1863 Leadership · Issue Paper No. 14
Ninety-Four Percent
In 1960, ninety-four percent of American doctors and lawyers were white men. Undoing that produced as much as two-fifths of the next fifty years' growth. Five repairs, and a market open to everyone.
Abstract
This is an argument about output, not about manners. In 1960, 94 percent of American doctors and lawyers were white men; by 2010, 62 percent. Since innate talent for medicine did not redistribute itself across groups in fifty years, the earlier figure records an enormous pool of people who were not doing the work they were best at. Four economists — two of them at Chicago — measured what correcting that was worth, and found between 20 and 40 percent of the growth in American output per person across those five decades. Discrimination is a misallocation of talent, and misallocation has a price. The argument that markets punish it on their own is Becker's, it is testable, and it has been tested: across 28 field experiments covering 55,842 applications, white applicants receive 36 percent more callbacks than Black applicants, with no measurable change since 1989. This paper proposes that full participation be stated as a condition of a functioning market rather than as a concession to fairness, that the line fall between what a person has done and what a person is, that access to capital and suppliers be treated as seriously as access to a job, and that the expressive exception recognized in 303 Creative be drawn narrowly and honestly by its supporters rather than stretched by them.
Key findings
- In 1960, 94 percent of American doctors and lawyers were white men. By 2010, 62 percent. Between 20 and 40 percent of growth in aggregate market output per person over that period is explained by the improved allocation of talent.1
- Across 28 field experiments — 55,842 applications for 26,326 positions — white applicants receive 36 percent more callbacks than African Americans and 24 percent more than Latinos.2
- There has been no change since 1989. Controlling for education, gender, method, occupation and local labor market conditions does almost nothing to that result.2
- Milton Friedman compared the Fair Employment Practices Commission to "the Hitler Nuremberg laws" — the same in principle. He also, in 1957, moved the American Economic Association out of segregated hotels.3
- 303 Creative protects a refusal to create a bespoke expressive work. It does not grant any business a general right to refuse service to a protected class, and the Court said so. 4
Section 1The question before us
In 1960, ninety-four percent of the doctors and lawyers in the United States were white men. By 2010 the figure was sixty-two. Nothing happened in those fifty years to redistribute the innate capacity for medicine or law across the American population. What changed was who was permitted to pursue it.
Which means the 1960 number is a measurement. It records a very large number of people who were not doing the work they were best at — not because they lacked the aptitude, but because a door was closed. In economics that has a name, and the name is not injustice. It is misallocation.
Every closed door is a person in the wrong job, and an economy is nothing but the sum of who is doing what.
This paper makes an argument about output. It is available to a reader who has no particular feelings about anyone, and it does not require agreement about history, about what is owed, or about anything that happened before the reader was born. It requires only the proposition that an economy does better when people do what they are good at.
1.2 The prediction that was made, and how it turned out
It would be convenient to write that the tradition that produced this argument got it right and was ignored. It did not.
National Review called Brown v. Board of Education an act of judicial usurpation — "shoddy and illegal in analysis, and invalid as sociology." It opposed the Civil Rights Act of 1964 and the Voting Rights Act of 1965. William F. Buckley dismissed the former as a law "artificially deduced from the Commerce Clause" whose marginal effect would be "to instruct small merchants in the Deep South on how they may conduct their business."3
Milton Friedman wrote in Capitalism and Freedom that legislation creating the Fair Employment Practices Commission was the same in principle as the Nuremberg laws. Speaking at Harvard in 1964 he put the objection as a symmetry: "If we pass a law saying that race shall not be a factor in employment, then what grounds do we have for opposing a law that race shall be a factor?"3
The fair version of the record matters as much as the damning one. Friedman detested segregation and acted against it — in 1957 he pressed the American Economic Association to stop meeting in hotels that excluded Black economists, applying his own student's theory that a discriminating business pays for the privilege. Barry Goldwater voted for the Civil Rights Acts of 1957 and 1960 and desegregated the Arizona Air National Guard; his objection in 1964 was to Title II and the reach of the commerce power. Friedrich Hayek supported the Act.3
These were not men who wanted Jim Crow. They ranked freedom of association above access to the market and were wrong about which of the two was load-bearing.
