal Uniform Guidelines on Employee Selection Procedures and is known as the four-fifths rule or the 80 percent rule. It is four steps and one division.

• Selection rate per group. Number selected divided by number who applied, calculated at one stage. If 100 people from a group reached the resume screen and 42 passed, the rate is 42 percent.
• Find the highest rate. Whichever group passed at the greatest rate becomes the benchmark. It is not fixed to any particular group in advance.
• Impact ratio.Each group's rate divided by the highest rate. A group passing at 21 percent against a benchmark of 42 percent gives 0.50.
• Compare with 0.80. Ratios below four fifths are generally treated as evidence of adverse impact worth investigating.

Two honest caveats, because this number gets over-read in both directions. The rule is a heuristic, not a legal verdict: enforcement and litigation also involve tests of statistical significance, and with small applicant numbers the ratio swings wildly, since a couple of people moving between outcomes can drag it across the line. Equally, clearing 0.80 is not a clean bill of health. It means this particular check did not flag this particular stage.

Measure it stage by stage, not end to end

The most common analytical mistake is comparing hires against total applicants. That produces one blended number that tells you something is wrong without telling you where, and it frequently hides the problem entirely.

Look at that table for a moment. The interview stage is clean. The offer stage is clean. Interviewers are being consistent and hiring managers are deciding evenly between the people in front of them. Everything downstream of the resume screen looks like a well-run process, because it is one. The entire effect was created at a single stage, by a single criterion, weeks before any human had a conversation with anybody.

This is the general pattern. Adverse impact usually enters at the cheapest, fastest, least documented stage of hiring, which is the resume screen, and then every later stage faithfully carries it forward while looking blameless in isolation.

Where adverse impact comes from in resume screening

These are the usual culprits. None of them is automatically unlawful, and several can be perfectly justifiable for particular roles. All of them should be able to survive the question: can we show this is necessary for the job?

• Degree requirements. The best-documented source. Educational attainment is not distributed evenly, so a requirement that is neutral in form is rarely neutral in effect. If the degree is standing in for a specific capability, screen for the capability, which is the whole argument in skills-based hiring.
• Minimum years of experience. A blunt proxy that penalises career breaks, which fall unevenly, particularly on people who took time out for caring responsibilities.
• Employer and school prestige. Scanning for recognisable names measures access to those institutions at least as much as it measures ability.
• Continuous employment history. Treating gaps as a negative screens on life circumstances rather than on capability. The candidate-side view of this is explaining employment gaps, and the recruiter-side lesson is that a gap is not evidence about the work.
• Location and postcode filters. Convenient, and in some labour markets closely correlated with demographics.
• Keyword filters that auto-reject. These reject on phrasing rather than capability, and phrasing correlates with who has been coached on resume writing. The mechanics are in resume parsing.
• Unstructured human judgement under time pressure. The default when no criteria exist. It produces different standards at hour two and hour thirty, and leaves no record of what was applied to whom.

Why an unread pile is the hardest thing to defend

Here is the part that connects adverse impact to how screening actually works at volume. To run any of the analysis above, you need to know what was applied to each applicant. A stage that produced decisions has data. A stage where 300 of 412 resumes were never opened, and the rest were assessed against criteria that lived in somebody's head, produced no data at all.

That is not merely an analytics inconvenience. If the effective determinant of whether an application was read was the order it arrived in and how tired the reader was, then the process has no articulable standard behind it. There is nothing to validate, nothing to explain to a hiring manager, and nothing to show anyone who asks how candidates were evaluated. The absence of a defined, consistently applied criterion is itself the exposure.

How to screen at volume without creating adverse impact

Job-relatedness: the defence that actually matters

Finding adverse impact does not mean a practice must be abandoned. Under the disparate impact framework, a practice that produces adverse impact can still be lawful if the employer shows it is job-related for the position and consistent with business necessity, and even then a challenger may point to an equally effective alternative with less impact.

Translated into recruiting terms, three questions decide whether a criterion survives:

• Can you articulate what it measures? Not what it correlates with. What ability does it establish?
• Can you show the job needs that ability? Ideally from an analysis of the work rather than from tradition or from what the last posting said.
• Is there a less exclusionary way to measure the same thing? This is where most proxies fail. If a degree stands in for statistical literacy, screening for statistical literacy directly is both more accurate and less exclusionary, which makes the proxy hard to justify.

That third question is why documenting your criteria is worth the effort. A ranked, explained screening decision, with the matched and missing requirements attached to each candidate, is an answer to all three questions. A shortlist that emerged from an afternoon of skimming is not.

What to monitor, and how often

• Selection rates by stage, by requisition. Application to screen, screen to interview, interview to offer, offer to acceptance. The stage view is the entire point.
• Impact ratios against the highest-passing group, reviewed per role family rather than per individual req where numbers are small, because tiny samples produce noisy ratios.
• Which criterion drove each rejection. If your screening records the reason, you can identify the specific requirement creating the gap instead of guessing at the stage.
• Aggregate trends over time. A single requisition rarely proves anything. The same criterion flagging across six requisitions is a finding.
• Coverage. What share of applicants were actually evaluated. A stage with 25 percent coverage cannot be analysed honestly, whatever the ratios say, which ties this directly to candidate experience and to whether anyone ever got an answer.

Adverse impact mistakes to avoid

• Only comparing hires with applicants. One blended ratio hides which stage created the gap, and the gap is nearly always created at one stage.
• Treating 0.80 as a pass mark. It is a prompt to investigate, not a certificate. Ratios just above the line on a criterion you cannot justify are still worth fixing.
• Running the analysis on tiny samples and drawing conclusions. With 12 applicants per group, one person changes everything.
• Assuming automation removes the issue. Consistent application of a criterion with adverse impact produces adverse impact consistently. What automation adds is the record.
• Assuming a manual screen is safer. It is simply unmeasured, which is a different thing from being fair, and it is much harder to explain afterwards.
• Fixing the number instead of the criterion. Adjusting who gets through to make a ratio look better, rather than removing an unjustified requirement, creates a new and more serious problem. Fix the rule, not the arithmetic.
• Never writing the criteria down. An undocumented standard cannot be validated, defended or improved, and it is the most common state of affairs in high-volume recruiting.

Bottom line: adverse impact is not a question of anyone's character, and treating it as one is why it goes unexamined for years. It is a measurement, it lives at a specific stage, and the stage is almost always the resume screen, where criteria are vaguest and coverage is worst. Define what the job genuinely requires, apply it to every applicant on the same basis, keep the reasons, and run the ratios when the requisition closes. Start with the screening half: put your next batch through Rankid's bulk resume screening tool and get every candidate scored against the same stated criteria, with the matches and gaps written down for each one.