Adverse Impact Ratio (Four-Fifths Rule)

The adverse impact ratio, commonly called the four-fifths rule, is a statistical test from the federal Uniform Guidelines on Employee Selection Procedures used to flag potential discrimination in hiring, promotion, or other selection decisions. If any group's selection rate falls below 80% of the highest-selected group's rate, the process is generally treated as showing adverse impact.

How Is the Adverse Impact Ratio Calculated?

Divide the selection rate of the group in question by the selection rate of the highest-selected group. If 60% of male applicants are hired but only 40% of female applicants are hired, the ratio is 40/60, or about 67% — below the 80% threshold, indicating adverse impact against the lower-selected group, even if no one intended to discriminate.

Is Falling Below 80% Automatically Illegal?

No. The four-fifths rule is a practical screening threshold used by the EEOC, not an automatic legal finding of discrimination. A ratio below 80% shifts the burden to the employer to show the selection procedure is job-related and consistent with business necessity — it opens the door to scrutiny rather than closing the case.

How Does This Relate to Disparate Impact and Disparate Treatment?

The four-fifths rule is the primary statistical tool used to establish disparate impact, a discrimination theory based on outcome rather than intent. It differs from disparate treatment, which requires showing intentional differential treatment. A neutral-seeming process can still fail the four-fifths test and trigger disparate impact liability.

Why Does This Matter for AI Hiring Tools?

Every major AI employment law — NYC Local Law 144, the Colorado AI Act, and similar frameworks — uses the same underlying logic to evaluate automated employment decision tools: calculate selection rates by group, compare them, and flag ratios below the threshold. An AI bias audit is, in practice, a four-fifths calculation applied to an algorithm's outputs instead of a human recruiter's decisions.

What Are Common Pitfalls in Applying This Rule?

Small sample sizes can produce misleadingly large swings in the ratio — a handful of hires in a small applicant pool can push a ratio above or below 80% without reflecting a real pattern. Statisticians typically pair the four-fifths test with a test of statistical significance before drawing conclusions, rather than relying on the ratio alone.

How Should HR Teams Monitor This?

Track selection rates by demographic group at every stage of the pipeline — application, interview, offer — not just the final hire rate, since adverse impact can appear and disappear at different stages. Recalculating regularly, rather than only when a complaint arises, catches drift early enough to correct a process before it becomes a legal exposure.

Discover how our HR solutions streamline onboarding, boost employee engagement, and simplify HR management

Request a Demo

Frequently Asked Questions

Q: What is the exact threshold for the four-fifths rule?
A: A selection rate below 80% (four-fifths) of the highest-selected group's rate generally indicates adverse impact.

Q: Does failing the four-fifths test mean an employer discriminated?
A: No, it shifts the burden to the employer to justify the selection procedure as job-related, not an automatic finding of discrimination.

Q: How is the adverse impact ratio different from disparate treatment?
A: The ratio measures outcomes regardless of intent; disparate treatment requires showing intentional differential treatment.

Q: Why do small applicant pools cause problems with this rule?
A: Small samples can swing the ratio significantly without reflecting a real, statistically meaningful pattern.

Q: Do AI bias audit laws use the same four-fifths calculation?
A: Yes, most AI employment laws apply the same selection-rate comparison logic to an algorithm's outputs.

Q: At what stage of hiring should this ratio be measured?
A: At every stage, application, interview, and offer, since adverse impact can appear at one stage and not another.