And there is a consequence we should state before somebody states it for us. The Claremont Review of Books — no organ of the left — observed that having opposed every major civil rights law, conservatives cannot reasonably be aghast that others decline to accept their reading of what those laws now require.3 Standing on this question was forfeited, and it will have to be earned rather than asserted.
1.3 Why the objection failed, which is the useful part
Friedman's position was not a preference. It was a prediction, and that makes it testable.
The prediction came from Gary Becker, his student, whose 1957 work argued that an employer who refuses to hire the best candidate pays for that prejudice in wages and output, and that competition should therefore grind discrimination down over time. Friedman believed it, said so at Harvard, and concluded that the law was unnecessary and dangerous.
Sixty years of data are now in, and Section 3 reports them. The erosion did not happen.
That is a better thing to say than that anyone was wicked. A serious man made a falsifiable claim in good faith; the claim was tested at length; it failed. What this paper proposes is not the abandonment of that tradition but the completion of an argument it started and then dropped.
Section 2What the closed door cost
Four economists — Hsieh and Hurst at Chicago Booth, Jones and Klenow at Stanford — asked what American growth owed to the opening of professions previously closed. Their paper appeared in Econometrica in 2019. Its keywords are "economic growth, discrimination, misallocation."
Their method assumes only that talent is distributed the same way across groups. If that is true, then the 1960 occupational distribution is evidence of constraint, and the convergence since is evidence of constraint lifting. Modelling that convergence, they find that between 20 and 40 percent of the growth in aggregate market output per person between 1960 and 2010 is explained by the improved allocation of talent.1
Somewhere between a fifth and two-fifths of half a century of American productivity growth came from letting people do the work they were suited for.
Read that as a cost rather than a benefit and it is the more arresting number. For every year before 1960, the country was operating with a large fraction of its talent pointed at the wrong work — and paying for it in output it never produced, inventions never made, businesses never founded, patients treated by the second best doctor available.
We would add one observation about the word people reach for. Racism and sexism belong to a family of ideas that includes fascism and communism, and the family resemblance is precise rather than rhetorical: each of them assigns a person's place by category rather than by conduct. Each substitutes a group for an individual as the unit of judgment. A market does the opposite — it prices the transaction and does not ask who is standing there — and anything that forces it to ask is not a market operating freely but a market operating under instruction.
Section 3The correction did not arrive
Becker's argument is sound in its own terms and we do not dispute it. An employer who passes over the better candidate is worse off. Over time, competitors who do not should outperform that employer and take the market.
The question is whether it happens. That is not a matter of theory.
Quillian, Pager, Hexel and Midtbøen pooled every available field experiment of hiring discrimination in the United States — 28 studies representing 55,842 applications submitted for 26,326 positions, in which otherwise identical résumés differ only in the signal of the applicant's race.
Since 1989, white applicants receive 36 percent more callbacks than African Americans and 24 percent more than Latinos.2
And there has been no change over twenty-five years. Accounting for applicant education, applicant gender, study method, occupational group and local labor market conditions "does little to alter this result." There is modest evidence of decline for Latino applicants. For Black applicants, none at all.2
A quarter century of controlled experiments, in a country that has become dramatically more competitive and more open in every other respect, and the penalty for the discriminating employer has not appeared.
The literature offers explanations and we do not need to adjudicate among them. Prejudice on the customer's side rewards the discriminating firm rather than punishing it. Search frictions mean the rejected applicant does not simply appear at a competitor's door. Many labor markets are less competitive than the model assumes. What matters for policy is the observation rather than the mechanism: the self-correction Friedman was waiting for has had sixty years and has not occurred.
Section 4Reform I: State the principle as a condition, not a courtesy
Full participation in the market should be stated in law and argued in public as a structural requirement of a functioning market — not as a kindness the market extends, and not as a civil rights claim that markets tolerate.
