The Debate over the Genetic Nature of Race and the Causes of Group Differences

The Postwar Consensus and Its Institutional Construction

The modern claim that race is a social rather than biological category did not emerge spontaneously from laboratory findings; it was in significant part manufactured by an international organization. Poul Duedahl's archival study documents how UNESCO, from 1945 onward, placed the redefinition of race at the center of its "mental engineering" program, with the explicit aim of dismantling doctrines of racial inequality in the wake of the Holocaust.¹ Julian Huxley, UNESCO's first Director-General, deliberately privileged the social sciences over the natural sciences on the grounds that social scientists "mastered the arguments" needed to combat inequality.¹ The 1950 Statement on Race, drafted largely by Ashley Montagu from a panel recruited almost entirely from anthropologists and sociologists who regarded race as a social construct, declared that race was "less a biological fact than a social myth" and recommended replacing the term with "ethnic group."¹

Duedahl shows that this statement was immediately attacked by physical anthropologists — notably William Fagg, editor of *Man* — precisely because no physical anthropologist of standing had been consulted, and that the resulting controversy forced UNESCO to commission a second, more cautious statement in 1951 under the geneticist L. C. Dunn.¹ The second statement conceded that dividing humanity into three main races made sense for physical, though not mental, traits — a retreat from the first, but also a reinforcement of the anti-racist point.¹ Duedahl's overall assessment is that UNESCO succeeded in imposing a new postwar view of man, that its statements were cited in American desegregation litigation including *Brown v. Board of Education*, and that the discipline of anthropology itself was reshaped so that physical anthropology became a subdiscipline while culture displaced race as the field's unifying concept.¹ He is also candid that the campaign was uneven, that its early focus was overwhelmingly on the United States and South Africa, and that only after 1960 did it acquire genuinely global reach.¹ Several of the other sources take this history as the origin of what they regard as an orthodoxy that has outlived its evidentiary base.²³

Lewontin's Apportionment and Edwards's Rebuttal

The single most influential technical argument in the debate is Richard Lewontin's 1972 finding that roughly 85 percent of human genetic variation lies within populations and only about 15 percent between them, from which he concluded that racial classification is "of virtually no genetic or taxonomic significance."² A. W. F. Edwards's short 2003 paper argues that this conclusion rests on a statistical fallacy: analyzing loci one at a time and then drawing taxonomic conclusions from the single-locus results ignores the information contained in the correlation structure across loci.² Edwards constructs a simple two-population example in which the within-population share of variance at each locus remains 84 percent, yet the probability of misclassifying an individual falls essentially to zero once a hundred loci are considered jointly.² He traces the underlying logic back to Fisher and to the 1920s debate over Pearson's coefficient of racial likeness, and notes that Cavalli-Sforza and Edwards had derived a convincing evolutionary tree from very similar data as early as the 1963 International Congress of Genetics — an analysis Lewontin, though present at the Congress, did not cite.² Edwards closes with the observation that it is dangerous to premise the moral equality of human beings on biological similarity, "because dissimilarity, once revealed, then becomes an argument for moral inequality."²

Nicholas Wade reproduces Edwards's argument and adds the observation that Sewall Wright, who invented the fixation index Lewontin used, regarded values in the 5–15 percent range as indicating moderate but by no means negligible genetic differentiation.³ Winegard, Winegard and Boutwell likewise cite Edwards, together with studies by Tang and colleagues showing that self-reported ethnicity corresponded closely with genetic clusters derived from 326 microsatellite markers.⁴ Charles Murray's approach is a direct extension of the same logic: he plots target allele frequencies for pairs of populations and shows that within-continent pairs correlate at around +.98, while cross-continent pairs for schizophrenia-associated SNPs correlate at +.70 to +.81 — high by social-science standards but consistent with very many large individual differences.⁵

The Chinese-language review by Xi Huanjiu and Li Wenhui gives Lewontin's finding a different weight, reporting it as evidence that within-group variation exceeds between-group variation and citing it among the reasons the concept of race has been questioned; the authors do not engage Edwards's rebuttal.⁶ This is a genuine point of divergence within the collection: the same 85/15 figure is treated by one set of sources as decisive and by another as a well-documented misreading.

