Genetic selection

Thesis

For roughly seven centuries before the Industrial Revolution, England was a Malthusian society: population was held near subsistence by periodic mortality crises, and surplus wealth translated into surplus children. In this regime, the rich reproduced more than the poor: Gregory Clark’s reconstructed probate data show that testators in the top wealth quartile had roughly twice the surviving offspring of testators in the bottom quartile over the period ~1250–1800. Because rich families could not all remain rich across generations (primogeniture, division, bad luck, regression to the mean), the descendants of the English middle and upper classes gradually diffused downward through the social structure. If the behavioral traits that made their ancestors rich — patience, literacy, thrift, work intensity, low violence — had any heritable component (genetic or cultural-transmission), then seven centuries of this process would have shifted the population distribution of those traits substantially. By 1800, England was populated by a downwardly-diffused descendant group of its medieval propertied classes. The IR was the payoff.

This is the strong-form Clark thesis (A Farewell to Alms, 2007). A more biological version — Cochran and Harpending, The 10,000 Year Explosion (2009) — makes the explicit case that the selection was partly genetic and identifies candidate alleles (though this is highly speculative). The position is treated at length here despite being heterodox because (a) the empirical finding it rests on (the differential-reproduction-by-wealth pattern in pre-industrial England) is robust and important, (b) the contemporary genomics literature on selection signals in human populations has direct bearing on whether the proposed mechanism is feasible, and (c) the methodological objections to it are themselves productive — they have shaped the cultural-transmission and human-capital literatures that now occupy much of the space the genetic version was trying to fill.

Lead proponents

Gregory Clark — A Farewell to Alms (2007) is the foundational statement. Clark followed up with The Son Also Rises: Surnames and the History of Social Mobility (Princeton, 2014), using surname-based social-mobility evidence across England, Sweden, the United States, China, India, Korea, Japan, and Chile to argue that underlying social “status” is far more heritable than conventional mobility statistics suggest — intergenerational status correlations of ~0.7–0.9 across all societies tested, far above the conventional 0.2–0.5 income/education correlations. Clark’s most recent work (The Inheritance of Social Status, PNAS 2023) extends the surname-tracking strategy to genealogical-network data and argues that the underlying status-heritability is consistent with a high heritability of an underlying behavioral-genetic trait.
Greg Cochran (with Henry Harpending) — The 10,000 Year Explosion (2009) is the explicit biological framing. Cochran–Harpending argue that human evolution has been accelerating since the agricultural revolution because population growth expanded the substrate over which mutations can arise and be selected on, and because new selective environments (sedentism, dense populations, novel diseases, complex social organization) created persistent directional selection pressures. Their well-known case studies are lactase persistence in dairying populations, malaria-resistance alleles, and (more controversially) the Ashkenazi cognitive-trait hypothesis (developed in their 2006 Journal of Biosocial Science paper).
Henry Harpending specifically brought the population-genetics expertise; his earlier collaboration with Alan Rogers on demographic-history inference from mitochondrial DNA was foundational in the field. The Cochran–Harpending biological version is more explicit than Clark about the genetic mechanism but covers fewer historical specifics.

The probate evidence

The empirical core of the Clark argument is the demographic finding from English probate (will) records.

The dataset. Clark uses ~3,000+ probated wills from rural English parishes (chosen because parish records of births, deaths, and marriages are relatively complete) over the period 1250–1800. From the wills, he extracts the testator’s recorded wealth (movable goods, debts owed, real property where mentioned) and matches the testator to parish records of surviving children. The matching is imperfect — parish records become more complete after the 1538 Cromwell injunctions; wealthy testators with London business interests are underrepresented; younger sons who died without leaving wills are missed entirely — but the dataset is the largest sustained reconstruction of wealth-correlated fertility in pre-industrial England, and probably the largest such dataset in any pre-industrial society.

The pattern. Across the full period, testators in the top quartile of probated wealth had roughly twice the surviving offspring of testators in the bottom quartile. The differential is robust to alternative wealth measures (movable goods only; movable goods plus real property; Bourdieuian “social capital” proxies based on signature legibility and witness lists). It is also robust to alternative time-window subsetting: the differential is present in the 1250–1500 window, the 1500–1700 window, and the 1700–1800 window. It does not depend on plague-driven peculiarities of the 14th-century data.

