On July 21, the EEOC voted 2-1 to publish a proposed rule rescinding the EEO-1 report, along with the EEO-2 through EEO-6 collections and the recordkeeping requirements attached to them. The proposal now sits in a 30-day comment period that closes August 24, with a public hearing scheduled for August 11. The vote does not end anything by itself, and a final rule still has to survive notice-and-comment and almost certainly litigation.
Whatever ultimately happens in the rulemaking process, the larger trend is already underway. The Commission’s direction is not ambiguous, and the only meaningful question left is timing: whether it finishes quickly enough to avoid collecting 2025 workforce data at all.
The public data is already disappearing regardless. The most recent aggregate statistics on the EEOC’s website cover 2023. The agency collected 2024 data but has not released it. For anyone who uses the EEO-1 as a research tool, the practical loss happened quietly sometime last year. The rulemaking is catching up to a reality that already exists.
Where this started
In the summer of 2020, I went looking for workforce data and found the EEO-1 aggregate public-use files. Every covered employer reported using the same ten job categories and the same demographic classifications, going back decades. It was not built for investors, it was not company-level, and it was not remotely user-friendly. It was also the most consistent workforce dataset I had ever seen.
I built two Tableau dashboards (here and here) to make it legible by race and ethnicity and by gender, broken down by industry, geography, and job level. They were crude, and they were not the point.
The point was what the exercise revealed about the broader problem. There was no shortage of workforce data. Government databases, company disclosures, regulatory filings, surveys, academic studies, third-party research, and alternative datasets already existed in abundance. They lived in separate silos, used different methodologies, and answered different questions. Almost nobody was doing the work of pulling them together into normalized datasets that decision-makers could actually use.
That realization became the foundation for DiversIQ, now PeopleReturn. The EEO-1 was never the business itself, but it was the dataset that showed me what the business needed to be.
Why it was the gold standard
The EEO-1 requirement dates to 1966. It covers employers with 100 or more employees and federal contractors with at least 50 employees, representing roughly 73,000 employers and more than 50 million workers in a typical year.
It was never intended to measure corporate performance. Its purpose was to help the federal government enforce equal employment opportunity law. Yet almost by accident, it became one of the most valuable workforce datasets ever created.
Because every employer reported using the same job categories and demographic classifications, it became the only workforce dataset that supported meaningful comparison across companies, industries, geographies, and job levels, with enough history to distinguish a trend from a temporary fluctuation. Researchers used it to study occupational segregation. Journalists used it to report on workforce composition. Companies used it to benchmark themselves against peers. Investors increasingly relied on it to evaluate workforce disclosures and test whether corporate narratives aligned with reality.
Its greatest strength wasn’t what it measured. It was that everyone measured it the same way. Voluntary disclosure never could, because companies choose their own definitions, methodologies, and timing.
The disclosure wave, and the retreat
When I first found the EEO-1 data in 2020, perhaps thirty to thirty-five public companies voluntarily published their reports.
That changed almost overnight. The largest asset managers and proxy advisors began pushing companies to disclose in the second half of 2020, and by the time we incorporated DiversIQ in January 2021, many of the largest public companies were already releasing their reports. Disclosure peaked in mid-2024: as of late July that year, 82% of the S&P 500 and 97% of the S&P 100 had publicly released a recent EEO-1 report, alongside a broader expansion in workforce representation, employee development, pay equity, and board demographic disclosure.
The shift also created an unexpected problem.
Investors were securing disclosure commitments faster than they could monitor them. One company uploaded a standalone PDF to its investor relations site. Another buried the data in a sustainability report. Some issued a press release; others posted the report with no announcement at all. There was no consistent location, format, or timing, and this was true across nearly every category of human capital reporting, not just the EEO-1.
Many of the investors who had successfully pushed companies to disclose suddenly needed a way to know whether those commitments had actually been fulfilled and what had changed from year to year. That became one of our first products, and some of our earliest institutional clients relied on us simply to track disclosures that, despite being public, had become remarkably difficult to find.
Over the past year, we have documented the reversal. Board demographic disclosure has collapsed. Workforce representation disclosure continues to decline. Pay equity reporting has become increasingly rare. Companies that once published EEO-1 reports have quietly stopped.
Now the federal collection itself appears headed in the same direction.
In 2020, company-level disclosure was limited, but the aggregate data remained comprehensive. If this proposal is finalized, both largely disappear. That leaves us in a worse position than the one this company was founded to solve.
The demand did not go away
One thing worth noting about the reaction to the vote: the objections are coming from allocators, not advocacy groups.
Responsible Investor surveyed US asset owners and managers last week and found a consistent position. CalSTRS said it needs consistent and comparable data across its portfolio to assess risk. Parnassus described the EEO-1 as one of the few standardized, third-party-verified workforce data sources it can get. Trillium uses it in voting on director elections, shareholder proposals, and executive compensation. New York City comptroller Mark Levine put the structural problem plainly: without a consistent regulatory requirement, companies will have “no incentive to willingly disclose complete data.”
