Adverse impact ratio

The adverse impact ratio is the selection rate of a given demographic group divided by the selection rate of the most-favored group; under the EEOC four-fifths rule, a ratio below 0.80 is treated as evidence of disparate impact.

Direct answer

The adverse impact ratio (AIR) measures whether a hiring screen passes one demographic group at a much lower rate than another. Compute it by dividing the selection rate of the group in question by the selection rate of the most-favored group. The EEOC's 1978 Uniform Guidelines say a ratio below 0.80 is treated as evidence of disparate impact and triggers a deeper look. It matters because the AIR is the actual number a Local Law 144 bias audit reports, and the number a plaintiff's expert witness cites. Who needs to know: anyone running a structured screen on more than ~30 candidates per group. This page is an explainer, not legal advice — get counsel for specific cases.

In detail

The arithmetic

For each demographic group: selection rate = (candidates from the group who passed the screen) / (candidates from the group who took it). Then take the selection rate of the group with the highest pass rate and divide every other group's rate by it. The result is the adverse impact ratio. Example: if 60% of Group A passes and 40% of Group B passes, B's AIR is 0.40 / 0.60 = 0.67. Below the 0.80 four-fifths threshold — adverse impact flagged. The Uniform Guidelines published by the EEOC, DOJ, OPM, and Department of Labor are the canonical source for the four-fifths rule.

When it applies

For LL144 bias audits, the AIR is computed across sex categories (M / F / unknown), race-and-ethnicity categories per the EEO-1 schema, and the intersection of the two (an independent auditor handles this). The categories the candidate self-identifies into come from the candidate-notice flow before the AEDT runs. Outside LL144, plaintiff's counsel will still use AIR as the standard four-fifths-rule test in federal Title VII employment discrimination matters.

Common misconceptions

Three things to keep straight. (1) A ratio above 0.80 is not a clean pass — it is the absence of an automatic flag. A regulator or court can still find disparate impact at a higher ratio if the sample is large enough and a statistical test reaches significance. (2) AIR is meaningless on tiny samples. Below roughly 30 candidates per group the number jumps around and the four-fifths rule should not be the only diagnostic. (3) AIR describes outcomes only, not the cause — it does not tell you whether the screen itself is unfair, just that the pass rates differ. The investigation comes next.

Related concepts

The AIR is the headline number in any AEDT bias audit. The input data comes from the rubric scoring output, applied across a population of AI first-round interview sessions. The framework around what gets logged and exported is on the compliance page.

How Vettika handles it

Vettika logs per-session selection outcomes and the candidate's self-identified demographic categories (where the candidate chooses to disclose). The independent bias auditor pulls that export to compute the AIR per group — we do not compute or self-certify the ratio ourselves, because that is the part the law requires be done independently.

Try a first-round interview, no card.

The audit log is on by default — every session records what the auditor needs.

Run your first 3 free