The mechanism. The differential reflects two compounding effects. First, infant and child mortality rates in pre-industrial England were severe (~25–35% pre-1800) and strongly inversely correlated with wealth: rich families’ children survived in greater numbers because of better nutrition, less crowded housing, and (later) some access to medical care. Second, marriage formation was wealth-gated: the median English marriage age was unusually late (~26 for men, ~24 for women) and marriage required setting up an independent household, which required wealth. Poor men and women often delayed marriage past peak fertility years or never married; rich men and women married younger and had longer fertile-marriage spans.

The downward-diffusion claim. Primogeniture concentrated inheritance in eldest sons. Younger sons of well-off families either entered the church, the army, the law, or commerce — or married downward into lower social strata. Daughters, with smaller dowries than the eldest son’s inheritance, married downward more reliably. Over generations, this produced systematic downward genetic and cultural flow from the propertied class into the broader population. Clark’s surname-tracking work (The Son Also Rises, 2014) supports the long-horizon persistence of “status” at the surname level: surnames associated with elite occupations (Oxbridge admissions, professional registrations, parliamentary representation) regress to the population mean very slowly, with apparent half-lives of 5–10 generations — far slower than conventional mobility statistics would suggest, and consistent with a high underlying heritability of whatever the surname is tracking.

The 10,000-Year Explosion mechanism

The Cochran–Harpending framework provides the explicit biological-genetic component that Clark’s framework is more elliptical about.

The core argument is that human biological evolution has been accelerating since the agricultural revolution, not stopping or slowing as mid-20th-century anthropology assumed. Three mechanisms drive the acceleration. First, population growth: a population of 7 billion has many more new mutations per generation than a population of 7 million; the substrate over which selection can operate is larger. Second, novel selective environments: agriculture, sedentism, dense urban populations, and complex social organization create persistent directional selection pressures absent in hunter-gatherer environments. Third, cumulative effect: even modest selection coefficients (~1% per generation) compound to substantial trait shifts over 100–500 generations.

The empirical exhibit cases are robust. Lactase persistence — the ability to digest milk-sugar in adulthood — is a derived trait that emerged in dairying populations (Northern Europe, Northeast Africa, parts of West Asia) within the past ~7,500 years, with selection coefficients estimated at 5–10% per generation. Malaria-resistance alleles — sickle-cell, β-thalassemia, G6PD deficiency, Duffy-negative blood — emerged independently in multiple populations exposed to malaria over the past ~5,000 years. Skin pigmentation lightening in northern populations occurred within the past ~5,000 years. The Cochran–Harpending claim that human evolution has accelerated is empirically secure for these well-documented traits.

The contested extension is to cognitive and behavioral traits. Cochran and Harpending argued in their 2006 paper “Natural History of Ashkenazi Intelligence” (Journal of Biosocial Science) that the higher mean IQ scores of Ashkenazi Jewish populations (~110, vs. European population mean of 100) reflect ~600 years of selection in medieval European urban occupations (moneylending, commerce, scholarship) that combined with the population’s reproductive isolation to produce measurable cognitive shifts. The argument identifies several Ashkenazi-prevalent disease alleles (Tay-Sachs, Gaucher, Niemann-Pick) as potentially heterozygote-advantageous for neuronal function, in the way that sickle-cell heterozygotes are protected from malaria. The argument is empirically suggestive (the disease-allele clustering is real; the IQ-score gap is real and replicable) but the causal claim that the disease alleles produced the IQ advantage is unproven and is treated by most population geneticists as speculative.

The application to the IR. Cochran and Harpending argue that if 600 years can produce a measurable cognitive shift in Ashkenazi populations through selection in commercial and scholarly occupations, then 700 years of differential reproduction in the English propertied class could plausibly have shifted the English population distribution of cognitive and behavioral traits relevant to industrial-economic outcomes. The argument is structurally parallel to Clark’s but with explicit genetic mechanism.

Modern PGS evidence and the genomics critique

Since 2015 the genomics literature has begun to provide direct empirical purchase on the Cochran–Harpending claim. The new tool is the polygenic score (PGS) — a genome-wide summary of allele frequencies weighted by their effect sizes on a trait of interest, built from genome-wide association studies (GWAS) on contemporary populations.

Polygenic scores for educational attainment (PGS-EA) — derived from GWAS on years-of-schooling outcomes in millions of contemporary European-ancestry individuals — capture roughly 11–13% of the variance in educational attainment in held-out samples. PGS-EA is the most-used cognitive-related polygenic score and is the natural empirical proxy for the kind of behavioral-trait selection Clark and Cochran–Harpending propose.

