The Bronze Age steppe migrations remade Europe. Around five thousand years ago, people related to the Yamnaya herders of the grasslands north of the Black and Caspian seas spread westward, bringing ancestry that became a major part of the European gene pool. The scale was enormous: Haak and colleagues estimated that the sampled Corded Ware people in Germany derived roughly 75% of their ancestry from a Yamnaya-related population.

Farther west, Olalde and colleagues found that the later Beaker-associated migration replaced about 90% of Britain’s gene pool within a few hundred years. Those migrants already carried a mixture of steppe and European farmer ancestry. It was a profound population turnover, even though the newcomers were themselves the product of earlier mixing.

We usually follow this story west, into Europe, or south towards India and Iran. But another part runs east, into the mountain valleys and deserts of what is now western China. Long before the Silk Road, Afanasievo-related people had brought Yamnaya-like ancestry into northern Xinjiang. Later arrivals introduced a different steppe profile, related to Sintashta and Andronovo.

“Western steppe herders” is the broader genetic category. “Aryan-related” here refers more specifically to that later Sintashta/Andronovo ancestry associated with Indo-Iranian dispersals. It connects to the larger steppe expansion through shared ancestry and subsequent mixing. Keeping those layers distinct lets us follow the eastern story without treating every steppe migrant as the same people.

I brought together 370 ancient Xinjiang genomes to follow both sides of this story: the arrival of steppe ancestry and its later loss of ground as East Asian ancestry increased. The sharpest later shift comes after the Iron Age. Comparing people buried at the same sites gives us a way to measure the change without confusing a later date with a different valley.

Two journeys east

There were two different steppe connections, separated by roughly a millennium. That distinction matters from the beginning: the first people to bring substantial steppe ancestry into northern Xinjiang were Afanasievo-related, while the later Andronovo connection belongs to a different expansion.

Zhang and colleagues (2021) found predominantly Afanasievo-related ancestry in northern Xinjiang individuals dating to about 3000–2800 BCE. These people were closely connected to the earlier Yamnaya-like expansion across the steppe. A possible link with the ancestors of Tocharian places them in another branch of the Indo-European story.

The later arrivals were genetically closer to Sintashta and Andronovo. They shared much of the older steppe ancestry, alongside an additional contribution related to European farmers, as shown by Allentoft et al. (2015) and Narasimhan et al. (2019). The DNA measures genetic relationships; identifying the languages these people spoke also requires historical evidence.

By the Late Bronze Age, Kumar and colleagues (2022) found individuals in Xinjiang with strong Andronovo/Sintashta-related affinities. Archaeology gives the movement a more tangible setting. Adunqiaolu, near the western edge of Xinjiang, has an Andronovo-affiliated settlement and cemetery, with dated contexts extending approximately from 1900 to 1300 BCE. Figure 1 places these connections alongside the earlier expansion.

Figure 1. Two steppe expansions into western China

The mummies complicate the story

The famous Tarim mummies might seem to offer a simple continuation of those migrations. Their genomes tell a different story. The earliest sampled Tarim individuals, dating to about 2100–1700 BCE, carried a deeply rooted local genetic profile. Zhang et al. distinguished them from the Afanasievo-related people farther north. Xinjiang already contained very different populations before the later steppe arrivals.

That regional diversity continued into the Iron Age. Kumar and colleagues identified Saka-related individuals and proposed a connection with the introduction of Khotanese, an Iranian language. Yang and colleagues’ 2026 study then drew a sharper east–west distinction: Andronovo-related ancestry in the western Tianshan, with Afanasievo-related ancestry persisting in eastern groups.

The published models make the contrast visible. In three selected western Iron Age models, the direct Andronovo coefficients are 56.2% at Balesa and 60.5% and 59.5% in two Nileke groups. The full period maps, with Andronovo and Afanasievo coloured separately, are in Figures 6A–6E below. These are published model coefficients with different source sets, so they should be read as a regional comparison rather than a single ancestry curve.

Upgrade to see where the Andronovo connection strengthened and what changed after the Iron Age, with the maps and full results from 370 ancient Xinjiang genomes.

Read more

In Rome and the Northern Catch Up Reanalysed, I used EA4, a polygenic score for educational attainment, based on genetic associations with years of schooling in modern people, to test an Iron Age Italian lead and later northern convergence in a much larger European sample. The northern side no longer meant mostly Denmark: it included six continental countries. EA4’s weights came from a conventional genome-wide association study comparing people from different families. With that score, I found a higher early Italian score and a relative northern rise over the full time span. What happens if the genetic weights are estimated within families instead?

For education, the choice of weights matters unusually much. A conventional genome-wide association study compares unrelated people, so its associations can reflect a person’s own genetic effects alongside parental genes that shape the home and educational opportunities, patterns of mate choice, and residual ancestry differences. Kong and colleagues found that even parental alleles a child did not inherit predicted the child’s education. In Okbay and colleagues’ study, accounting for parental scores roughly halved the education score’s association with education. Tan and colleagues instead estimated genetic effects from comparisons within families, making the resulting weights less vulnerable to those indirect and population-level influences.

