What is the Diversity Index?
The diversity index is a single number that measures how varied a population is across a set of categories. Feed it the headcount in each demographic group, and it returns a score that tells you the odds that two randomly chosen people belong to different groups.
The math came out of ecology. Edward Simpson published his version in 1949 to measure species variety in a habitat. Sociologist Peter Blau brought the same formula into organizational research in 1977 to measure heterogeneity across categorical traits like race, gender, and job function. That is why you will see the identical formula called four different names: Simpson's Diversity Index, Blau's Index, the Gini-Simpson index, and the Gibbs-Martin index. They produce the same number. In HR and management research, Blau's Index is the standard.
Most HR teams use a diversity index to summarize representation. A demographic breakdown gives you a dozen percentages. The index compresses all of them into one figure you can put on a slide, track quarterly, and compare across departments or levels.
It sits inside the broader practice of people analytics, alongside representation rates, pay equity ratios, and the four-fifths rule. Each of those answers a narrow question. The diversity index answers a structural one: how evenly is the workforce distributed across the categories that matter to you?
Interest in rigorous, quantitative measures has climbed for a specific reason. Public diversity programs have pulled back under legal and political pressure. The Conference Board found that diversity mentions in S&P 500 filings dropped 68% between 2022 and 2024. Companies disclosing women-in-management data fell from 71% in 2024 to 55% in 2025. As the language softened, the appetite for a neutral, defensible statistic went up. A diversity index is exactly that: a workforce composition metric with an academic pedigree and no advocacy attached.
The Diversity Index Formula
The standard HR formula is Simpson's Diversity Index, also written as Blau's Index:
Diversity Index = 1 − Σ(pᵢ²)
Here is what each piece means and how to work through it.
Step 1: Sort your workforce into categories. Pick one attribute, such as race/ethnicity, gender, age band, or job family. Count the headcount in each group. Call each group's count nᵢ.
Step 2: Find the total. Add every group's headcount to get N, the total number of people.
Step 3: Calculate each group's share. For every category, divide its headcount by the total: pᵢ = nᵢ / N. This is the proportion of the workforce in that group.
Step 4: Square each share. Multiply each pᵢ by itself to get pᵢ².
Step 5: Sum the squares. Add up all the squared proportions: Σ(pᵢ²). This intermediate figure is Simpson's raw index, also called D. On its own it runs backward: a higher D means less variety.
Step 6: Subtract from 1. Diversity Index = 1 − Σ(pᵢ²). Now the score runs the intuitive direction. Higher means more diverse.
A few variations exist. When you only have percentages rather than raw counts, the proportional form above works fine. For very small teams, some analysts use the finite-sample version, D = Σ[nᵢ(nᵢ − 1)] / [N(N − 1)], which corrects for sampling a small population without replacement. For large workforces the two give nearly the same answer. A separate cousin, the Inverse Simpson Index (1 / D), reports diversity as an "effective number of equal groups" instead of a 0-to-1 score. We use the 1 − Σ(pᵢ²) form because it produces a bounded, readable number and matches the academic convention HR research has used since Blau.
Worked Example
Dana runs People at a PE-backed building-products manufacturer with 1,200 employees across four plants. Her operating partner wants one diversity figure for the value-creation deck, and it has to hold up if anyone challenges it.
She pulls race/ethnicity headcount from the HRIS:
- White: 600 employees
- Hispanic/Latino: 300
- Black/African American: 180
- Asian: 84
- Two or more / Other: 36
Total: 1,200.
Now the shares and their squares:
- White: 600 / 1,200 = 0.500, squared = 0.2500
- Hispanic/Latino: 300 / 1,200 = 0.250, squared = 0.0625
- Black/African American: 180 / 1,200 = 0.150, squared = 0.0225
- Asian: 84 / 1,200 = 0.070, squared = 0.0049
- Two or more / Other: 36 / 1,200 = 0.030, squared = 0.0009
Sum of the squares: 0.2500 + 0.0625 + 0.0225 + 0.0049 + 0.0009 = 0.3408.
Diversity Index = 1 − 0.3408 = 0.66.
With five categories, the highest score possible is 1 − 1/5 = 0.80, reached only if all five groups were exactly equal. So 0.66 sits at about 82% of the ceiling. For a single attribute, that reads as a genuinely mixed workforce.
Then Dana does the move that makes the metric worth the effort. She runs the same calculation on her 60-person leadership tier:
- White: 48 (0.80), squared = 0.6400
- Hispanic/Latino: 6 (0.10), squared = 0.0100
- Black/African American: 3 (0.05), squared = 0.0025
- Asian: 3 (0.05), squared = 0.0025
Sum: 0.655. Diversity Index = 1 − 0.655 = 0.34.
The company-wide 0.66 looked healthy. The leadership 0.34 tells a different story. The variety lives on the plant floor and thins out as you climb. That gap is the finding. It points straight at promotion pipelines, succession plans, and where the next hiring focus belongs. A single org-wide number would have buried it.