Several studies have looked for selection signals on PGS-EA in modern genome data. Kong, Frigge, Thorgeirsson et al. (PNAS, 2017) used the Icelandic genealogy database to estimate that PGS-EA has been under negative selection in Iceland over the past century: higher-PGS-EA individuals have fewer children, by enough to reduce the population mean PGS-EA by ~0.04 standard deviations per generation. Beauchamp (PNAS, 2016) found a similar contemporary negative selection signal on PGS-EA in US data. Hyytinen, Bingley, Christensen et al. (2019) extended the analysis to Danish and Finnish data with consistent results.

The implication for the Clark thesis is mixed. Yes, contemporary genome data can detect selection on PGS-EA — the Cochran–Harpending claim that cognitive-trait selection is empirically tractable is vindicated in principle. But the direction of the contemporary selection signal is opposite to what the Clark thesis predicts: in modern post-demographic-transition societies, higher cognitive PGS individuals have fewer children, not more. The Clark thesis applies specifically to pre-demographic-transition societies where higher wealth correlated with higher fertility, but no contemporary genome data set is from such a population.

The remaining empirical question is whether genome data from historical European populations can detect a positive selection signal on PGS-EA across the period 1250–1800 — directly testing the Clark mechanism. Recent ancient-DNA work (Mathieson et al., Nature, 2015 and follow-ups; the Reich Lab’s work on European population history) has the methodological capacity to look at this question, but no published study has yet found a clean positive selection signal on cognitive-trait PGS in pre-industrial European samples. The absence is not yet a refutation — the historical samples are small and the signal-detection threshold is demanding — but it is the empirical place where the Clark thesis is most directly testable and where the evidence so far does not corroborate it.

Heritability methodology debates

Even granting the differential-fertility evidence and the in-principle feasibility of selection on behavioral traits, the Clark/Cochran argument requires that the relevant traits be substantially heritable. This is the most methodologically contested step in the chain.

Twin-study estimates. Decades of twin and adoption studies report heritability estimates of 0.4–0.8 for cognitive ability (IQ), 0.3–0.6 for personality traits (Big Five), 0.4–0.6 for educational attainment, 0.3–0.5 for income, and 0.3–0.5 for measured patience and time-preference. These estimates underwrite the standard population-genetic calculation that 25 generations of consistent directional selection on any trait with heritability 0.3+ produce shifts of ~0.2–0.5 standard deviations.

The objections to twin-study heritability are well-known. The “equal environments assumption” (that monozygotic twin pairs and dizygotic twin pairs experience equally similar environments) may be violated, biasing heritability upward. Heritability estimates from population-typical environments do not generalize to environments outside that range. Heritability is a population-and-environment-specific quantity, not a property of the trait itself. None of these objections fatal to the standard estimates, but all qualify them.

The GWAS-vs-twin gap. Empirical GWAS estimates of heritability (the variance explained by all measured SNPs) are systematically lower than twin estimates — for educational attainment, GWAS heritability is ~0.20–0.25 in best estimates, vs. twin estimates of ~0.4–0.6. The gap (sometimes called “missing heritability”) may reflect rare-variant contributions, gene-gene interactions, structural-variant contributions, or methodological-artifact contamination of twin estimates. The Clark/Cochran argument needs the twin estimate to be approximately right; the GWAS evidence narrows but does not close the case for that.

The cultural-transmission alternative. The same downward-diffusion mechanism Clark proposes works under pure cultural transmission — parents transmit norms, attitudes, time-preferences, and skills to children through socialization rather than through genes. Cultural transmission can be approximately as durable as genetic transmission across small numbers of generations and produces patterns observationally similar to genetic transmission (within-family resemblance, surname-correlated outcomes, regression to the mean). The cultural-transmission version of the Clark mechanism is much less controversial — economic historians broadly accept that cultural transmission of norms and skills mattered for industrialization — but it is also much less distinctive: it is a Bourdieuan cultural-reproduction account dressed in population-genetic vocabulary, and it does not require the biological-genetic claim that gives the Clark/Cochran framework its provocative edge.

de la Croix and the formal critique

David de la Croix, Matthias Doepke, and Joel Mokyr’s “Clans, Guilds, and Markets: Apprenticeship Institutions and Growth in the Pre-Industrial Economy” (QJE, 2018) is the most influential recent formal critique. The paper does not engage Clark directly but builds an alternative skill-transmission framework — apprenticeship institutions, guilds, and markets — that explains the same long-run human-capital accumulation Clark attributes to differential reproduction, without requiring any genetic component.