That gain comes with a loss of predictive accuracy. In a modern benchmark discussed by Tan and colleagues, a direct-effect education score explained 1.8% of the variation in education, compared with 5.3% for a population-effect score.

Below, I ask how closely the family-based score tracks EA4 in the same ancient people; whether Republican-era Rome still stands apart; how Italy and Northern Europe compare across five periods; and whether their later convergence reflects a northern rise, an Italian decline, or both. I also test how much those answers depend on ancestry adjustment and on the studies that supplied the ancient remains.

Upgrade to read which parts of Rome’s early lead and the northern catch-up survive when the genetic weights come from within families.

Read more

That unevenness was central to Does Ancient DNA Track Human Progress, or Just Time?. The question was whether genetic scores linked to education were associated with how far a society had moved through the archaeological sequence. Different societies reached farming, metalworking and later technologies at different times. Around 650 BCE, for example, one burial in Latvia is classified as Late Bronze Age while another in Italy is classified as Iron Age. Did genetic scores differ between societies that had reached different stages at roughly the same time? That is a more demanding test than simply comparing older populations with more recent ones.

There is a human question behind the statistical one. Farming requires people to plan around delayed returns. Metalworking depends on skills that must be learned, remembered and transmitted. More elaborate societies bring new demands on coordination and accumulated knowledge. It is reasonable to ask whether traits associated with reasoning, learning and future orientation played a part in those changes. Ancient DNA gives us a way to investigate that question, although genetic scores measured in modern populations remain an indirect guide to people who lived thousands of years ago.

The evidence has now improved in a fairly literal way. I went back to the papers and their supplementary files, looking for the archaeological descriptions behind the genetic records. That expanded the source-supported European sample well beyond the initial 1,730 records. It also exposed a problem with counting skeletons: a cemetery with a hundred sequenced burials may still supply only one archaeological phase.

I grouped individuals by site and by the archaeological phase identified in the original study, then treat each group as one observation.. I also used newer genetic scores for fluid intelligence, alongside educational attainment and delay discounting. That makes the comparison more interesting, because education, reasoning ability and willingness to wait are related ideas without being interchangeable measurements.

The intelligence result remains worth pursuing. But the latest checks change the explanation of why some models looked stronger than others. The new results suggest that intelligence-linked genetic scores capture something beyond the passage of time. How much survives closer scrutiny?

One burial is classified as Late Bronze Age, another as Iron Age. Both have a recorded date estimate of 650 BCE (Figure 1). Can ancient DNA tell us anything about that difference once the calendar is held constant?

Figure 1. Two individuals at the same recorded date in different archaeological periods

Upgrade to explore the larger source-study dataset, compare intelligence, education and patience in the new effect plots, and follow what happens when I average within archaeological phases and allow the relationship between date and archaeological period to follow a smooth curve.

Read more

I also ask whether the shift towards lighter-skin genetics happened faster in places with more or less UVB. Adding an “interaction” between time and UVB lets the rate of change differ with local sunlight. A positive interaction means that people living in higher-UVB environments shifted towards lighter-skin genetics faster; a negative interaction means the shift was faster where UVB was lower. Farming remains in the model, so this comparison accounts for how long agriculture had been established locally.

The proposed biological model balances the benefits of sunlight against ultraviolet damage. Melanin absorbs UV and may help protect folate, which is important for reproduction and early development. Yet UVB also starts vitamin D production in the skin, and melanin can limit the radiation available for that process. Where UVB is scarce, lighter skin could improve vitamin D production. UV exposure also raises skin-cancer risk, although its role in ancient selection is less certain because many cancers occur after the main reproductive years. Jablonski and Chaplin (2010) discuss this evolutionary balance.

Farming could change that balance through diet. If a shift towards cereals reduced vitamin D intake from foods such as oily fish, people would become more dependent on producing it in their skin. Under those conditions, variants associated with lighter skin could become advantageous, particularly where UVB was already limited.

Farming may also have changed how much sunlight reached people’s skin. Settled life brought more indoor activities, potentially making vitamin D harder to obtain—although working the fields still meant plenty of time outdoors. Together with changes in diet, this gives us a reason to ask whether farming helped favour lighter skin.

That farming signal turns up again in East Asia when I use a genetic score developed from East Asian data and account for relatedness between the ancient individuals. In both Europe and East Asia, the scores also point towards lighter skin over time. But sunlight throws a curveball: in East Asia, higher UVB goes with a higher score for lighter skin—the reverse of the European pattern.

Different genes can lead to lighter skin

A polygenic score weights and adds genetic variants associated with a trait in modern people. A skin-colour score that works in Europe is not automatically the best instrument for studying East Asian prehistory. Similar pigmentation can be reached through partly different genetic routes.