That is the mechanism. Reporting is mandatory; publishing has always been voluntary. The federal requirement did not force disclosure, but it standardized the artifact, which is what made voluntary publication worth anything to an investor. Remove the standard and what remains is a company’s own definitions on a company’s own timeline.
Our own numbers show it already happening. The figure above measures something different from the mid-2024 snapshot: the share of current index constituents that have publicly released data for each individual reporting year, which is the cleaner way to see the trend because it removes the timing lag between when a company files, when its data is certified, and when it posts.
On that basis, S&P 500 disclosure peaked at 72.6% for reporting year 2023 and fell to 58.0% for 2024. The S&P 100 fell harder, from 87.0% to 60.0%, a 27-point drop in a single reporting year. The pattern holds down-cap: the Russell 1000 went from 44.1% to 35.4%, and the S&P 400 from 16.8% to 14.0%. Note that the 2024 figures may tick up slightly as late postings arrive, and that the near-zero 2025 column is not a further collapse in willingness. The EEOC has never opened the 2025 filing window, so there is nothing for companies to certify and post.
Two years ago this week, we wrote about why that drop was coming.
Through the early 2000s, more than a hundred retailers published monthly same-store sales, including nearly every large US chain. Entirely voluntary, entirely standard. Walmart stopped in May 2009, and the industry reverted to quarterly reporting. The automakers ran the same sequence: monthly sales by make and model were universal until GM dropped the practice in 2018, and the rest followed by 2021.
At the time I wrote that, EEO-1 disclosure looked settled at 97% of the S&P 100. But GM and Nvidia had both quietly pulled their reports, having last published 2021 data, and that is the condition under which these regimes break. The cost of disclosing is relative. It is cheap when everyone does it and expensive when you are conspicuous, so participation holds until a large enough company defects and then unwinds fast. The 2024 column is what that looks like on the way down.
The federal requirement was never what forced companies to publish. It was what made publishing legible: a common form, a common set of job categories, a common filing year. That is the part being removed now, and it removes the floor under an already-unwinding voluntary regime.
Boston Common’s Amy Orr made the point that matters for interpreting all of this: companies may well keep collecting workforce data for internal management and other regulatory requirements. Whether public disclosure stays consistent and comparable is the open question.
Several investors told RI they will respond by engaging companies directly for the same information. That is the actual outcome here. The data does not stop existing. The cost of getting it moves from a standardized federal filing onto individual stewardship teams, one company at a time, with no guarantee the answers are comparable when they arrive.
The part that does not add up
The Commission’s stated rationale rests on burden and legal risk. In the proposal itself, it preliminarily concluded that the reports are inconsistent with equal employment opportunity law and potentially unconstitutional, and that whatever marginal benefits they produce are outweighed by the burdens they impose on employers and on the agency. Both halves are worth checking against the agency’s own numbers.
Start with burden. The EEOC puts the annual cost of the collection at roughly $275 million for employers and about $4 million for itself, and separately estimates that 110,000 private employers file the EEO-1. That works out to no more than $2,500 per employer per year, and in practice less, because the $275 million figure also covers the EEO-2 through EEO-6 filers. For many employers covered by the rule, that works out to roughly the cost of a few hours of outside legal advice.
The agency’s own $4 million share is likewise less than 1% of its $435.4 million FY 2026 appropriation.
Those are the numbers being offered to justify ending a sixty-year dataset, and they do not carry the weight placed on them. If the burden case were the real case, the obvious response would be to simplify the form, raise the filing threshold, or move to biennial collection. Eliminating the collection outright is not a proportionate answer to a $2,500 problem.
Which leaves the legal argument. Chair Andrea Lucas has argued that requiring the reports encourages employers to make race-conscious employment decisions in an effort to improve their numbers.
For decades, this same dataset has been one of the Commission’s primary tools for identifying employers whose workforce patterns may warrant investigation. That is the obvious objection, and the usual response to it is that little really changes: employers will continue collecting demographic information and can still be required to produce it during an investigation.
That is not quite what this proposal says.
It rescinds both the reporting requirement and the associated recordkeeping and record-preservation requirements. Commissioner Kalpana Kotagal made exactly this point in her dissent, arguing that eliminating the reports does not eliminate the legitimate legal and business reasons employers maintain workforce demographic data, from disparate-impact analysis to compliance with state law. Chair Lucas responded by pointing to the Justice Department’s recent Office of Legal Counsel opinion on the Uniform Guidelines on Employee Selection Procedures, and signaled that the Commission intends to revisit those requirements as well.
The assurance is therefore less complete than it initially sounds. The only federal agency charged with investigating employment discrimination is proposing to eliminate one of the primary datasets it has historically used to identify where those investigations are warranted.
What is left
Not everything is disappearing, and the remaining sources are worth understanding.