The argument. Pre-industrial skill transmission ran through three institutional channels: clans (within-family transmission, dominant in pre-modern China), guilds (within-craft transmission across families, dominant in medieval Europe), and markets (open apprenticeship contracts, dominant in early-modern Northwestern Europe). The market channel was uniquely capable of accumulating skill across generations because it transmitted skills across family lines, mixing the best masters with the most promising apprentices regardless of family origin. The clan channel locked skills within lineages, limiting innovation. The guild channel restricted entry, slowing skill diffusion. The market channel — which Northwestern Europe developed specifically — produced the human-capital accumulation needed for industrialization, without requiring any biological-genetic mechanism.

The implication is that the apprenticeship-institutions framework provides a complete alternative explanation for the long-run human-capital accumulation in pre-industrial Northwestern Europe, with much firmer empirical foundations (apprenticeship records exist; ancient-DNA selection signals on cognitive traits do not). To the extent the de la Croix–Doepke–Mokyr account is correct, the residual that Clark/Cochran genetic-selection is needed to explain shrinks substantially or disappears. The mainstream uptake of the apprenticeship framework — see the upper-tail human capital position — has correspondingly reduced the perceived need for the genetic-selection account.

Key arguments

English probate records show differential reproduction by wealth. Clark’s data, 1250–1800: testators in the top quartile of probated wealth had ~2× the surviving offspring of the bottom quartile. This differential persisted for roughly 25 generations.
Rich family lineages diffused downward. Primogeniture concentrated inheritance in eldest sons; younger sons and daughters of well-off families married into lower social strata, taking their traits and (per the hypothesis) their genes with them. Surname-tracking work (The Son Also Rises, 2014) supports the long-horizon persistence of “status” at the surname level, with regression-to-the-mean half-lives of 5–10 generations.
Selection over centuries can produce substantial trait shifts. Lactase persistence and malaria-resistance alleles document that human evolution has produced substantial trait shifts within historical timescales. If patience, literacy, violence-aversion, and work intensity have heritability ~0.3–0.5, 25 generations of consistent selection would produce shifts of ~0.5–1.0 standard deviations — enough to plausibly change the social and economic aggregate character of a population.
The IR fits the pattern. The traits selected for are exactly those associated with market-based production, capital accumulation, and deferred consumption. A population with a long-diffused middle-class genetic/cultural inheritance would industrialize more readily than one without.
Only the densely documented English case is testable. Clark’s evidence is particular to England (rich probate records, stable surnames). The theory requires (and implicitly predicts) similar differential reproduction in other IR-precursor societies, which is much harder to reconstruct.

Key evidence

Clark’s probate dataset. ~3,000+ pre-1800 English wills reconstructing wealth-correlated fertility. This is the central empirical finding and is not seriously disputed in its descriptive form.
Surname-based social mobility evidence (The Son Also Rises, 2014). Very slow regression-to-the-mean of elite-associated surnames (Oxbridge, high professional registrations, parliamentary representation) over 5–10+ generations, implying high “underlying” status heritability across multiple societies (England, Sweden, US, China, India, Korea, Japan, Chile).
Twin and adoption studies. Decades of behavior-genetic work showing moderate-to-high heritability (~0.3–0.7) of personality, IQ, education, and income. The Clark thesis does not require high heritability, but does require some non-zero heritable component on relevant traits.
Documented selection signals on derived traits. Lactase persistence, malaria-resistance alleles, skin pigmentation — all documented to have undergone substantial population-level selection within historical timescales. Provides existence-proof for the Cochran–Harpending claim that selection on relevant timescales is feasible.
Continental comparisons. Clark discusses (but has much less data for) France, Japan, and China, suggesting the differential-reproduction pattern was weak or absent in Qing China, which might explain why similar material conditions did not produce an IR there. The continental data is much thinner than the English data and is the place where the cross-cultural generalization is empirically weakest.

