Does Board and Management Diversity Improve Financial Performance? The Evidence Is Weak

COMMENTARY Markets and Finance

Does Board and Management Diversity Improve Financial Performance? The Evidence Is Weak

Nov 18, 2026 7 min read
COMMENTARY BY
Allen Mendenhall, PhD

Research Fellow & Senior Advisor, American Investor Initiative

Allen Mendenhall is a Research Fellow for the Institute for Economic Policy Studies, and Senior Advisor of the American Investor Initiative.
The analysis may document correlations inside a selected dataset, but it does not establish that management diversity improves corporate performance. Kenishirotie / Getty Images

Key Takeaways

A new report, Capturing the Diversity Benefit, claims that companies with more racially diverse management teams perform better financially.

At a 5% significance threshold, false positives become not merely possible but inevitable when researchers search across that many combinations.

Corporate leaders should demand better evidence before accepting a financial claim simply because it supports a preferred social or political conclusion.

In 1954, the economist Darrell Huff published How to Lie with Statistics, a slim volume that became, improbably, one of the best-selling statistics books in American publishing history. Huff’s central insight—that correlation dressed in the vestments of causation can seduce even careful readers—has aged better than most mid-century paperbacks. It’s due for a revival.

Especially in the wake of a new report, Capturing the Diversity Benefit, which claims that companies with more racially diverse management teams perform better financially.

The report is not without ambition. It draws on federal EEO-1 workforce filings matched to financial data for 1,482 publicly traded companies between 2016 and 2024, and it contains some genuine quantitative analysis.

But its empirical research design is deeply flawed, and in ways Huff would have recognized instantly. The analysis may document correlations inside a selected dataset, but it does not establish that management diversity improves corporate performance, that DEI programs create shareholder value, or that companies incur financial risk by scaling those programs back.

Consider, first, the sample, always the place a skeptical reader should look, and always the place such reports hope one will not. The number of observed companies changes substantially by year, reaching 1,059 workforce observations in 2020 and falling to 523 in 2024. Earlier data combine voluntary disclosures with records obtained through a Department of Labor Freedom of Information Act release. Later data rely heavily on voluntary disclosure alone, a shift roughly analogous to switching, mid-poll, from a mandatory census to a call for volunteers.

>>> DEI’s Days in the Corporate World Are Numbered

The report itself concedes the consequence: companies choosing to disclose had 4.5 percentage points more BIPOC management representation than companies that never disclosed. Selection into the sample is therefore entangled with the very variable under study, which is a polite way of saying the study measures, in part, itself.

Nor does the report tell readers how its companies compare with the Russell 3000 or the CRSP-Compustat universe, the benchmarks against which any claim to represent “corporate America” must be tested. Readers do not learn whether the sample differs in size, industry, age, profitability, exchange listing, geographic exposure, or prior stock performance.

These are not incidental details; they are precisely the characteristics that predict both workforce composition and financial outcomes. Pooling a large technology company with a small regional manufacturer and regressing profits on a single demographic measure does not make those differences disappear. It merely makes them invisible.

The changing sample, moreover, weakens whatever trends the report reports. An average can move for two quite different reasons: because the same firms changed, or because different firms wandered into and out of the dataset. To its credit, the report acknowledges that sample composition may explain part of the sharp decline in female representation it reports for financial services between 2020 and 2021.

That caveat, once granted, applies throughout the time series, a fact the report is less eager to grant. A balanced panel and a decomposition of within-firm change from sample entry and exit are prerequisites, not refinements, for interpreting these averages as corporate trends rather than statistical artifacts of who happened to show up.

Measurement supplies a second problem, and a subtler one. EEO-1 filings are more standardized than ordinary ESG ratings, which is a modest compliment, but they still require decisions about establishments, corporate ownership, mergers, missing filings, and the aggregation of executive, senior, first-line, and mid-level managers. The report provides too little information to reproduce those choices. Its Racial Justice Initiative score is even less transparent. It combines 27 indicators—ranging from promotion data and EEO-1 disclosure to public statements, donations, and references to political movements—with some categories assigned five times the weight of others. One is invited to trust the recipe without seeing the kitchen.

The ESG measurement literature explains, with some precision, why these choices matter. Florian Berg, Julian Kölbel, and Roberto Rigobon compared six major ESG rating systems and found correlations ranging from 0.38 to 0.71, hardly the tight clustering one would expect if these systems were measuring the same underlying reality. They attribute 56% of rating divergence to measurement, 38% to scope, and 6% to weighting.

Any claimed DEI-performance relationship therefore raises a prior question, one the report never quite gets around to asking: What DEI score is being used, which indicators enter it, and how are they weighted? A composite social score cannot be treated as an objective corporate trait without its coding rules, missing-data procedures, reliability checks, and sensitivity to alternative weights. Put simply, DEI and ESG scores vary so markedly across providers that they have largely lost their meaning as fixed facts about a company.

Related research by Christos Makridis and Majeed Simaan demonstrates that this measurement noise is not merely an academic nuisance; it has real consequences for actual investment decisions. Using more than 22,000 firm-year observations and 21 years of out-of-sample portfolio tests, they find that portfolios incorporating Refinitiv ESG scores do not consistently outperform return-based strategies on a risk-adjusted basis. ESG-only rules can produce higher average returns in some specifications, true enough, but they also generate greater volatility and downside risk (the fine print investors are seldom shown).