This is a change in argument rather than in statute, and we think it matters. For sixty years the case against discrimination has been made in the language of dignity and repair, and the case for permitting it has been made in the language of liberty and property. That framing hands the economic argument to the wrong side of the question, and it is backwards.
A market allocates capital and labor to their most productive use. It can only do that if every participant can reach it. A market that a category of people cannot enter is not a freer market — it is a market with a smaller supply of talent, fewer bidders for every job, fewer founders, fewer customers, and a lower ceiling. The Econometrica result is the measurement of exactly that.
So the sentence we want in circulation is not that discrimination is unkind. It is that a market everyone cannot enter is not performing the function that justifies leaving it alone.
Section 5Reform II: The line falls between what a person has done and what a person is
Discrimination in commerce should be permitted on characteristics related to performance of the transaction, and prohibited on characteristics unrelated to it.
Every hiring decision is discrimination. An employer chooses among applicants on skill, experience, credentials, judgment, reliability, professionalism, aptitude, and the thousand small signals that indicate whether a person will do the work well. That is the entire purpose of the exercise and nothing in this paper restricts it.
What should be out of bounds is the set of characteristics that tell you nothing about whether the work will be done: race, colour, sex, sexual orientation, gender identity, religion, national origin, age, disability where it does not bear on the job, and genetic information.
Judge the performance, not the person. It is the same test a market applies to every other input, and the only one it has ever needed.
We choose the relatedness test deliberately over the more intuitive one. People often say the protected characteristics are those a person cannot change, and that is nearly right and fails at the first case: religion is chosen, and is protected, and should be. It is protected because it has nothing to do with whether the accounts balance. That is the criterion, and it also happens to be the standard the law already applies when it asks whether a characteristic is a genuine requirement of the job.
The test also tells you where the boundary genuinely is. A church may require that its minister share its faith. A film may cast an actor of a particular description. A women's shelter may hire women for particular roles. Those are not exceptions to the principle. They are applications of it, because in each case the characteristic is doing real work in the job.
Section 6Reform III: Capital and suppliers, not only hiring
The prohibition should apply with equal force to access to credit, to investment, to supplier and distributor relationships, to commercial leases and to the ordinary infrastructure of doing business — and enforcement effort should be allocated accordingly.
Employment discrimination denies a person a job. Credit discrimination denies a person the ability to create jobs, which compounds over a lifetime and across a family. A person refused a position loses a salary. A person refused a loan loses the business they would have built, the people they would have hired, and whatever their children would have inherited.
If the argument of this paper is about misallocation — and it is — then capital is where misallocation does the most damage, because capital is the input that multiplies. Yet enforcement attention and public argument have concentrated overwhelmingly on hiring.
We note that the evidence base here is thinner than for employment and that this is itself a finding. Paired testing in lending and supplier markets is harder to run and less often funded, which means the area where the stakes are highest is the area we understand least. Reform V addresses that.
Section 7Reform IV: The expressive exception, drawn narrowly by its friends
There is a genuine liberty interest on the other side of this and we would rather define it ourselves than have it defined by the least careful person who invokes it.
In 303 Creative v. Elenis the Supreme Court held that Colorado could not compel a website designer to create custom wedding sites carrying a message she rejected. The reasoning is narrower than the headlines suggested, and the narrowness is the part that matters.
Justice Gorsuch concluded that the designer objected to sending certain messages, not to serving certain customers — and cited as evidence that both parties stipulated she would gladly sell LGBT customers her other products. The Court also said plainly that public accommodations laws are vital to the civil rights of all Americans and that governments have a compelling interest in eliminating discrimination in places of public accommodation.4
The ruling does not grant any business a general right to refuse service to a protected class. It protects the creation of a bespoke expressive work.
That is the line and we adopt it: selling what you already sell, to anyone who comes in, versus being compelled to create a specific work conveying a specific message. Refusing to make a custom piece celebrating something you object to is protected. Refusing to sell the thing on the shelf to the person holding the money is not.