Major critiques

The selection pressure is smaller than claimed. Standard population-genetic math suggests 25 generations of even strong wealth-correlated fertility differentials produce shifts of perhaps 0.2–0.4 standard deviations on heritable polygenic traits — not nothing, but far short of what would be needed to cause an Industrial Revolution on its own. The Clark thesis requires the mechanism to be load-bearing; the population-genetic feasibility analysis suggests it is plausibly contributory but not load-bearing.
Heritability of “bourgeois traits” is speculative. What, precisely, is being selected for? “Patience” is partly heritable; “industriousness” is a construct; “market orientation” is culturally coded. Without specifying the trait and its genetic architecture, the selection claim is hand-wavy. The polygenic-score-for-educational-attainment work (Lee et al., Nature Genetics, 2018; Okbay et al. follow-ups) provides a tractable proxy but is itself only modestly powered (PGS-EA explains ~11–13% of variance in EA), and the mapping from PGS-EA to the historical traits Clark proposes is ambiguous.
Contemporary selection signals run the wrong way. Kong et al. (2017) and Beauchamp (2016) detect negative selection on PGS-EA in contemporary post-demographic-transition populations. The Clark thesis applies to pre-demographic-transition Malthusian populations specifically; the absence of any historical-genome-data positive selection signal on cognitive-trait PGS is the most direct empirical absence in the case. Recent ancient-DNA capability could in principle test this; no clean positive result has been published.
Cultural transmission is a better fit. The same differential-descent mechanism works under pure cultural (parent-to-child, elite-to-non-elite) transmission of norms, attitudes, and skills, without requiring any genetic change. This version is much less controversial but also less distinctive — and is essentially what the de la Croix–Doepke–Mokyr (2018) apprenticeship framework provides.
The cross-cultural story is thin. Why didn’t Tokugawa Japan, Song China, or the high Dutch Republic — all with propertied elites, high fertility differentials among the well-off, and centuries of stable reproduction — produce similar outcomes? The theory is tested essentially on England; everywhere else is a just-so explanation of why it didn’t happen. The Tokugawa Japan case is particularly hard for the Clark thesis because Japanese demographic data show comparable wealth-correlated fertility patterns without producing comparable subsequent industrialization.
The de la Croix–Doepke–Mokyr formal alternative. The apprenticeship-institutions framework (QJE, 2018) provides a complete alternative explanation for the long-run human-capital accumulation in Northwestern Europe, with firmer empirical grounding (apprenticeship records exist; selection signals don’t). To the extent that account is correct, the residual that Clark genetic-selection is needed for shrinks toward zero.
Allen and McCloskey’s methodological objections. Robert Allen has criticized the Clark framework for treating differential fertility as causally prior when it is plausibly consequent on the same wealth differentials that drove industrialization through other channels (capital accumulation, institutional formation). Deirdre McCloskey treats the Clark thesis as a category error: economic-cultural change at civilizational scale operates through ideas, institutions, and rhetoric, not through allele-frequency shifts that would take many more generations to accumulate to consequential magnitudes than the historical timeline allows.
Ideological and methodological suspicion. The Clark thesis — and especially Cochran’s biological version — sits uncomfortably adjacent to older racialist theories of European superiority that have been consensus-rejected for good reason. The empirical mechanisms proposed (differential reproduction → selection → economic outcomes) are plausible in principle but the specific Clark/Cochran claims about which traits, in which direction, with what magnitude, are speculative in ways that go beyond ordinary scientific dispute. The reasonable response is sceptical engagement on the methodological merits, not preemptive dismissal — but the political loading of the topic shapes how it is received in the field.
Clark’s own later work softens the claim. In The Son Also Rises (2014) and in subsequent papers, Clark has shifted toward a “status heritability” framing that’s less explicitly biological. Many readers interpret this as a retreat toward a Bourdieuan cultural-reproduction model dressed in population-genetic language. The 2023 PNAS paper restates the underlying-trait-heritability claim more strongly than the 2014 book did but still carefully avoids committing to a specific genetic-vs-cultural decomposition.

Status

Heterodox. Clark’s book is widely read and widely cited as an important provocation; the core empirical finding (differential reproduction by English wealth class over centuries) is broadly accepted. The causal claim — that this drove the IR — is accepted by essentially no one outside Clark’s immediate orbit and Cochran’s, and is treated by the rest of the field as a speculative hypothesis deserving of engagement but not adoption. Its value has been partly methodological: forcing mainstream economic historians to think harder about the very-long-run accumulation of human capital and behavioral capacity, and about pre-industrial selection pressures generally. The most productive intellectual successors to the Clark/Cochran framework are not in genetics but in the cultural-transmission and apprenticeship-institutions literatures (de la Croix–Doepke–Mokyr 2018; the upper-tail human capital position more broadly), which have absorbed the long-run-accumulation insight while replacing the genetic mechanism with a documented institutional one. The genetic-selection version has been productively wrong: it pointed at a real long-run accumulation question, and the field’s answer to that question has turned out to be institutional rather than genetic.