Clines, Clusters, and the Status of "Race"

Several sources converge on a moderate position: races are not discrete types with sharp boundaries, but neither is the absence of sharp boundaries an argument against their usefulness. Wade reviews the clustering literature — Bowcock and Cavalli-Sforza's 1994 tandem-repeat study, the Rosenberg–Feldman analyses using 377 and then 993 markers, the Li and Myers SNP survey of 51 populations, and Tishkoff's identification of fourteen ancestral groups within Africa — and reports that all of them recover the continental groupings, with additional clusters (South Asia, the Middle East) emerging as more markers are used.³ He quotes Neil Risch to the effect that these studies "recapitulated the classical definition of races based on continental ancestry," and Marcus Feldman's remark that the data back up Risch's claim.³ Wade answers the standard objections directly: that variation is clinal (true, but categories can be useful without being discrete), that classifications are arbitrary (they are pragmatic but constrained by ancestry, genetic profile, and phenotype), and that boundaries cannot be drawn (races are identified by clusters of traits, not by any single necessary feature).³ He devotes particular criticism to Jared Diamond's argument that different equally valid classification procedures yield contradictory results, calling the malaria-resistance example evolutionarily incoherent because resistance arose independently and recently in each population.³

Winegard, Winegard and Boutwell advance essentially the same moderate position by analogy with film genres: categories that are fuzzy, interest-relative, and lacking metaphysical essences can nonetheless be predictive and non-arbitrary.⁴ They prefer the term "human population" while insisting that the substantive argument is unaffected by nomenclature.⁴

Xi and Li's review reaches a superficially similar but differently weighted conclusion. They trace the concept from Bernier in 1684 through Linnaeus, Blumenbach, Cuvier and Coon, catalogue the shifting criteria (skin color, head form, nasal index, blood groups), and emphasize that classification standards differ between living subjects and skeletal material, that a single skull can be assigned to different races depending on which measurement is privileged, and that the same Anyang and Shanpula remains have been variously classified by different scholars.⁶ They report the American Association of Physical Anthropologists' 1996 statement rejecting the race concept as lacking scientific utility, and note that Braveman argued in 2021 that the concept originated in the slave-trading era and should be abandoned.⁶ Their own judgment, however, is that although racial boundaries are blurring, differences are narrowing, and the concept is fading, it cannot yet be abolished — because of its continuing utility in medicine, forensics, epidemiology and human-factors engineering.⁶ They give examples: elevated hypertension and several cancer mortality rates among African Americans, an eighty-fold difference in basal-cell and squamous-cell carcinoma rates between whites and African Americans, and differences in lactose tolerance.⁶ Their most distinctive contribution is a critique of terminological confusion, arguing that the Chinese *minzu*, the Western *ethnic group*, and *race* have different histories and referents, that American ethnic groups are constituted across racial lines while Chinese minority nationalities are not, and that the imported term "族群" cannot simply substitute for "民族."⁶