We would add that the exception should be construed against the person claiming it, which is how exceptions to general rules are ordinarily construed. A business that defines everything it makes as expressive has not found a principle. It has found a way around one.
Justice Sotomayor read the same facts differently, warning that the Court had for the first time granted a business open to the public a constitutional right to refuse members of a protected class.4 A reader should know that both readings exist and that the boundary will be litigated for years.
Section 8Reform V: Measure it, continuously and publicly
Paired testing should be funded as routine measurement across employment, lending, housing, and supplier markets, with results published in aggregate.
Everything known with confidence in this area is known because somebody sent out matched résumés and counted the callbacks. That is why the finding in Section 3 exists and why nobody can seriously argue about it. The method produces a direct measure with strong causal validity, which is rare in social science and rarer in this subject.
Two reasons to make it routine. It replaces argument with arithmetic, which benefits whichever side is right. And it would tell us something nobody currently knows: whether the gap is narrowing anywhere, in which sectors, and under what conditions — which is the information any serious policy would be built on and which we do not have.
We propose it as measurement rather than as enforcement. Using a test as the basis for an individual prosecution raises real questions of notice and entrapment that we do not think are resolved. Publishing sector-level results raises none.
Section 9 · The strongest case against
9.1 Friedman's symmetry problem
The objection deserves better than the dismissal it usually receives. If the state may forbid race as a factor in employment, on what principle does it lack the power to require it? The machinery is identical; only the direction differs.
Our answer is that the symmetry is rhetorical rather than real. Every prohibition can be inverted in words — a law requiring contracts be honoured can be described as differing from a law requiring them be broken only in direction. What distinguishes them is that one expands the set of people who can transact and the other contracts it. A rule that opens the market and a rule that closes it are not the same instrument pointed two ways.
We think that answers it. We also think a reader who finds the answer too convenient is entitled to, and that Friedman's version of the worry has been vindicated often enough in other domains to deserve respect.
9.2 Freedom of association is a real liberty, and this restricts it
It does. A person who wishes to trade only with people they choose is prevented from doing so in the commercial sphere, and calling that a market correction does not make it less of a restriction.
We accept that trade openly. The claim is not that no liberty is lost. It is that market access is the more fundamental of the two liberties, because without it the others have nothing to operate on — and that the loss falls on the conduct of a business rather than on anyone's belief, association, speech or worship, all of which remain untouched.
9.3 Disparate impact is where this becomes dangerous
This paper proposes a rule against treating people differently on grounds unrelated to the work. It does not propose that statistical disparity is itself proof of discrimination, and those are very different doctrines with very different consequences for an employer.
A reader who supports the first and fears the second is in a coherent position, and the honest difficulty is that intent is hard to prove and outcome is easy to measure, so enforcement drifts toward the second whatever the statute says. We do not have a mechanism that prevents that drift.
9.4 Compliance produces defensive behaviour
Firms exposed to litigation risk document, standardize and hedge. Some of that is good practice; some of it is expensive theatre, and small firms bear it worst because they cannot amortize a compliance function across ten thousand employees. There is also evidence in adjacent literatures that employers sometimes respond to hiring regulation by avoiding the regulated category altogether, which would be the opposite of the intended effect.
9.5 The expressive exception may swallow the rule
Justice Sotomayor's dissent in 303 Creative warns precisely of this, and the mechanism is easy to see: a business that characterizes its ordinary output as expressive claims the exception. Florists, bakers, photographers, caterers, venue operators and web designers can all make some version of the argument, and the line between selling a cake and designing one is not self-executing.4
Section 7 says the exception should be construed against the party invoking it. A reader may reasonably think that instruction will not survive contact with litigation.
9.6 The evidence on remedies is weaker than the evidence on the problem
Section 3 establishes that discrimination persists. It does not establish that any particular legal remedy reduces it, and the honest state of the literature is that we know the disease far better than the cure. A reader could accept everything in Sections 2 and 3 and still doubt that the instruments proposed here would move the callback ratio.
