The short answer
An impact ratio is a group's selection rate divided by the selection rate of the most selected group, or a group's scoring rate divided by the scoring rate of the highest scoring group. That is the definition in New York City's rules for Local Law 144 (6 RCNY 5-300). A ratio of 1.00 means parity with the top group; the lower the number, the bigger the gap, which is why every impact ratio bias audit report leads with them.
Impact ratio explained
There are two versions, depending on what the tool outputs:
- Tools that select or classify (pass/fail, advance/reject): impact ratio = selection rate for a group / selection rate of the most selected group.
- Tools that score (a number for each candidate): impact ratio = scoring rate for a group / scoring rate of the highest scoring group. The scoring rate is the share of a group that scored above the median score of the whole sample.
The top group always lands at exactly 1.00, and no group can go above it, because every rate is compared with the highest one. A NYC bias audit has to report these ratios separately for sex, for race/ethnicity and for intersectional categories (for example, Hispanic or Latino women).
Example
The city's rules include worked examples. Here are two rows from each:
In the scoring case, 44.7 / 54.3 = 0.82. In plain terms, women in that sample scored above the median about four-fifths as often as men.
How to read impact ratios
- Go to the lowest number. Sex-level ratios can look fine while an intersectional row sits much lower.
- Check the group size. A ratio built on a handful of people moves a lot with one decision. NYC lets an auditor leave out a category under 2% of the data, but the summary must give the reason, the count and the rate.
- Look for the unknown count. The audit must say how many people were assessed but left out because their category was unknown.
- Know there is no NYC pass mark. Readers often compare ratios with 0.80, the federal four-fifths rule, but Local Law 144 itself sets no threshold.
Our guide to reading AI bias audits has a fuller sample table and the questions to ask a vendor.
Related terms
- Selection rate: the numerator and denominator behind most impact ratios.
- Four-fifths rule: the 0.80 rule of thumb many readers apply.
- Adverse impact: what a low ratio may indicate.
- Bias audit: the report where impact ratios are published.
- Automated employment decision tool: the kind of tool NYC requires these ratios for.
Questions
What is a good impact ratio?
Closer to 1.00 is better. Many employers and auditors treat a ratio below 0.80 as a flag because of the federal four-fifths rule, but New York City's Local Law 144 sets no pass mark, and small samples can make any single ratio unreliable.
Can an impact ratio be higher than 1?
Not under the NYC definition. Each group is divided by the most selected or highest scoring group, so that group is exactly 1.00 and every other group is at or below it.
What is the difference between a selection rate and a scoring rate?
A selection rate is the share of a group a tool moves forward or classifies. A scoring rate, used for tools that output scores, is the share of a group that scores above the median score of the full sample.
See how a live adaptive AI interview produces the scores an auditor would test: book a 20-minute demo, or try a sample interview.
This is not legal advice.