Recent Selection and the Evidence for Population Differences

A second line of argument in the collection concerns the tempo of human evolution. Wade summarizes the genome-scan literature of the mid-2000s: Pritchard's finding of roughly 200 selected regions in each of the three major races with little overlap; Sabeti's 412 fine-mapped regions divided almost evenly among Europeans, East Asians and Africans; Akey's estimate that at least 8 percent of the genome shows evidence of recent selection, with about 80 percent of those signals being local adaptations.³ Crucially, Wade introduces the distinction between hard sweeps, in which a single new mutation goes to or near fixation (rare in humans, and the reason races do not differ by fixed alleles), and soft sweeps acting on standing variation across many loci, which he argues is the dominant mechanism and explains how populations can differ substantially in polygenic traits while sharing the same gene repertoire.³ Winegard and colleagues make a parallel argument from the animal literature, tabulating rapid evolution in Italian wall lizards, Australian snakes responding to cane toads, house sparrows in North America, Darwin's finches, peppered moths and guppies, and arguing that thirty generations can suffice for measurable morphological and behavioral change.⁴ They are careful to reproduce the original authors' own caveats — Herrel et al. note that phenotypic plasticity and maternal effects need further investigation, and Phillips and Shine acknowledge they lack direct data distinguishing selection from plasticity — and they explicitly warn that natural selection should not be the default assumption when group differences appear.⁴

The Akbari et al. ancient DNA study provides the most substantial new empirical contribution in the collection and, in several respects, adjudicates the older debate. Analyzing 15,836 West Eurasians spanning 18,000 years using a generalized linear mixed model that corrects for population structure, the authors identify 479 independent loci with a posterior probability of directional selection above 99 percent, and project at least 3,800 independent episodes at a 50 percent false discovery rate — a twentyfold increase over previous scans.⁷ They confirm that classic hard sweeps have been rare and that only about 2.16 percent of allele frequency change is attributable to directional selection, but argue there is no conflict: allele frequency change has been so pervasive that even a small fraction corresponds to thousands of loci.⁷ They document selection for light skin at ten loci and none for dark skin, selection at HLA-DQB1 for coeliac risk, reversals of selection at TYK2 and HFE, and — importantly for the debate over multiple sclerosis and steppe ancestry — a rebuttal of the claim that the north–south MS gradient was driven by selection on the Yamnaya steppe, showing instead that selection began south of the Caucasus in people without steppe ancestry and was subsequently stronger in northern Europe.⁷ This is a case where a widely publicized population-difference explanation is corrected by better data.

Polygenic Signals for Cognitive and Behavioral Traits

The most contested territory concerns traits related to cognition and behavior. Akbari et al. report that they found no enrichment of genome-wide significant selection signals among variants affecting mental, psychiatric or behavioral traits, in contrast to strong enrichment for blood, immune and inflammatory traits.⁷ Yet at the polygenic level they detect significant directional selection against predictors of bipolar disorder and schizophrenia, and for predictors of intelligence-test scores, household income and years of schooling, with all three of their statistics significant.⁷ They explain the apparent tension by noting that brain traits have a qualitatively different genetic architecture: extremely polygenic, with per-allele effects too small for individual detection — one must drop 825 loci for the schizophrenia signal to vanish and 449 to 1,056 for the behavioral traits.⁷ They replicate the years-of-schooling signal using East Asian GWAS effect sizes, whose population structure is uncorrelated with West Eurasian structure, and using family-based GWAS immune to stratification.⁷

The authors surround these findings with unusually strong caveats. They stress that the effects were measured in industrialized societies and that "it remains unclear how these relate to phenotypes that were adaptive in the past"; that scores on intelligence tests, income and schooling "are only relevant to modern societies and would have been unmeasurable in preliterate societies"; that the alleles driving these signals are highly correlated with those reducing type 2 diabetes-related traits, complicating interpretation; and that selection pressures have changed, citing evidence of selection *against* the educational-attainment predictor in twentieth-century Iceland.⁷ They also failed to distinguish cognitive from non-cognitive components of the educational attainment signal and found no signal for brain volume.⁷ The paper thus supplies evidence that is simultaneously the strongest in the collection for polygenic selection on behavior-related variants and the most explicitly hedged about what that evidence licenses.

