Rohini Pande in Science Magazine: "On the Economic Costs of Ending DEI"

This article first appeared in Science Magazine on May 15, 2025

In January 2025, Donald Trump signed executive orders 14151 and 14173, which were intended to eliminate all diversity, equity and inclusion (DEI) activities in the Federal bureaucracy and non-discrimination requirements for Federal contractors and subcontractors, respectively. These executive orders follow the Supreme Court decision in Students for Fair Admissions v. Harvard, in which the conservative majority concluded that race-based affirmative action in academia violates the Fourteenth Amendment’s Equal Protection Clause. 

As I set out in my last Science article, there is substantial evidence that even seemingly meritocratic recruitment into high-quality education and demanding jobs does not effectively allocate talent. Economies and countries are most productive when talent is optimally allocated; that is, when the person who would be most productive in a job gets that job. In practice, however, structural frictions such as social norms, negative gender and racial biases and discrimination, an over-reliance on limited networks, and economic inequality that prevent talented individuals from investing in skill development all contribute to substantial talent misallocation in most economies. 

Meritocracies that do not intentionally correct for the effects of structural frictions are not effective meritocracies and produce suboptimal economic outcomes. DEI initiatives seek to compensate for these structural frictions in hiring and in promotion. In the United States, these initiatives are undertaken in compliance with Title VII of the Civil Rights Act,, which focuses on discrimination in employment, and with Title VI, which focuses on discrimination in Federal contracting and funding, and may include actions such as outreach to under-represented groups, ‘positive discrimination’ such as employment or contracting quotas, or ‘diversity training’ for employees  The Supreme Court’s decision in Steelworkers vs Weber in 1979 confirmed that affirmative action designed to correct existing structural inequalities in the labor market is legal under the Act, and does not itself constitute discrimination against those who do not benefit from the affirmative action. 

The Trump administration nevertheless asserts that these initiatives are inherently discriminatory, anti-meritocratic, and illegal under the Civil Rights Act. This runs contrary to  extensive evidence for the continued existence of structural frictions that disadvantage numerous groups in the labor market. The administration’s position may eventually be challenged in court.  Nevertheless, it has already had a chilling effect. Following the issuance of Trump’s executive orders, several  American corporations, including IBM, Warner Brothers, State Street, Victoria’s Secret, Paramount, Bank of America, and Citigroup, reportedly decided to discontinue some or all of their efforts to increase diversity in hiring and/or promotion. It is unclear in each case whether these corporations made this decision to increase or maintain their attractiveness as potential contractors to the new administration, to avoid anticipated legal attacks by the administration, or because their management believes DEI is ineffective, counter-productive, and unnecessary in the current political climate. Research shows that hiring practices in several Fortune 500 companies discriminate against minorities. What is then almost certain, however, is that the withdrawal from DEI in its entirety will harm productivity and economic growth in the United States, as well as impede US companies’ ability to compete globally. 

Following the passing of the Civil Rights Act, we have seen the positive economic effects of initiatives to reduce structural frictions in the United States’ economic progress during the past few decades. According to Hsieh et al., better job allocation accounted for up to 40% of the increase in GDP per person between 1960 and 2010. This is primarily driven by a reduction in institutional barriers to women's and Black Americans' advancement in the labor market. For example, as the authors note, in 1960, 94% of doctors and lawyers were white men. By 2010, the proportion was 60%. 

Preliminary work from Chiplunkar and Kleineberg, using a similar methodology across 91 countries, shows that overall reduced structural barriers to female labor force participation across these countries have resulted in higher female employment and have played a substantial  role in recent economic growth in these countries. They do, however, highlight substantial variation across countries. This variation in structural frictions, and subsequent economic impacts, can also be observed within countries. In their preliminary study, Goldberg et al. use a similar methodology to Hsieh at al. to show that the least progressive Indian states could raise their productivity by 15% if they reduced structural frictions to the same level as those in the most progressive states. 

In richer countries, even after decades of progress, inefficient talent allocation persists.  Hsieh et al. estimated that, in 2010, improved talent allocation might still result in a 10% rise in GDP per person in the United States. This finding indicates that structural inequalities in the US labor market are still very much present, and so that, contrary to the administration’s claims, barring a new Supreme Court decision that overturns precedent, affirmative action aligned with the Equal Employment Opportunities Commission guidelines is likely still legal.