Kim et al. (2024) measured skin colour in 48,433 East Asians. Their study found substantial differences from European pigmentation genetics. Some genetic variation associated with lighter skin is shared, including an allele near KITLG, while variants involving SLC24A5 and MC1R show population-specific patterns. Their East Asian light-skin score explained approximately twice as much variation when applied to East Asians in UK Biobank as when applied to Europeans.

I use the same score as in my European analysis, alongside one developed from East Asian data.

The panels capture different parts of pigmentation genetics. A contrast between their ancient trends is consistent with the genetic differences described by Kim and colleagues, although this comparison alone cannot establish why their results differ. Neither panel reconstructs a person’s exact complexion.

The long trend is towards lighter pigmentation

First, consider the broad chronological pattern before asking what farming or sunlight might explain. I use gamma, γ, for the change in the lighter-skin score as time moves towards the present. Positive gamma means an increase in variants associated with lighter skin pigmentation.

Figure 1 shows the result. A genetic relationship matrix (GRM) accounts for genome-wide similarity between people. With that adjustment and study effects, European Skin587 has an overall gamma of +1.20 score standard deviations per 10,000 years [95% interval: +1.00, +1.40]. East Asian Kim has a gamma of +0.78 [95% interval: +0.21, +1.34], with p=.007. Both directions are also positive without GRM adjustment (Table 5).

Skin587 gives a different answer in those same East Asian genomes: its trend is uncertain, with an interval spanning zero. The East Asian score detects a trend towards lighter skin pigmentation that Skin587 does not clearly capture. The choice of score therefore matters to the historical story.

Figure 1. Overall trends in lighter pigmentation scores

Figure 2 puts those trends back into the individual genomes. Each dot shows a person’s skin score after I account for differences between studies and, in the full sample, imputation uncertainty. I leave the dates unchanged. Europe uses Skin587 and East Asia uses Kim’s 15-variant score. The fitted lines rise towards the present in both populations, with and without genetic relatedness adjustment; the orange points and lines show the alternative genotype-confidence screen.

Figure 2. Adjusted lighter-skin scores through time

There are 6,692 European genomes and 1,384 East Asian genomes in the main comparison. Europe extends back approximately 17,100 years; the East Asian sample extends back approximately 9,900 years. The comparison follows the same modelling approach within each population, with different geographical and chronological coverage.

Below, I test whether farming helps explain the shift towards lighter skin on both sides of Eurasia, and investigate why the East Asian sunlight result turns the familiar evolutionary story on its head.

Read more

A language can spread with migrants who mix with local populations along the way. The resulting communities can share a language without forming a tight genetic cluster. Their proposed expansion could therefore help explain the diversity I found.

The regional pattern in my earlier results

I compared ancient and later populations using the Allen Ancient DNA Resource. One set of tests measured which of two reference groups a population shared more genetic history with; another measured overall genetic distances. The Celtic references included Iron Age Britain, Gaulish and Central European La Tène groups, and Hallstatt. [3]

The clearest continuity on the Celtic side was local. The medieval Irish and Pictish groups in my comparison leaned towards Iron Age Britain over Iron Age Scandinavia. They did not show the same clear preference for every continental Celtic reference. The later Saxon and Viking comparisons gave a more consistent Scandinavian pull. [3]

Small ancient samples, local drift and uneven sampling can also exaggerate genetic distances, as I noted in the earlier post. The stronger evidence for local continuity came from the directional allele-sharing comparisons. [3]

What the migration study adds

McColl and colleagues use shared stretches of DNA to reconstruct finer population connections. Their preprint links a westward ancestry expansion to Knovíz, an archaeological group in Bohemia within the Late Bronze Age Urnfield world, and traces connections into later Hallstatt and La Tène contexts. [1]

The accepted abstract distinguishes three relevant movements: Bell Beaker-associated dispersal after about 2450 BCE, an earlier southwestern connection into Britain around 1850–1250 BCE, and the Knovíz/Urnfield-related expansion around 1250–850 BCE. It favours the last as the most economical explanation for Celtic dispersal. Figure 1 locates those broad connections. [2]

Figure 1. Three movements discussed in the Celtic homeland debate

Figure 2 separates the older Bell Beaker transformation from the later Bronze Age movements. The arrival of steppe-related ancestry predates the expansion favoured for Celtic dispersal.

Figure 2. Three demographic episodes that could matter for Celtic

How the DNA segment method works

DNA is inherited in stretches, which recombination breaks into smaller pieces over generations. A matching stretch inherited from a common ancestor is an identity-by-descent (IBD) segment. Longer shared segments generally point to more recent common ancestry.