Murray's chapter approaches the same question from a different direction, working from raw target allele frequency differences in the GWAS Catalog rather than from selection statistics. Using an operational definition under which a difference counts as "large" if it exceeds 99 percent of within-continent differences (approximately .186, rounded to .20), he reports that roughly a third of cross-continental target allele differences qualify as large for physiological traits, diseases, biomarkers, cognitive disorders, mental abilities and personality features alike, with Africa–Asia comparisons consistently showing the most and Asia–Europe the fewest.⁵ He is explicit that he is not presenting proof of phenotypic causation, only describing the raw material.⁵ He also foregrounds the problem that motivates caution: polygenic scores lose predictive validity as genetic distance from the discovery sample increases, and the European bias of genomic databases means such comparisons cannot yet be trusted.⁵ His argument is that this bias is now being criticized on health-equity grounds, that large non-European samples will soon exist, and that population differences will therefore be studied intensively whether or not anyone wishes it.⁵ He catalogues differences already established — lactase persistence, sickle cell, high-altitude adaptation via distinct routes in Tibetans, Andeans and Ethiopians, Ashkenazi genetic disorders, the prostate cancer risk loci, immune and inflammatory susceptibility — and closes by arguing that the sex difference in violence, with roughly 90 percent of homicides committed by males, is far larger than any population difference likely to be found, and that if that difference can be lived with, so can smaller ones.⁵

The Heritability Argument and Its Philosophical Critics

Sesardic's paper is the collection's most sustained methodological intervention, and it is directed not at geneticists but at philosophers of science. He identifies a "master argument," derived from Lewontin's 1970 seed-in-two-soils thought experiment, which holds that hereditarians infer between-group heritability directly from within-group heritability and are thereby refuted.⁸ He quotes Papineau, Richardson, Block, Sarkar, Daniels, Longino and Kitcher endorsing versions of this claim, and Richardson explicitly attributing Jensen's alleged blindness to latent racism.⁸ Sesardic then argues that the attribution is textually false: Jensen repeatedly and explicitly stated that heritability coefficients by themselves cannot answer questions about group differences, and that his argument was a two-step one combining high within-group heritability with independent empirical evidence bearing on candidate environmental explanations.⁸ Sesardic distinguishes environmental hypotheses invoking factors that vary within groups from "X-factor" hypotheses invoking factors uniformly present in one group and absent in the other, and produces a table showing how rising heritability increases the between-group environmental difference required to explain a fifteen-point gap.⁸

Crucially, Sesardic's own position is agnostic on the substantive question: he states that purely environmental scenarios are not ruled out and that his aim is only to show that the philosophical critique is misconceived.⁸ His strongest supporting witness is James Flynn, a committed environmentalist, who wrote that Jensen's critics had largely evaded the argument by seizing on Lewontin, and that Lewontin's example shows only that high heritability can be ignored under highly specific and unusual circumstances.⁸ Sesardic also documents secondary errors: Sarkar's inference of political motive from the fact that heritability studies examine the five-factor personality traits (which are salient because of psychometric consensus, not politics), Sarkar's and Dummett's misattribution to Herrnstein of racial hereditarianism he had explicitly disclaimed in the early 1970s, and Gould's *Mismeasure of Man*, whose central attack on Jensen he argues depends on the false premise that Jensen's case rests on *g*, and whose Morton craniometry chapter he reports was contradicted by Michael's 1986 remeasurement of 201 specimens.⁸ He notes that Ulric Neisser, an opponent of hereditarianism, conceded the small overall trend in mean measured brain and skull sizes.⁸

The Ethics of Inquiry

The collection's final axis of disagreement concerns whether these questions should be pursued at all. Noah Carl catalogues the position he opposes — Kitcher's argument that evidentiary standards should scale with political cost, Block and Dworkin's call for voluntary abstention, Gillborn's proposal that all research on human capabilities be predicated on a statement that assertions of fixed group differences are racist by nature, and Kourany's proposal for a National Science Advisory Board imposing restrictions.⁹ He documents the treatment of E. O. Wilson, Jensen, Murray and Alison Stanger, and argues that the assumed asymmetry between costs and benefits has never been empirically demonstrated.⁹ His three main claims are that equating scientific statements with racism holds morals hostage to facts and commits the moralistic fallacy; that the blank-slate view has its own pernicious history, having been invoked to justify totalitarian projects of remaking humanity and persecutions of successful minorities; and that stifling debate has done demonstrable material harm, citing racial differences in drug metabolism and the British grooming-gang inquiries in which officials reported fear of appearing racist as an obstacle to intervention.⁹ He also cites evidence that opposition to political correctness independently predicted support for Donald Trump.⁹ Carl is careful to state that he has not shown the benefits outweigh the costs, only that one cannot take for granted that stifling debate is the ethical course.⁹