9.7 And the misallocation finding has a rival explanation
The Econometrica result assumes talent is distributed identically across groups and attributes the occupational convergence to falling barriers. An alternative account attributes more of it to rising human capital — better schools, better health, changed preferences — which would mean the growth contribution reflects investment rather than the removal of discrimination. The authors model both channels; the decomposition is a modelling choice and other specifications produce other numbers, which is why the range is 20 to 40 rather than a single figure.1
9.8 What we concede, and what we do not
We concede that this restricts freedom of association. We concede that Friedman's symmetry objection is serious and that our answer to it will not satisfy everyone. We concede that the line between disparate treatment and disparate impact is unstable in practice. We concede that compliance burdens fall hardest on small firms. We concede that the expressive exception may prove unmanageable. We concede that the evidence for remedies is thinner than the evidence for the problem, and that the misallocation estimate is a range because it rests on modelling choices.
We do not concede that a market which a category of people cannot enter is a free market. It is a market operating under instruction, with a smaller supply of talent, fewer bidders, fewer founders and a lower ceiling — and the country has already run the experiment of finding out what that costs.
Section 10What we are not claiming
We are not claiming that anyone alive is responsible for what was done before them. The argument here is prospective and it concerns output.
We are not claiming that disparity proves discrimination. Section 9.3 says the opposite, and the evidence this paper relies on is experimental precisely because disparity alone cannot settle the question.
We are not claiming markets are hostile to open participation. Becker's mechanism is real and the discriminating employer really is worse off. What we claim is narrower and better supported: that the mechanism has not been strong enough to clear the field in sixty years of trying.
And we are not claiming anyone must approve of anyone. Belief, speech, worship and private association are untouched by anything proposed here. The obligation attaches to a transaction, not to an opinion, and a person may hold whatever view they like about the customer whose money they take.
Section 11The argument you can carry
The paper compressed to what a person can remember and repeat.
And the one that indicts the argument's own side. National Review called Brown shoddy and invalid as sociology. Friedman likened fair employment law to the Nuremberg statutes. Goldwater — who had voted for the 1957 and 1960 Acts and desegregated the Arizona Air National Guard — voted no in 1964. They did not want Jim Crow. They ranked freedom of association above access to the market, and they were wrong about which one was load-bearing.
Section 12Conclusion
There is a version of this argument made in the language of dignity and historical repair. It is not the argument of this paper, and the reason is tactical as well as intellectual: for sixty years that framing has conceded the economics to the other side, as though openness were a cost the market absorbs out of decency.
It is not a cost. It is the mechanism.
A market works by moving resources toward their most productive use. Labour is a resource. A rule that keeps a category of people out of a profession does not make that profession freer — it makes it smaller, less competitive, more expensive, and worse. The country ran that experiment for most of its history and four economists have now put a number on the bill.
Somewhere between a fifth and two-fifths of fifty years of growth was sitting in people who were not allowed to do the work they were best at. That is what the ninety-four percent was: not merely an injustice, though it was that. A permanent reduction in what the country was capable of producing.
That tradition was in a position to see this and did not, because it treated the right to refuse as the liberty at stake and missed the larger one underneath. A person who cannot get into the market has no freedom of contract worth the name — only the theoretical right to make bargains nobody will make with them.
Notes
- Chang-Tai Hsieh, Erik Hurst, Charles I. Jones and Peter J. Klenow, "The Allocation of Talent and U.S. Economic Growth," Econometrica 87(5): 1439–1474 (September 2019), onlinelibrary.wiley.com; working version at web.stanford.edu. Source for the 94 percent figure in 1960, the 62 percent figure in 2010, and the finding that across various specifications between 20 and 40 percent of growth in aggregate market output per person is explained by the improved allocation of talent. Hsieh and Hurst are at the University of Chicago Booth School of Business; Jones and Klenow at Stanford. Earlier versions of the paper reported a narrower range of 15 to 20 percent using a different specification, which is the source of the variation noted in Section 9.7; any published version should state which specification it is quoting.