For each individual, the authors summarise how much DNA is shared with each genetic cluster, adjusting for the number of people sampled in that cluster. They then fit this sharing profile as a mixture of profiles from candidate source populations to estimate their relative ancestry contributions. The model uses sharing across the wider dataset, rather than just a single pair of genomes. [1]

A shared migration without a uniform population

A migration can contribute to several populations without making them each other’s closest relatives. My comparisons measured the relationships among later populations; they could not determine which earlier movement brought a language into each region.

I described Hallstatt and La Tène as pointing to the original Celtic homeland. McColl and colleagues favour an expansion that began earlier, in the Late Bronze Age Urnfield world. If it carried Celtic, the language was already spreading before those later archaeological traditions. Their association with Celtic-speaking regions cannot, on its own, locate the first Celtic speakers. [1, 3]

Why Britain still makes me cautious

In Large-scale migration into Britain during the Middle to Late Bronze Age, Nick Patterson and colleagues identified substantial continental migration involving people genetically similar to ancient individuals from France. They proposed a plausible route for early Celtic speech. Their results place an important ancestry change in southern Britain around 1000–875 BCE, incorporating arrivals over preceding centuries. [4]

Lara Cassidy and colleagues’ Continental influx and pervasive matrilocality in Iron Age Britain then argued for further continental influence during the Iron Age, varying between regions. They discuss how later contact might have affected Celtic languages already spoken around the Channel. [5]

Celtic could have developed further east and reached Britain through France. Later migrants could then have reinforced an existing language or introduced a different variety. Britain’s repeated continental contacts make it difficult to assign its Celtic speech to one arrival.

The local continuity in my Irish and Pictish comparisons is compatible with that history: much of their ancestry could have persisted while speech changed through continental contact.

Where the language question remains open

The preprint argues sharply against western and central origin models. The accepted abstract explicitly accommodates an earlier southwestern movement compatible with Atlantic formation scenarios, while still preferring the later eastern expansion. [1, 2]

I find the eastern expansion a plausible vehicle for Celtic dispersal. Establishing that its migrants spoke Celtic still requires evidence about language. They could have carried it west, adopted it along the way, or spread a later variety among communities already speaking Celtic.

Methods and source notes

The earlier post was published on 11 May 2026; the McColl preprint dates to 2025. The earlier comparisons have not been rerun. [1, 3]

Public preprint v2 analyses 4,587 individuals: 578 newly generated individuals from PRJEB87274 and 4,009 previously published individuals. The 578 belong to an accompanying study’s larger set of 712 new genomes. The accepted abstract also reports 4,587 individuals. Its complete manuscript is embargoed; detailed methods here are attributed to preprint v2. [1, 2]

The earlier analysis used allele-sharing tests (f4 statistics) and Hudson genetic distances (FST); the migration study relies on shared inherited DNA segments for its principal fine-scale modelling. [1, 3]

Dates in Figures 1 and 2 are rounded from calibrated years before 1950: 4400 BP to about 2450 BCE, 3800–3200 BP to about 1850–1250 BCE, and 3200–2800 BP to about 1250–850 BCE. They are approximate demographic periods, not confidence intervals or direct dates of language change. Map arrows show schematic regional connections rather than exact routes; the later window does not date every regional endpoint. [2]

References

[1] McColl, H., et al. (2025). Tracing the Spread of Celtic Languages using Ancient Genomics. bioRxiv. https://doi.org/10.1101/2025.02.28.640770. Version used here: public full text v2. Read source

[2] Willerslev, E., McColl, H., Kroonen, G., Kristiansen, K., & Sikora, M. (2026). Tracing the spread of Celtic languages using ancient genomics. Cambridge Apollo accepted-version record, repository DOI 10.17863/CAM.135565; journal listed as Nature. Public abstract and metadata only; author listing follows the repository record. Read source

[3] Piffer, D. (2026, May 11). Were Celts and Germanics Genetically Different? PifferPilfer. Read source

[4] Patterson, N. et al. (2021, online publication). Large-scale migration into Britain during the Middle to Late Bronze Age. Nature. Print edition: 601, 588–594 (2022). Read source

[5] Cassidy, L. M., Russell, M., Smith, M., et al. (2025). Continental influx and pervasive matrilocality in Iron Age Britain. Nature, 637, 1136–1142. Read source

David Reich’s new preprint, Hypothesis: A modern human range expansion ~300,000 years ago had a critical role in Neandertal origins, proposes a history in which a successful early human population expanded into Europe and Africa, mixing with different local populations in each region. The African side of that reconstruction contains the most striking possibility: descendants of lineages that separated around 1.5 million years ago came together roughly 300,000 years ago. [1]

That leaves about 1.2 million years between separation and reunion. If this reconstruction is correct, our ancestry records successful reproduction across an exceptionally deep divide. It raises questions about fertility and, more speculatively, attraction between people whose populations had followed different evolutionary paths.