David Reich's op-ed occupies a middle position that all sides have found quotable. He accepts that race is a social construct and that Lewontin's finding of overall similarity is correct, but argues that the consensus has "morphed, seemingly without questioning, into an orthodoxy" holding that average genetic differences among racially defined groups are too trivial to matter.¹⁰ He describes his own laboratory's identification of a genomic region with 2.8 percent excess African ancestry in African American prostate cancer patients, containing at least seven independent West African–common risk factors sufficient to account for the disparity.¹⁰ He notes that the line into behavior and cognition has already been crossed by Benjamin's educational-attainment study, Kong's Icelandic follow-up showing that the same variants nudge carriers to delay childbearing, and Posthuma's intelligence-test work.¹⁰ Reich is equally emphatic in repudiating Nicholas Wade's book, which he calls unfounded and irresponsible for claiming that modern research supports traditional stereotypes, and in recounting James Watson's evidence-free assertions about Jewish and East Asian populations.¹⁰ His central prediction is that whatever we currently believe about population differences is most likely wrong — he cites his own 2016 finding that "whites" descend from four ancient populations as different from one another as Europeans and East Asians are today — and his recommendation is to model the response on the accommodation of sex differences: recognize that differences exist and accord equal freedoms regardless.¹⁰ Murray adopts a similar framing, arguing that the study of population differences is not an exercise in ranking and that no population will prove free of defects or possessed of all virtues.⁵ Winegard and colleagues make the strongest positive case for discussion, arguing that a vacuum left by responsible scholars will be filled by extremists, and that celebration of biological diversity is compatible with tolerance.⁴

Points of Genuine Disagreement

Beneath the shared vocabulary the sources differ substantially. On the reality of race, Duedahl's UNESCO tradition and the AAPA statement cited by Xi and Li hold that the concept lacks scientific utility; Edwards, Wade, Winegard and Murray hold that continental clustering is real and useful; Xi and Li and Reich occupy intermediate positions, retaining the term for practical purposes while denying it picks out discrete biological kinds.^12345610 On the tempo of evolution, Wade and Winegard emphasize rapid soft sweeps, while Akbari et al. supply direct measurement showing both that hundreds of loci have been under strong directional selection in the Holocene and that only about two percent of allele frequency change is directional.³⁴⁷ On the causes of group differences in cognition, Akbari et al. present polygenic signals hedged with extensive caveats, Murray presents raw allele-frequency imbalances he explicitly declines to interpret causally, Sesardic defends the legitimacy of the hereditarian inference without endorsing it, Reich warns that no such stereotype-confirming evidence exists, and Xi and Li reject the older psychometric arguments outright as products of unequal conditions.⁵⁶⁷⁸¹⁰ On the ethics, Carl and Winegard argue for open inquiry, Duedahl documents the reasons for caution, and Reich and Edwards both warn that resting moral equality on empirical sameness is a dangerous strategy.¹²⁴⁹¹⁰