Return-based factor models achieve comparable or better ex-post ESG standing without using ESG ratings directly at all. Their panel analysis also finds that market volatility predicts later ESG ratings, while ESG ratings do not predict later volatility, suggesting the ratings absorb information the market already knew rather than furnishing anything so useful as foresight.

Capturing the Diversity Benefit then commits its largest error, and here Huff would surely have reached for his red pen: it estimates simple bivariate regressions. Financial performance depends on firm size, book-to-market ratios, profitability, investment, leverage, industry, firm age, prior performance, and macroeconomic conditions. The main analysis controls for none of them. This omission is especially damaging because the report’s own tables show sharply different results across market-capitalization groups. The authors call firm size “almost certainly” a confound—then proceed, undeterred, to present unadjusted aggregate estimates anyway.

Income after tax, measured in dollars, illustrates the difficulty. Large companies generally earn more dollars and may also employ more diverse managers. A positive coefficient can therefore capture company scale rather than operating performance—correlation masquerading, once again, as insight.

Raw multiyear stock-price appreciation is similarly inadequate to the task. Finance researchers evaluate abnormal returns only after accounting for established sources of risk and return. The Fama-French five-factor model incorporates market exposure, size, value, profitability, and investment; cross-sectional work commonly employs the Fama-MacBeth method to test whether a given characteristic predicts returns after competing explanations have had their say. The report employs none of this.

Timing presents yet another serious obstacle. Matching workforce composition in a given year with performance measured in that same year does not reveal which came first. Strong companies may simply have more resources to recruit broadly, invest in management systems, and publish workforce data in the first place. Both diversity and performance may reflect nothing more exotic than management quality itself. Showing that a correlation existed before 2020 does not resolve the question of reverse causality. It merely pushes it back a few years. At a minimum, the analysis should use lagged representation to predict future outcomes, control for prior performance, and account for the swath of unobserved differences across firms over time.

The report also runs hundreds of statistical tests across demographic groups—approximately 15 outcomes, 11 sectors, three size categories, and nine years. At a 5% significance threshold, false positives become not merely possible but inevitable when researchers search across that many combinations. It would be more surprising if nothing turned up significant by chance alone. Yet the tables display only plus signs, minus signs, and blanks. They omit coefficients, standard errors, confidence intervals, and economic magnitudes—the very information a reader would need to judge whether “significant” means “large” or merely “detectable.” The analysis should identify a small set of primary outcomes and adjust for multiple testing accordingly.

Prior research illustrates why this quantitative discipline matters so greatly. Jeremiah Green and John Hand attempted to reproduce McKinsey’s influential diversity-performance findings using observable S&P 500 data and found no statistically reliable relationship between executive racial and ethnic diversity and industry-adjusted profitability, sales growth, return on assets, return on equity, or total shareholder return. Their critique also identifies reverse causality in relating later diversity measures to earlier performance—the chicken-and-egg problem dressed in academic clothing.

>>> The Myth of “Voluntary” ESG

Alex Edmans, Caroline Flammer, and Simon Glossner draw an equally important conceptual distinction. Their employee-survey measure of DEI has little correlation with demographic representation as such. It is associated with stronger future accounting performance, earnings surprises, and valuation ratios while demographic diversity, on its own, is not. The DEI measure also does not predict future stock returns. Representation, inclusion, and corporate programs are, in short, different constructs measuring different things, and the report moves among them with a freedom the evidence does not license.

Which raises the damning question: if DEI scores do not reliably reflect a company’s actual demographic composition, doesn’t the entire argument collapse under its own premise?

None of this forecloses the possibility that a serious analysis might yet find that management diversity contains financially relevant information. The current report has simply not demonstrated it. The authors should compare their sample with the public-company universe, publish their data construction and coding rules, use a balanced panel, estimate models with firm and year fixed effects, control for standard company characteristics, evaluate factor-adjusted returns, lag the workforce variables, and correct for multiple testing. They should also report effect sizes rather than colored symbols, which are decorative rather than dispositive.

Until then, the defensible conclusion is a modest one: among the company-years observed in this selectively disclosed dataset, some measures of management representation correlate with some financial outcomes. That could well be the case. But it does not establish a “diversity benefit,” and the distance between those two claims is the whole of the matter. Corporate leaders should demand better evidence before accepting a financial claim simply because it supports a preferred social or political conclusion. The same standard, it hardly needs saying, should apply when the conclusion cuts in the opposite direction. Huff, one suspects, would have appreciated the symmetry.

Heritage Offers

Activate Your 2026 Membership

Activate Your 2026 Membership

By activating your membership you'll become part of a committed group of fellow patriots who stand for America's Founding principles.

The Heritage Guide to the Constitution, 3rd Edition

The Heritage Guide to the Constitution, 3rd Edition

Receive a clause-by-clause analysis of the Constitution with input from more than 100 scholars and legal experts.

American Founders

American Founders

In this FREE, extensive eBook, you will learn about how our Founders used intellect, prudence, and courage to create the greatest nation in the world.