The paper is a hypothesis, released as a preprint without peer review. Its proposed history deserves attention precisely because the biological implications are substantial and several parts remain uncertain. This proposed early expansion is compatible with the much later Out-of-Africa expansion that supplied most of the ancestry of living non-Africans. [22]

The mixture behind the headline

Reich begins with a puzzle about Neanderthals. Across most of their genomes they are closest to Denisovans, yet their mitochondrial DNA and Y chromosomes share more recent ancestry with modern humans. These two markers follow the maternal and paternal lines, respectively. Earlier research inferred that an early modern human lineage contributed both markers, alongside about 3–7% of the remaining Neanderthal genome. [1]

He proposes that expanding bands repeatedly absorbed archaic Europeans. The bands could continue spreading while local people supplied most of their eventual ancestry. His simulations explore how particular mating patterns or reproductive advantages might preserve the expanding population’s maternal and paternal lineages.

For Africa, Reich connects this scenario to a separate study by Trevor Cousins, Aylwyn Scally and Richard Durbin. Their model infers two ancestral populations that separated about 1.5 million years ago and mixed around 290,000 years ago, contributing roughly 79% and 21% of living people’s ancestry. The majority lineage is closer to Neanderthals and Denisovans. [2]

Figure 1 shows Reich’s proposed assignment of that mixture: approximately 80% from the early modern human lineage and 20% from the deeply separated African archaic lineage. The “early modern” label refers retrospectively to an ancestral branch extending back before the appearance of anatomically modern humans.

Figure 1 Two proposed mixtures with very different ancestry proportions

Under this reconstruction, the deep African mixture contributed to the ancestry of all living humans, including Europeans and East Asians. Reich places the proposed earlier source around northeast Africa or the Near East; its precise location remains unresolved. The later Out-of-Africa expansion follows a separate, much more recent stage of this history. [1]

A deeply separated African lineage may have supplied about 20% of our ancestry. Neanderthals left roughly 2% in many living non-Africans. Why did the deeper divide leave the larger genetic legacy?

Upgrade to read the Khan–Tabin debate, follow the human branches and migrations in a second diagram, and explore what fertile animal hybrids reveal about reproductive barriers—and what ancient interbreeding might imply about attraction.

Read more

In Does Ancient DNA Track Human Progress, or Just Time?, I asked whether genetic scores associated with human capital track the development of complex societies after chronology is taken into account. This follow-up shifts the emphasis towards a different trait: patience.

A farmer has to wait for a harvest. Someone building an institution has to think beyond the next meal. That makes delayed gratification a plausible contributor to social complexity. Here I examine hierarchy, one aspect of that broader concept. The first results suggest a positive association between a genetic score oriented towards patience and both administrative and settlement hierarchy. The evidence is preliminary, and its strength varies across models.

The link could be direct, if patience helps people sustain specialised institutional roles whose rewards arrive later. It could also be indirect: greater delayed gratification might support investment and specialisation, producing more complex societies that develop additional layers of administration and settlement. These are possible mechanisms to investigate. The present associations do not tell us which pathway operates, whether either is causal, or whether other historical processes account for the pattern.

The crucial test is whether patience contributes anything beyond intelligence and education. I put all three scores into the same models, then checked which results survive stricter genetic-quality filters and the removal of whole regions. Below I show the full comparison, the regional stress tests and what happens when I ask the scores to predict history.

Read more

A seed and an examination paper belonged to very different chapters of East Asian history. Farming arrived thousands of years before the schools and court traditions that we associate with Confucianism. Yet an ancient genome from a later century sits after both events. That makes it surprisingly easy to mistake one long genetic trend for evidence of two different causes.

In Did Farming Make Us Smarter?, I found a change in an education-related genetic score around local crop arrival. In Did Confucianism Select for Education?, I found another near regional Confucian milestones. Now I put both dates into the same analysis and add a score linked to fluid intelligence. If the Confucian bend was farming in disguise, it should shrink. It barely does. The farming signal, meanwhile, tells a more complicated story.

Read more

One recurring idea is a feedback loop: the conditions that help a society expand can be changed by its own success. Security and prosperity alter incentives, weaken older restraints or undermine cooperation, allowing a new ruling group to emerge. These traditions offer different accounts of political and dynastic turnover; they do not imply that every society follows the same timetable.

Edward Dutton and Michael Woodley of Menie developed a genetic account of civilization cycles in At Our Wits’ End. They propose that selection favouring intelligence contributes to a civilization’s rise, while greater prosperity relaxes survival pressures and changing fertility patterns eventually reverse the direction of selection. In this account, the achievements of a civilization can weaken the conditions that helped produce them. Over generations, declining cognitive capacity would make innovation and the maintenance of technological achievements harder. Their discussion of Rome and other civilizations applies this rise-and-decline argument to the ancient world as well as the modern West. A renewed rise after decline would require conditions favouring those traits to return. These are proposed mechanisms to test against historical evidence. Figure 1 illustrates this proposed feedback and the conditions a renewed rise would require.