Conclusion

Taken together, the collection describes a field in which the technical and the political have never been cleanly separable. The postwar consensus that race is a social construct was, on Duedahl's account, deliberately engineered by an international body pursuing a defensible moral aim through selective recruitment of experts.¹ Its central statistical prop, Lewontin's apportionment, has been argued by Edwards to rest on a fallacy that ignores correlation structure.² Genomic clustering, recent selection scans, and now ancient DNA time series have established that human populations differ in allele frequencies at many loci associated with disease, physiology and, more contentiously, with traits measured in modern societies as cognitive or behavioral.³⁵⁷ What remains genuinely open — and what the most careful sources in the collection, including Akbari and colleagues themselves, insist remains open — is what those frequency differences mean for phenotypes, how they were shaped by selection pressures no longer operating, and whether polygenic predictors developed in one population transfer to another at all.⁵⁷ The sources that argue most forcefully for open inquiry and those that argue most forcefully for caution converge on one point: that political equality is a moral commitment rather than an empirical hypothesis, and that grounding it in claims about biological sameness leaves it hostage to whatever the next decade of sequencing reveals.²⁹¹⁰

Sources

1. Duedahl2020-RaceConceptUNESCO copy.pdfPoul Duedahl, Changing the concept of race: On UNESCO and cultural internationalism, 2020Archival history of UNESCO's four race statements (1950–1967), the recruitment politics behind them, and their uneven political and legal impact in the United States, South Africa and Europe.
2. Edwards2003-LewontinFallacy.pdfA.W.F. Edwards, Human genetic diversity: Lewontin's fallacy, 2003Argues that Lewontin's 85/15 apportionment ignores the correlation structure across loci and therefore does not license the conclusion that racial classification lacks genetic significance.
3. TroublesomeInheritance-Chapter5.pdfNicholas Wade, The Genetics of Race, in A Troublesome Inheritance: Genes, Race, and Human History, 2014Popular synthesis of genomic clustering studies, genome scans for recent selection, and the hard-sweep/soft-sweep distinction, together with rebuttals of standard arguments against the reality of race.
4. Human Biological and Psychological Diversity (2017).pdfBo Winegard, Benjamin Winegard and Brian Boutwell, Human Biological and Psychological Diversity, 2017Proposes revising standard evolutionary psychology's assumption of a panhuman nature, using animal examples of rapid evolution and candidate cases such as self-construal styles and Ashkenazi intelligence.
5. HumanDiversity-Chapter9.pdfCharles Murray, The Landscape of Ancestral Population Differences, in Human Diversity: The Biology of Gender, Race, and Class, 2020Quantifies cross-continental target allele frequency differences in the GWAS Catalog, discusses population stratification and polygenic score portability, and catalogues already-established population differences.
6. 关于人种问题的讨论.pdf席焕久 and 李文慧, Discussion on racial concept, 2023Chinese-language review tracing the history of racial classification, surveying critiques of its biological basis, and distinguishing the concepts of race, minzu and ethnic group while defending the concept's continued practical utility.
7. Akbari2026-DirectionalSelectionWestEurasia.pdfAli Akbari et al., Ancient DNA reveals pervasive directional selection across West Eurasia, 2026Reports 479 loci under directional selection from 15,836 ancient West Eurasian genomes, including polygenic signals for pigmentation, body fat, schizophrenia and cognitive-performance predictors, with extensive caveats on interpretation.
8. sesardic2000.pdfNeven Sesardic, Philosophy of Science that Ignores Science: Race, IQ and Heritability, 2000Argues that the standard philosophical refutation of hereditarianism attacks a position Jensen never held, and documents further factual and interpretive errors in the philosophical literature on race and IQ.
9. How Stifling Debate Around Race, Genes and IQ Can Do Harm (2018).pdfNoah Carl, How Stifling Debate Around Race, Genes and IQ Can Do Harm, 2018Challenges the assumed asymmetry between the costs and benefits of discussing taboo topics, arguing that the moralistic fallacy, the blank slate's own history, and documented institutional failures cut against censorship.
10. Reich2018-GeneticsRace.pdfDavid Reich, How Genetics Is Changing Our Understanding of 'Race', March 23, 2018Op-ed by a leading geneticist accepting that race is a social construct while arguing that average genetic differences across ancestral populations are real and must be discussed candidly, and repudiating Wade's and Watson's claims.