Diversity Recruiting Metrics: What to Track and How to Calculate Them

Diversity recruiting metrics count who applies and who advances at each hiring stage, grouped by sex and by race or ethnicity. They show which stage loses candidates from a group, so you know where to look more closely.

Say you're filling several backend software engineer roles. The hiring manager asks how the pool breaks down by sex and by race or ethnicity. You need a few numbers you can calculate, a fair way to compare groups, and clear limits on who sees the data. Those limits matter because federal guidance treats counting outcomes across a group very differently from using demographic data on one candidate.

Which diversity recruiting metrics to track at each stage

The core diversity recruiting metrics are the number of candidates from each group at each stage and the share who move forward. A pass-through rate divides the number who advance by the number who entered the stage, so groups of different sizes compare fairly. If you already report a recruiting funnel, these diversity hiring metrics use the same stages with one added breakdown.

The table shows what to count at each stage, the formula, and what a gap between groups may point to.

In the backend engineer search, applicant shares might look reasonable while the technical assessment pass rate differs between groups. That points you to one stage instead of the whole process, which is why you calculate rates stage by stage. If you recruit for several clients, a separate view per client keeps one employer's numbers from blending into another's.

Collect demographic data separately from the application

Data for diversity recruiting metrics should come from a self-identification form kept separate from the application. That way the people screening candidates never see it. The EEOC's guidance on race says it's generally assumed that pre-employment questions will "form the basis for hiring decisions" about applicants. Employers that track applicant flow can keep race information separate, for example on a tear-off sheet.

The same guidance says asking about race by phone could almost never be justified. So a screening call isn't the place to fill gaps in the data. Avoid guessing race or sex from names or photos as well. A guessed count is unreliable, and it suggests demographics shaped your view of the person.

Some candidates won't answer the form. Report how many declined alongside the group counts, so the hiring manager knows how complete the data is.

How to run the four-fifths rule on a hiring stage

The four-fifths rule compares each group's selection rate at a stage with the rate of the group that did best. Under the Uniform Guidelines, federal enforcement agencies will generally treat a rate below 80% of the top rate as evidence of adverse impact. It's a general screen the agencies apply, not a finding of discrimination. An adverse impact analysis built on it tells you where to look, not what caused the gap.

Take a hypothetical assessment stage in the backend search, using round, made-up numbers:

• Group A: 200 candidates took the test and 100 passed, a selection rate of 50 percent.
• Group B: 100 candidates took the test and 35 passed, a selection rate of 35 percent.
• Ratio: 35 divided by 50 gives 70 percent of the top rate, which falls below the four-fifths line.

The guidelines would generally treat that result as a sign of adverse impact at this stage. The next step is to review the test, and the guide to pre-employment assessments covers how to check that it measures what the job needs.

Small numbers change how you read the result. The same regulation says a larger gap may not count when it rests on small numbers and isn't statistically significant. A smaller gap can still count when it's significant in both statistical and practical terms. With 10 candidates in a group, one person passing or failing can move the ratio across the line.

A low ratio can also have an innocent cause, such as one channel sending many early-career applicants to a senior-level test. The regulation says that if the whole process shows adverse impact, each of its components "should be evaluated" too. It also expects employers to keep records of impact by race, sex, or ethnic group. Tracking across several searches gives a steadier picture.

Check your sourcing channels before you blame a later stage

When the gap starts at the application stage, the cause may sit in where you search and how the job post reads. Later stages can only work with the pool you send them. Breaking applicants down by channel shows whether one referral network or job board supplies most of the pool. Your sourcing metrics may already hold that view.

A labor-market benchmark gives your diversity recruiting metrics rough context. BLS survey data for 2025 shows 20.3% of employed software developers were women, 5.4% were Black or African American, and 6.1% were Hispanic or Latino. Those figures describe people already working in the occupation nationally, at every experience level, not applicants. They're 11-month averages that exclude October, and Hispanic or Latino people can be of any race, so the groups overlap.

Treat the benchmark as a loose reference for the backend search, not a target. If your applicant sits well below it, widen the channels you use and look for requirements the role doesn't need. A clearer job description can make the role easier to judge for qualified people outside your usual profile.

Keep diversity data out of individual hiring decisions

Diversity recruiting metrics describe a process in aggregate, so they shouldn't be used to choose, rank, or reject one candidate. EEOC technical assistance issued in 2025 says DEI-related actions can be unlawful when race, sex, or another protected trait motivates them, in whole or in part. The covered actions include hiring, internships, and selection for interviews. They also include placing candidates on a slate or leaving them off one.

The same guidance flags "diverse slate" policies, since they can require employers to collect race information before hiring. If that information is used in selection with groups excluded, the guidance warns that those inquiries can be evidence of discrimination. This is agency guidance, not a statute or court ruling. Still, it applies directly when a hiring manager or client asks you to meet a slate requirement, and the Rooney Rule guide explains how those policies work.

Recommended practice limits who sees what:

• Separate the roles: Where possible, someone outside the hiring decision, such as an HR analyst, pulls the report.
• Share group totals only: The hiring manager gets counts and rates by group, never a candidate list with demographic labels.
• Combine small groups over time: When a group has only a few candidates, pool several searches so the report can't point to one person.
• Fix the stage for everyone: If interviews show a gap, a structured interview scorecard applies one standard to every candidate.

Check the guardrails before you share the numbers

Stage-by-stage rates show where candidates from a group drop out, and the four-fifths check shows which drops deserve a closer look. Neither explains the cause, and both get shaky when the counts are small. How you collect and share the data matters too, since numbers that help fix a stage create legal risk once they touch one person's decision. Before you send the hiring manager anything, confirm the data came from a separate form and the report shows only group totals.