Figure 1. The proposed civilization cycle

Modern Icelandic research by Kong and colleagues found that people with higher education-associated genetic scores tended to have children later and have fewer children overall. That observation makes reproductive differences a plausible mechanism for changing score averages. Whether a comparable process operated during ancient expansion, imperial decline and recovery requires historical evidence.

Ancient DNA gives us a way to ask whether anything resembling this sequence actually happened. In Rome Was Ahead. Then the Barbarians Caught Up, I followed education-associated genetic scores through the Roman transition in Italy and northern Europe. That raised a larger question: was the Roman-era pattern one stretch of a much longer European rise, decline and recovery?

The Roman comparison also raises a regional question: how much does the European pattern differ from place to place? Different political and demographic histories could leave different genetic trajectories. I test that directly by asking whether allowing regional differences improves on one European trend.

The larger question calls for a European view, extending back into prehistory. I compare two polygenic scores (PGS): one associated with educational attainment (EA4), and another associated with fluid intelligence (FI), or reasoning ability.

From the Holocene to the Black Death

The historical hypothesis has three landmarks. The first is the beginning of the Holocene, around 9750 BCE, or 11,700 years before present. The second is 27 BCE, the beginning of the Roman imperial period. The third is 1350 CE, around the Black Death. These dates give the theory a concrete sequence to confront: a long rise, a decline across the imperial and medieval centuries, then a renewed rise.

The 1350 date comes from the question I explored in Did the Black Death Select for Intelligence?. Here the plague hypothesis becomes part of a longer story. Instead of asking only whether scores rose after the pandemic, I ask whether that rise followed an earlier reversal.

At each date, the fitted line can change direction or speed without making a sudden jump. This is what a statistical “hinge” does. The dates are historical reference points chosen for the hypothesis; the DNA still has to tell us whether the proposed rise and fall are present.

Figure 2 puts the two scores side by side.

Read more

A Tang tomb epitaph could record a man’s ancestors, his father’s office, and how far he climbed. In a study of 3,640 of these inscriptions, passing the imperial examinations became an increasingly strong predictor of an elite career. Aristocratic pedigree lost some of its force. The inscriptions show how the rewards for learning changed.

The institutional history predates this series. Ping-ti Ho studied examinations and social mobility; Benjamin Elman traced their cultural history. The biological conjecture has its own history: Pan Guangdan read Confucianism in sociobiological terms, and Peter Frost later proposed that examination culture might exert selection on cognition. Those proposals preceded an ancient-DNA test.

A career is not a family tree. Still, if skill at learning influenced marriage and the number of children people raised, centuries of changed rewards might leave a trace in descendants’ DNA. Could ancient genomes show one?

That was the question behind my earlier post, “Did Confucian institutions select for educational attainment?” An education-linked genetic score appeared to rise faster after regional dates chosen to mark the spread of Confucian institutions. I have now tested a score linked to fluid intelligence in the same 1,384 people and re-examined the dates behind that apparent turn. The result is intriguing, but one study carries more of the evidence than I first realised.

Read more

Thirteen years ago, we finally had three genome-wide-significant genetic hits for education. Three! I used their frequencies across 14 populations from the 1000 Genomes project in my 2013 paper to investigate whether education-associated alleles followed a pattern across populations that tracked IQ. [1, 2]

Since then, the GWAS samples have grown enormously. EA3 arrived in 2018, followed by EA4 in 2022. And now, in 2026, we finally have a much larger study of fluid intelligence. [3, 4, 5] I wanted to see what happened when those original three SNPs met their successors.

There is also a new Norwegian education GWAS in the mix. Kvalvik and colleagues published it in September 2026, using the Norwegian Mother, Father and Child Cohort Study, usually shortened to MoBa. Here, “Norway” means their years-of-education score, based on 120,527 genotyped adults. The paper also studied educational milestones and twins; the score used here comes from the adult MoBa GWAS. [6]

For a comparison with the older intelligence results, I also include the Pan-UK Biobank fluid-intelligence score. Its weights come from the European-ancestry analysis of the earlier UKBB test data, covering 135,088 participants. The Pan-UKBB resource paper appeared in 2025. [7]

They agree remarkably well. Across 46 population samples, the old score correlates 0.939 with EA3, 0.779 with EA4 and 0.874 with the new fluid-intelligence score. It also correlates 0.810 with the population IQ estimates, across 39 samples. After thirteen years, those three variants are still keeping up.

Meet the GWAS

Table 1A puts the education studies on a timeline, and Table 1B does the same for intelligence. The growth is quite something: from three education hits in 2013 to thousands. The sample sizes belong to the discovery studies, not to the population samples in our scatterplots.

Table 1A. The education GWAS behind the scores

Table 1B. The intelligence GWAS behind the scores

The scores used in this post retain EA1’s 3 SNPs, 2,589 from EA3 MTAG, 2,933 from EA4, 42 from MoBa, 69 from Pan-UKBB, 454 from new FI and 502 from FI plus COGENT.