FOIA requests from journalists and non-profits resulted in the release of roughly 60,000 EEO-1 reports from more than 20,000 federal contractors covering 2016 through 2020. As You Sow sued the Department of Labor in June seeking the release of the 2021 and 2022 reports, after a FOIA request filed in April 2024 produced nothing through two years of appeals. Several states, including Illinois and California, require companies that file federally to submit state reports. Most of those mandates do not make the filings public. Illinois is the exception, and Colorado has enacted its own state-level EEO-1 requirement that survives the federal rescission entirely.
Over the past five years, we have filtered those releases to publicly traded companies and combined them with state filings and company disclosures into a single normalized, entity-linked dataset: 13,281 EEO-1 reports covering 2,168 public companies. Of those reports, 8,698 came from OFCCP releases, 1,193 from Illinois, and 3,390 directly from the companies themselves.
Why the FOIA data matters more than the voluntary data
We supplied the EEO-1 data behind As You Sow’s 2026 Capturing the Diversity Benefit report, published earlier this summer. We were a data source and nothing else. We did not conduct the analysis, and we take no position on its regression findings, which rest on bivariate models without controls for firm size or sector.
Two descriptive results in that report do not depend on the regressions at all, and both are worth sitting with.
The first concerns who discloses. Of 1,090 companies with management-level EEO-1 data for 2016–2020, 637 appeared only in the DOL FOIA release and have never voluntarily published their reports. Those companies averaged 19.3% BIPOC management. The 453 that did publish voluntarily averaged 23.8%, a gap of 4.5 percentage points that held across sectors.
Voluntary disclosure, in other words, is not a smaller version of the full picture. It is a systematically flattering one. The companies that decline to publish are, on average, the companies whose numbers most warrant looking at, and the only reason anyone can demonstrate that is the FOIA release that pulled the non-disclosers into view.
The second is a straightforward count. Usable company-year management observations in that dataset fell from 907 in 2020 to 523 in 2024, with the 2024 figure resting entirely on voluntary disclosure. That is what the loss of a mandatory collection looks like in practice: not an argument about transparency, but a sample that halves.
The ecosystem is fragmenting rather than disappearing. Government collections are shrinking, voluntary disclosure is declining, and what remains comes increasingly from state requirements, FOIA litigation, and the companies that still choose to publish. That does not make workforce analysis impossible. It makes it more difficult and more expensive, and it shifts the cost of assembly from the agency that mandated the data onto everyone downstream.
Five years ago, the EEO-1 taught me that the problem was never a lack of workforce data. It was a lack of structure. Today, the challenge is changing again. Standardized public datasets are becoming scarcer, while the remaining information is scattered across state filings, litigation, company disclosures, and alternative sources. The work is no longer discovering the data—it is preserving it, connecting it, and making it comparable. That was the idea that built PeopleReturn. It is simply a harder problem now.
Rescinding the collection does not reduce workplace discrimination. It removes one of the best tools available for measuring it.
What we’re reading:
- Big Companies Are Starting to Hire Again, Defying Predictions of AI Wipeout For months, major employers treated hiring as an expensive last resort. Now, a shift is emerging across industries. Companies ranging from railroad giant CSX to Google parent Alphabet have told investors in recent days that they plan to hire to meet growth goals or to seize on emerging technologies. (Wall Street Journal)
- Labor and the Corporate Information Environment This survey analyzes literature at the intersection of corporate disclosure and labor economics. While accounting research has long studied how executives and directors influence corporate reporting, recent work has started to explore lower-level employees, e.g., the “rank-and-file,” as both producers and users of corporate information. (Harvard Business School)
- Hybrid work has a $9 million problem. The fix depends on who you ask The everyday friction of hybrid work—from finding an open conference room to managing schedules across time zones—costs the average mid-sized company $9 million annually, according to a new study from workplace management platform Robin. (Fortune)
- Corporate America may be using AI to cut jobs, but small businesses are using it to keep them Last year, most small businesses were using AI to get answers to questions, review contracts, create policies and rewrite emails. Now they’re starting to move into real-life applications that are showing true return on investment. (The Guardian)
- Recruiters Turn to ID Checks as AI-Generated Resumes Surge More than a third of recruiters are now spending half of their workweek filtering “spam and junk” submissions, according to a recent survey by hiring software firm Greenhouse. Now, they’re fighting back. (Bloomberg)
- Diversity, Equity, and Inclusion and Firm Value: Organizational Benefits and Market Implications Diversity, equity, and inclusion (DEI) have become an increasingly important corporate and financial strategy, yet their impact on firm value and stock performance remains complex and context-dependent. This paper reviews empirical research and case evidence to evaluate whether DEI initiatives enhance financial outcomes or introduce new risks. (Journal of Business Diversity)
- The Rise of Million-Dollar Companies With Just One Employee Once upon a time, running a business of a certain size required a team. AI is turning that assumption upside down, and more aspiring entrepreneurs are going it alone. (Wall Street Journal)