The formula fits in one line. The hard part is knowing which rates to compare, which groups to look at and when the result can be trusted. Here are four cases that come up in real reviews of hiring and employment decisions, with the numbers worked out.
The formula
- Rate for each group = people with the favorable outcome ÷ people in the group (for example, selected ÷ applied).
- Reference = the group with the highest rate.
- Impact ratio = each group's rate ÷ the reference rate.
- A ratio below 0.80 (four-fifths) is generally regarded as evidence of adverse impact under the federal Uniform Guidelines (29 CFR 1607.4(D)).
If you need the basics first, start with the four-fifths rule, explained.
Example 1: a tool that scores instead of deciding
Many hiring tools don't reject anyone; they give each candidate a score and a recruiter decides. There is no "selected" column, so you use a scoring rate: the share of each group that scores above the median score of everyone assessed. That is the definition NYC Local Law 144 uses for scoring tools (6 RCNY 5-300).
Illustrative numbers, not a real employer.
48.0 ÷ 52.0 = 0.923. This tool passes the four-fifths benchmark for sex. Now the question is whether it also passes for race, ethnicity and the combinations of both, which is the next example.
Example 2: groups that pass separately and fail together
The same process, checked by sex and by race separately:
Illustrative numbers, not a real employer.
Both ratios are above 0.80. Now the intersectional categories, compared with the highest one, White men:
Illustrative numbers, not a real employer.
Black women are selected at 20.0%, and 20.0 ÷ 32.0 = 0.625: well below 0.80, in a process that looked fine on each characteristic alone. This is why Local Law 144 requires intersectional categories, and why a review that stops at sex and race separately can miss the group that is actually affected.
Example 3: terminations, where the direction matters
For layoffs, terminations or discipline, the outcome you measure is bad news. If you compute the ratio on the share kept, the gap can shrink until it disappears:
Illustrative numbers, not a real employer.
On the share kept, the ratio is 0.926 and passes. On the share let go, Group B was let go at 2.4 times the rate of Group A (12.0% vs 5.0%). Same numbers, very different story. When the outcome is adverse, look at both rates and at the gap in absolute terms, not only at the four-fifths ratio.
Example 4: small groups
Take a reference group selected at 30.0% and a small group of 12 applicants:
Illustrative numbers, not a real employer.
One hire more or less moves the ratio from 0.556 (fails) to 0.833 (passes). The Uniform Guidelines themselves say that with small numbers the four-fifths rule may not be reliable, and that differences can matter, or not, depending on whether they are significant in statistical and practical terms (29 CFR 1607.4(D)). Report the counts next to every ratio, and don't read a ratio for a group of a dozen people as a verdict either way.
Local Law 144 has its own rule for very small categories: the auditor may exclude a category that represents less than 2% of the data from the impact ratio calculations, but must explain why and still report its number of applicants and its rate (6 RCNY 5-301(d)).
Five mistakes that hide a gap
- Comparing with the average instead of the highest group. The Guidelines compare with the group with the highest rate. Comparing with the overall average makes every ratio look better.
- Mixing stages. Applicants in one group and interviewees in another give a ratio that means nothing. Use the same stage of the process for every group.
- Dropping unknowns quietly. If many people have no recorded race or sex, say how many. Local Law 144 requires the number of people in the unknown category to be reported.
- Checking each characteristic alone. As in Example 2, the gap can live in a combination.
- Reading the ratio without the counts. As in Example 4, 0.83 for 12 people and 0.83 for 12,000 people are not the same finding.