Back when three SNPs were a big deal

Rietveld and colleagues studied 126,559 people to find those three hits. [2] A genome-wide association study, or GWAS, searches for genetic variants associated with a trait. The three SNPs passed the significance threshold for years of education or college completion. In 2013, that was enough to get started.

A polygenic score adds up the associated alleles, giving each one a weight based on its estimated effect. With population allele frequencies, we can calculate the average score for each sample. For this comparison, I took the original three variants and weighted them using Rietveld’s published effects on years of education. I call that score EA1.

My original paper examined the three significant SNPs as well as a broader set of ten variants. The main 1000 Genomes analysis covered 14 populations, with another analysis of 11 HapMap populations. ALFRED extended the education analysis to 50 populations using seven markers, including proxies for all three lead SNPs. [1]

Eight of those ALFRED population groups are represented in today’s full 51-sample roster: Yoruba, Italians/Tuscans, Dai, Han Chinese, Japanese, Koreans, Mongolians and Papuans.

Three SNPs meet a few thousand

Start with Figure 1. Each dot is a population sample, with EA1 on the horizontal axis and the newer education score on the vertical axis. The EA3 plot is striking: r = 0.939. A score built from three variants follows one containing 2,589 variants remarkably closely.

EA4 agrees too, at r = 0.779, although the points are more spread out. A larger GWAS does not automatically give the highest correlation in every comparison. There is also a difference in what these scores measure: the EA3 panel combines education with cognitive and mathematical traits, whereas EA4 comes from the later education analysis.

Figure 1. Three original SNPs versus EA3 and EA4

None of the three original SNP identifiers appears in either of these newer score files. That rules out the simplest explanation, that we are just adding the same three SNPs back into a bigger sum. Nearby variants can still be inherited together, so some of the newer markers may track the same genetic regions.

Now intelligence joins in

The new paper by van den Berg, Malawsky, Huang and colleagues gets much more out of UK Biobank by combining measured fluid-intelligence tests with estimates for missing scores. Fluid intelligence means reasoning and solving unfamiliar problems. Their combined analysis reaches 455,402 people; adding the Cognitive Genomics Consortium, COGENT, brings it to 490,700. [5]

Figure 2 is the comparison I was most interested in. EA1 correlates 0.874 with the new UKBB intelligence score and 0.846 with FI plus COGENT. The first three education hits still track a much larger intelligence GWAS thirteen years later.

Figure 2. Three original SNPs versus the new intelligence scores

The FI scores use reconstructed panels of significant variants: 454 SNPs for FI and 502 for FI plus COGENT.

UK Biobank was not in the first study

This is where the new FI result becomes especially interesting. UK Biobank was absent from Rietveld’s 2013 study. I checked the original cohort list, and the 2016 education paper explicitly used UKBB as an independent replication sample. [8] Another UKBB study that year also replicated all three original education hits, including two for verbal–numerical reasoning. [9]

So the UKBB-only FI result is an independent replication with respect to the GWAS discovery sample. The genetic associations were estimated in people from a cohort that did not contribute to EA1. Yet the population scores correlate 0.874.

Adding COGENT changes that answer. Its GWAS includes Helsinki, Lothian 1936 and the Minnesota twin and family cohort, which also contributed to EA1. [2, 10] An earlier COGENT paper actually removed Helsinki and Lothian to avoid overlap in its independent comparison. [11] FI plus COGENT therefore has some cohort overlap with the first study. The published lists do not tell us exactly how many individuals were shared.

EA3 and EA4 also reuse earlier education cohorts, and UKBB contributes to both. Their agreement is useful evidence, with that shared history taken into account. UKBB-only FI gives us the clearer comparison with an independent discovery cohort.

And EA2? I started with the EA3 and EA4 population scores from the previous FI post, so I have not calculated an EA2 score here. The 2016 study was an important step between EA1 and EA3, expanding the discovery sample to 293,723 people and finding 74 significant loci. Its absence from these plots reflects the scores available for this comparison. [8]

The old hits still fit the newer scores

Table 2 puts all seven scores and population IQ together. I also included the Norwegian MoBa education score [6] and the older Pan-UKBB FI score. [7] The Norwegian score tracks EA1 very closely, at r = 0.895. Its correlation with EA3 is even higher, 0.948.

Table 2. The population correlation matrix

Figure 3 shows EA1 against the Norwegian MoBa education score and the older Pan-UKBB FI score. Pan-UKBB has the weakest correlation with EA1 among these six comparisons, at 0.390. The new FI result is much closer to the pattern in the original education hits.

Figure 3. Three original SNPs versus older FI and Norwegian education

For IQ, we need to compare the scores on the same populations. Table 2 uses every available pair, so the number of samples varies. Table 3 fixes that by using the same 39 samples for all seven scores.

