Updated 2026-09-16
Adverse impact ratio compares the selection rate of one demographic group to another after a hiring step — commonly using the four-fifths (80%) rule — to detect whether that step disproportionately excludes candidates from protected groups.
In U.S. federal employment discrimination guidance, a selection procedure may be flagged for adverse impact if a group's selection rate is less than four-fifths (80%) of the rate for the group with the highest selection rate. If seventy percent of Group A advances past a phone screen but only fifty percent of Group B advances, the ratio is 50/70 ≈ 0.71 — below 0.80 — triggering further review.
The rule is a screening tool, not a automatic legal finding. Statistical significance, sample size, job relatedness of the procedure, and business necessity defenses all enter formal adjudication. Small samples produce noisy ratios — ten candidates per group can swing percentages wildly without meaningful pattern.
Employers using automated or AI-led screening should monitor adverse impact at each gated stage: application knockout, assessment, AI interview score threshold, human panel. Impact can appear at one stage even when overall hiring looks balanced.
Define the stage
Example: 'Advanced past AI screen' among all who completed the AI screen.
Compute selection rate per group
Selected count ÷ eligible count for each demographic group you are permitted to analyse under local law.
Identify the highest rate
The group with the highest selection rate becomes the denominator reference.
Compute ratios
Each other group's rate divided by the highest rate. Any ratio below 0.80 warrants investigation.
Algorithms trained on historical hire data can inherit past bias — if previous human decisions under-hired a group, a model may learn to replicate that pattern. Structured rubrics and blinded scoring reduce but do not eliminate risk if the rubric itself rewards proxies for protected traits.
NYC Local Law 144 requires annual bias audits for certain automated employment decision tools used in the city — a related but distinct compliance obligation from ongoing adverse impact monitoring. See our glossary entries on EEOC hiring compliance and NYC Local Law 144 for context.
Remediation paths include revising questions, recalibrating score thresholds, adding human review for borderline cases, changing validation studies, or discontinuing a step that cannot be justified as job-related.
Not automatically. It signals disproportionate outcomes worth investigating. Legal outcomes depend on job relatedness, alternatives, and jurisdiction-specific standards.
In U.S. EEOC contexts, race, sex, and other protected categories when sample sizes support analysis. Local laws define protected classes elsewhere — including India, where categories differ.
They can reduce unstructured bias when rubrics are job-related and consistently applied — but a biased rubric or biased scoring still produces adverse impact.
Many teams review quarterly or after major process changes — new assessment vendor, new score cutoff, new question set — and at minimum annually for high-volume automated steps.
Head back to Hiring glossary or start now.
· 3 free credits · pay per interview · nothing recurring