HRBench Benchmark Data
Across all companies in the HRBench dataset, the national diversity index benchmark looks like this:
A median near 0.49 puts the typical company close to an even split on a two-category attribute, where the maximum is 0.50. Read your own score against this range, but weight your internal trend more heavily. A move from 0.41 to 0.47 over four quarters tells you more than where you land against a national figure.
What Data Do You Need to Calculate the Diversity Index?
The input list is short, but the data quality decisions matter more than the arithmetic.
You need a clean headcount by category for one attribute, drawn from a single point in time. Decide your snapshot date first, then pull every active employee as of that date so the denominator is stable.
Your category schema has to be consistent. If race/ethnicity uses seven buckets in one system and four in another, the score is meaningless across them. Map every source to one agreed set of categories before you calculate anything.
Handle "declined to answer" deliberately. You can drop those records, which shrinks N, or treat non-disclosure as its own category, which changes the math. With a high non-response rate, the second choice can swing the score by several points. Pick one approach and apply it the same way every quarter.
Watch the edge cases that distort headcount everywhere: contractors and temps who should usually sit outside the count, recent acquisitions whose demographic fields were captured differently, and rehires double-counted across systems. For very small teams, a single departure can move the index sharply, so use the finite-sample correction or report the raw counts alongside the score.
Why HR Leaders Need to Track the Diversity Index
It gives the board one number instead of a wall of percentages. Operating partners and directors want a metric they can track from quarter to quarter without reading a demographic table. The index delivers that, with a defined scale and a known maximum.
It survives scrutiny. A neutral statistic built on a 1977 academic formula reads as measurement, not advocacy. In a climate where diversity programs face legal review, a rigorous composition metric is the kind of figure that holds up when questioned.
It exposes homogeneity that averages hide. Run the index by level, function, or location, and a healthy company-wide score often splits into a diverse front line and a uniform leadership tier. That split is the actionable insight, and it is invisible without the breakdown.
It connects to workforce planning. Pair the index with promotion rates and average age by level, and you can see whether your pipeline will widen or narrow the gap over the next three years. That feeds directly into workforce planning and succession decisions.
It tracks against outcomes you already report. Trended alongside employee turnover and the engagement index, the diversity index becomes part of the story behind retention, not a standalone vanity figure. Research from McKinsey has repeatedly linked executive-team diversity to financial performance, which is why the number belongs in an EBITDA conversation, not just an HR review.
Benchmarks and Interpretation
There is no universal "good" diversity index, and anyone who quotes one is skipping a step. The score's ceiling depends on how many categories you measure.
For a single attribute, the maximum is 1 − 1/k, where k is the number of categories. Two categories cap at 0.50. Three cap at 0.667. Five cap at 0.80. Ten cap at 0.90. A gender index of 0.48 and an ethnicity index of 0.48 are not equally diverse, because they sit against different ceilings.
That has a direct consequence: do not compare raw scores across attributes with different category counts. Either normalize each score as a percentage of its own maximum, or only compare like with like. Gender against gender. Ethnicity against ethnicity. This category-count bias is well documented in organizational research and is the single most common error in practice.
As rough directional bands for one attribute, a score below 0.30 signals heavy concentration in one group, 0.30 to 0.60 is moderate, and above 0.60 is relatively even, always read against the ceiling. Your own trend line beats any external benchmark. The question that matters is whether the number is moving the direction you want.
Common Mistakes
Comparing scores across attributes with different category counts. A 0.50 gender index is at its maximum. A 0.50 ethnicity index across five groups is well short of 0.80. Treating them as equal misreads both.
Reporting one org-wide number and stopping. The aggregate score is where homogeneous leadership tiers go to hide. Always compute by level and function before you draw a conclusion.
Confusing Simpson's D with the Diversity Index. The raw sum of squared shares (D) runs backward: higher means less variety. The Diversity Index is 1 − D. Reporting one when you mean the other inverts the whole message.
Changing the category schema between periods. If you add or merge buckets, the score is no longer comparable to last quarter. Lock the categories before you start trending.
Ignoring non-disclosure. Dropping versus categorizing "declined to answer" can move the score by several points. Decide once, document it, and stay consistent.
Treating index movement as causation. A rising score does not prove a program worked. It is a structural measure, not an attribution model. Pair it with what actually changed in hiring and promotion.
Calculating on tiny populations without a correction. On a 12-person team, one hire can swing the index 10 points. Use the finite-sample form or show the counts.
Related Metrics
Representation rate: the percentage of a specific group in the workforce; the raw input the index summarizes into one figure.
Four-fifths (adverse impact) rule: flags whether selection rates for one group fall below 80% of the highest group's rate; a compliance lens on the same demographic data.
Pay equity ratio: compares pay across groups doing similar work; pairs with the index to separate representation from fairness.
Span of control: the structure metric that shapes how many leadership seats exist for a pipeline to fill.
Employee retention rate: trended against the index, it shows whether a diverse workforce is also a stable one.
Shannon index: an alternative diversity measure that weights rare groups more heavily, useful for surfacing tokenism the Simpson family underweights.
Inverse Simpson index: reframes the same data as an "effective number of equal groups" for a more tangible read.