How Selection Rates and Impact Ratios Are Calculated

When the Audit Summary Says 0.79 and No One on Your Team Knows Why

Your vendor's most recent bias audit summary lands in your inbox. It's two pages, mostly tables, and one column reads "Impact Ratio: 0.79" next to a demographic category you're responsible for explaining to your CHRO if anyone asks. You don't run the audit — an independent auditor does that, and you're not supposed to second-guess their engineering. But "0.79" is now a number sitting in your compliance file, and if someone asks what it means, "the auditor calculated it" isn't going to satisfy a curious board member, an HR consultant reviewing your roster, or a candidate who wants to know why an alternative process exists in the first place. By the end of this piece, you'll be able to read any published selection-rate table and reconstruct the impact ratio yourself — not audit it, just follow it.

What a Selection Rate Actually Measures

A selection rate is the simplest number in a bias audit, and also the one most people skip past. For any demographic category the audit reports on, the selection rate is just: the number of people in that group who were selected (hired, promoted, moved forward) divided by the number of people in that group who were considered by the AEDT. Nothing more.

That's it — a fraction, expressed as a percentage. The selection rate and scoring rate are where every other number in the audit summary originates. If the auditor's underlying selection-rate table is wrong, aggregated oddly, or missing a category, everything downstream — the impact ratio, the pass/fail read, your public summary — inherits that problem. This is also why an operations-side workbook that tracks vendor audit status and documents your rationale for reading a summary a certain way is not a substitute for the audit itself: the workbook helps you follow the arithmetic and keep a record of what you checked, not perform the calculation for the vendor.

Turning Selection Rates Into an Impact Ratio

The bias audit selection rate impact ratio calculation itself is one division. Take the selection rate for the group being compared, and divide it by the selection rate of whichever group had the highest selection rate in that category. The group with the highest rate always becomes the denominator — that's the reference point every other group is measured against.

Impact ratio = (selection rate for group X) ÷ (selection rate for the highest-selection-rate group)

That single fraction is the number that gets flagged, discussed, and — under the four-fifths framework — compared against a threshold. It is worth saying plainly here that none of this is legal advice, and this article isn't performing or certifying anything: it's a plain description of arithmetic that an independent auditor runs, so you can follow their published output. If you need a determination about your own AEDT's results, that's a conversation for your auditor or outside counsel, not a blog post.

A Worked Example: Walking Through the Four-Fifths Rule

Numbers make this concrete faster than definitions do, so here's an illustrative example with round, made-up inputs — not a real audit's data, just a way to see the mechanism work.

Say a resume-screening AEDT considered 100 applicants in Group A, and 50 were advanced. That's a 50% selection rate. In Group B, 80 applicants were considered, and 32 advanced — a 40% selection rate.

Group A has the higher rate, so it becomes the reference: 40% ÷ 50% = 0.80.

Under the four-fifths rule — a threshold drawn from the EEOC's Uniform Guidelines on Employee Selection Procedures — a selection rate for any group below 80% of the highest group's rate can indicate adverse impact worth investigating (via Assessment Systems, 2024). This example lands exactly at 0.80, which is precisely why auditors and regulators treat it as a threshold to watch, not a hard line that guarantees anything either way. A 0.79 is a flag. A 0.81 isn't automatically clean. The impact ratio in a Local Law 144 bias audit works the same way — it's a signal that invites a closer look, not a verdict on its own.

Where the Math Gets Murky

The arithmetic above is clean. What auditors actually work with often isn't. A few realities worth knowing before you take any published impact ratio at face value:

Small sample sizes distort selection rates badly. If only six people in a category were considered, one hire moves the selection rate by almost 17 percentage points, and the resulting impact ratio can swing wildly without telling you much about the tool's actual behavior.

Category aggregation choices matter more than most published summaries let on. A 2025 academic review, "Auditing the Audits," found that many published Local Law 144 audits may under-report disparities because of missing demographic data, opaque aggregation of categories, and metrics that don't reflect how the AEDT is actually deployed day to day (ACM FAccT, 2025). Two auditors working from the same underlying hiring data could publish different impact ratios depending on how they grouped intersectional categories.

And transparency itself has been thin. A separate 2024 study of 391 employers using AEDTs found that only 18 had posted an audit report and only 13 had posted the required transparency notice (ACM FAccT, Wright & Muenster et al., 2024). If you can't find your own vendor's published summary at all, that's a more urgent problem than any single ratio in it.

None of this means the four-fifths math is decorative. It means the number is only as trustworthy as the category definitions and sample sizes feeding it — which is exactly what you should be checking, not recalculating from scratch.

Reading This in a Published LL144 Summary

When you pull up a bias audit summary — your vendor's, or one you're vetting before signing a new AEDT contract — walk it in this order: find the selection rate table first, confirm each category's sample size is large enough to trust, identify which group is the reference (highest rate), then check the published impact ratio against your own division. If the numbers don't match, that's worth a question to the vendor before it's worth a panic.

Our guide on how to read a bias audit summary walks through the rest of the document — the audit date, the AEDT distribution date, and the parts of a Local Law 144 disclosure that aren't about ratios at all. And if you're still mapping out the full annual cycle — the audit itself, the public posting requirement, and the candidate notice that has to go out at least 10 business days before an AEDT is used (Crowell & Moring LLP, 2023; Epstein Becker Green, 2023) — the NYC Local Law 144 compliance guide lays out where this calculation fits into that larger obligation. Civil penalties for missing pieces of that cycle run up to $500 for a first violation and $500 to $1,500 for each subsequent one, with violations accruing per day (Office of the NY State Comptroller, 2025) — a good reason not to let an unreadable audit summary sit in a folder unchecked.

Your First Move: Build the Reading Habit Before You Need It

You will not calculate your vendor's impact ratios. That's the auditor's job, and it should stay that way — Local Law 144 exists specifically because an independent third party is supposed to do this work, not the employer and not the AEDT vendor. But you should be able to open a published summary, find the selection rate table, and check the division yourself, every year, before you post the summary publicly and before a candidate or a regulator asks you to explain it.

The Four-Fifths Rule & Impact-Ratio Reading Guide is built for exactly that habit: a short PDF walkthrough plus a companion calculator worksheet where you can drop in a vendor's published selection rates and see the impact ratio come out the other side, so you're checking the math instead of trusting the summary blind. Download the reading guide and keep it next to whatever audit summary lands in your inbox next.