- Lincoln Quillian, Devah Pager, Ole Hexel and Arnfinn H. Midtbøen, "Meta-analysis of field experiments shows no change in racial discrimination in hiring over time," Proceedings of the National Academy of Sciences 114(41): 10870–10875 (2017), pnas.org. Source for the 28 studies, 55,842 applications, 26,326 positions, the 36 percent and 24 percent callback differentials since 1989, the absence of change for African American applicants, the modest evidence of decline for Latino applicants, and the finding that covariates do little to alter the result. The foundational single study is Marianne Bertrand and Sendhil Mullainathan, "Are Emily and Greg More Employable Than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination," American Economic Review 94: 991–1013 (2004). The underlying theoretical claim this evidence tests is Gary S. Becker, The Economics of Discrimination (University of Chicago Press, 1957).
- Milton Friedman, Capitalism and Freedom (University of Chicago Press, 1962), chapter 7, "Capitalism and Discrimination," for the comparison of fair employment practice legislation to the Nuremberg laws. The 1964 remarks — "If we pass a law saying that race shall not be a factor in employment, then what grounds do we have for opposing a law that race shall be a factor?" — were made to the Harvard Young Conservative Club and reported in The Harvard Crimson, reproduced at irwincollier.com. On Friedman's 1957 action regarding the American Economic Association and segregated hotels, see thedailyeconomy.org, published by an organization sympathetic to Friedman. On Goldwater's votes for the 1957 and 1960 Acts, his desegregation of the Arizona Air National Guard, and the Title II basis of his 1964 opposition, and on Hayek's support for the Act, see the discussion at econlib.org. On National Review's characterization of Brown as "an act of judicial usurpation," "shoddy and illegal in analysis, and invalid as sociology," Buckley's description of the Civil Rights Act, and the observation regarding forfeited standing, see Claremont Review of Books, "Civil Rights and the Conservative Movement," claremontreviewofbooks.com. We have deliberately cited sources sympathetic to these figures alongside sources critical of them; every quotation above should be verified against the original published text before publication, as several circulate in paraphrase.
- 303 Creative LLC v. Elenis, 600 U.S. 570 (2023). Majority by Justice Gorsuch; dissent by Justice Sotomayor, joined by Justices Kagan and Jackson. Source for the holding that the First Amendment prohibits compelling a website designer to create expressive designs conveying messages with which she disagrees; for Gorsuch's reasoning that the objection ran to messages rather than customers, supported by the parties' stipulation that she would serve LGBT customers other products; for the Court's statement that public accommodations laws are vital to civil rights and that governments have a compelling interest in eliminating discrimination in public accommodations; and for Sotomayor's dissenting characterization that the Court had for the first time granted a business open to the public a constitutional right to refuse service to members of a protected class. See supreme.justia.com and the Congressional Research Service analysis at congress.gov. On the contested scope of the holding, see Yale Law Journal, "'We Do No Such Thing': 303 Creative v. Elenis and the Future of First Amendment Challenges to Public Accommodations Laws," yalelawjournal.org, which argues the majority overstates the reach of the compelled-speech prohibition.
A note on the author
Issue papers are published under the name of 1863 Leadership rather than an individual byline.
A note on sources
The two load-bearing citations in this paper are a top-five economics journal and the Proceedings of the National Academy of Sciences, and they point in the same direction from opposite methods — one modelling what openness was worth, the other measuring what closure still costs. Note 3 is the most difficult in the series to date, and we have cited sources sympathetic to Friedman and Goldwater alongside sources critical of them, flagging that several of the quotations circulate in paraphrase and require verification against original texts. Note 4 cites the Yale Law Journal arguing that the decision we adopt as our boundary was wrongly reasoned, because a reader deciding where the line falls should see that the line is contested.
Recommended citation
1863 Leadership. "Ninety-Four Percent." Issue Paper No. 14.
September 2026. 1863leadership.org
Corrections: None to date. Errors of fact are corrected on this page within one business day of notice, with a dated note describing the change.