A particular concern, even in these richer countries, is the relative lack of female entrepreneurs. For example, even in Denmark, a country with relatively high levels of gender equality, just 25% of entrepreneurs are women. And research finds that investors prefer pitches made by men to identical pitches made by women. This is important economically not only because talent allocation into entrepreneurial activity is likely to be suboptimal, but also because female entrepreneurs are generally better at giving jobs to women who deserve them. Initial findings from Goldberg and Chiplunkar, for example, find that removing barriers to female entrepreneurship in India would result in higher female labor force employment, and overall economic gains, as more productive female-led firms replaced less productive male-led firms. 

Given this evidence, it’s likely that Trump’s executive orders  - which will likely, in practice, worsen the functioning of initiatives such as the Women-Owned Small Business program and Economically Disadvantaged Women-Owned Small Business program that attempt to compensate for structural barriers to female-led (or minority-led) firms in Federal contracting -  will have the opposite effect. 

We know that innovation is a driver of growth. Women and minorities are substantially under-represented in commercial patenting - indicating a likely misallocation of talent, and a brake on innovation. If the administration - as seems likely - decides to end programs designed to help correct this misallocation -  such as the WAMS (Woman and Minorities in STEM) program at USDA or the NIH F-31 diversity grants for PhD. students from marginalized backgrounds, this brake on innovation will likely continue to be a brake on growth.

Another sector of concern, particularly for the future, is machine learning (‘ML’) related jobs. Researchers have found that women are substantially under-represented in the ML industry and preliminary work suggests they use 'generative AI' far less than men, even when practical access is equalized. So far, there is no good evidence as to why this disparity has arisen. One hypothesis is that Large Language Models constructed on decades of data with substantial implicit bias end up being less useful to women than men. Another issue is that using ML and participating in the ML industry have so far appeared to be relatively risky activities in some ways; early versions of LLMs, in particular, were likely to produce apparently convincing but inaccurate answers to queries; corporations developing ML have tended, in practice, to ignore substantial safety concerns. A reluctance to take risks is reasonable in the business environment where women are often more harshly treated than men.  

Given that there is no evidence that women are inherently less capable at using or implementing ML than men, it is likely that there is currently substantial talent misallocation in this field, and countries and companies that fail to understand and correct this misallocation will find themselves at a competitive disadvantage as ML technology is used in an increasing number of fields. 

So how should US companies that genuinely want to remain competitive, US universities that want to recruit the most talented students, and US policy-makers who want their country to maintain its economic progress, react to a US administration that talks about merit but appears to be recreating the economic inefficiencies of the 1960s?

First, they should carefully examine the empirical evidence around each element of DEI programs. It’s likely that not everything labeled as ‘DEI’ improves talent allocation - and some elements may exist to give the illusion of action while preserving ‘old boy’ networks. There is little, if any, evidence, for example, that so-called ‘diversity training’ actually contributes to creating a more diverse workplace.

Second, they should take the administration at its word and implement evidence-based recruitment practices that not only give the appearance of recruiting on merit, but genuinely, provably, recruit on merit - which entails controlling for, and compensating for, the complex and interacting effects of structural frictions such as economic inequality, bias, and gender norms.

Third, they should remember that polls indicate that a majority of Americans still support DEI policies. And that the abrupt firing of well-qualified minority and female senior officers in the military and scientists on the NIH review board suggests the attack on DEI may allow those seen as better aligned with the current administration a larger part of what will end up as a smaller economic pie. This might benefit those individuals - and mimicking the ideological bent of the administration might bring short-term gains for individual companies - but worsening talent allocation long-term certainly won’t benefit shareholders, consumers, or the country at large.

As I suggested in my previous ‘Science’ article, there is certainly room for improvement both in the way that we structure this compensation for structural frictions in the labor market, and in the way that we present the rationale for doing so. For example, by first ranking the achievements of a candidate for a demanding job, or a coveted higher education placement, against a pool of economic and social peers, before using this pool-based ranking to determine their place in an overall ranking for the position, we can effectively distinguish merit (innate talent, and the ability to work hard) from structural advantages that a candidate may have benefited from, in a way that treats every candidate, explicitly, equally and fairly.