Table 3. IQ correlations on the same 39 populations

New FI leads at 0.870, followed by EA3 at 0.855 and FI plus COGENT at 0.849. Our three-SNP EA1 score gives 0.810, and Norway 0.792. EA4 is lower, at 0.745, while the older Pan-UKBB FI score gives 0.313.

The newer FI scores fit these IQ estimates much better than the older one. EA1 does surprisingly well for such a tiny score. The ordering is descriptive: I have not tested whether each gap in Table 3 is statistically significant, and these correlations concern population averages rather than prediction in individuals.

The IQ association is still there

Figure 4 returns to the question behind my 2013 paper. The correlation between EA1 and population IQ is 0.810 across 39 samples. Giving all three increasing alleles equal weight produces almost the same result, 0.816. The pattern hardly changes when we remove the effect-size weights.

Figure 4. Three original SNPs versus population IQ

These IQ estimates were assembled before this study, mainly from Lynn and Vanhanen, David Becker’s National IQ dataset version 1.3.5 and other publications. They combine national and subgroup estimates collected in different ways. That is worth bearing in mind when reading the plots.

I also checked the other IQ measures used in the previous post. Table 4 shows that the Parra and Kirkegaard compilation gives 0.822, or 0.803 when the Ashkenazi estimate is added. [12] The online BRGHT and IIQTEST measures give lower correlations, 0.698 and 0.688. They cover different populations and rely on self-selected test takers, so we should expect some variation.

Table 4. EA1 against the five IQ measures

For the Parra and Kirkegaard comparison, the U.S. estimate goes with a demographic composite of the African-American, Mexican-American and Utah White scores. Mexico uses Mexico_overall. I retain 110 for the Ashkenazi estimate, based on the historical data discussed in Dunkel and colleagues. [13]

The Dutch sample is an unusual point in several plots. Removing it raises the EA1–IQ correlation from 0.810 to 0.833, so it is certainly not responsible for making that association look strong.

What ancestry changes

There is a substantial ancestry pattern in these plots. To see how much of the agreement remains within the broad groups, I subtracted each group’s average from both measures and correlated what was left. Table 5 gives the results.

Table 5. How much agreement remains within broad ancestry groups

EA1 still correlates 0.665 with EA3 and 0.465 with new FI after that calculation. Its correlation with IQ falls much more sharply, from 0.810 to 0.205. Much of the raw IQ association therefore lies between the broad ancestry groups. Those groups are coarse and some are small, but the drop is substantial and deserves attention.

In 2013, I proposed using population allele-frequency patterns to investigate recent polygenic selection. [1] The newer scores strengthen the evidence that the pattern recurs across GWAS. Establishing its cause takes more than these correlations: shared ancestry and the way genetic markers track one another can also produce agreement, while environmental differences affect the IQ estimates.

What happens when we account for LD

There is a catch with GWAS hits: the SNP we score may simply be travelling with the variant that affects the trait. We call it a tag SNP. Its estimated effect then partly reflects how well it tracks that causal variant in the discovery sample. This tendency for alleles to be inherited together more often than chance predicts is linkage disequilibrium, or LD.

Imagine a tag allele that usually accompanies an intelligence-increasing causal allele in Britain. Elsewhere, the same tag might accompany that allele much less often, or even accompany the opposite allele. Applying the British weight unchanged can then overstate its contribution or get its direction wrong, even if the causal effect itself is identical. A tag that works well in Britain can be a poor guide elsewhere. [14]

For population averages, there is another wrinkle: a difference in tag-allele frequency need not mirror a difference in causal-allele frequency. Summing tags with British weights can therefore distort comparisons between populations. Different LD patterns can also reduce individual prediction accuracy. These are related concerns, and an adjustment that changes population-score correlations still needs testing in individuals.

For each SNP, I compared its local LD profile in Britain with its profile in each of the other 25 populations in 1000 Genomes. The correlation between those profiles became a population-specific multiplier on the GWAS effect. This is an experimental weighting rule, rather than an established correction that makes a score “LD-free”.

One detail matters for reading Figures 5 and 6: these weighted scores are deviations from the British reference, GBR. For each population, I subtract a matched British score calculated with exactly the same retained SNPs and the same LD weights.

A positive value means a higher weighted sum than the matched British reference; a negative value means a lower one, and zero means no net weighted difference. These are score units, not IQ points. Because the weights and retained SNPs can differ between populations, each deviation uses its own scoring rule. Subtracting the matched reference makes the comparison explicit; it does not turn these into one identical score definition for everyone.

Our weighting asks how similar each SNP’s surrounding LD pattern is to Britain’s, using nearby variants as a practical proxy for the unknown causal links. It does not identify the causal variant or directly measure how well the score SNP tags it. That is why I treat this as an exploratory sensitivity analysis.

The three old hits are still very much in the game. Figure 5 shows EA1 correlating 0.874 with LD-weighted FI and 0.906 with LD-weighted FI plus COGENT, across 25 population samples. Table 6A adds the other scores: the weighted FI plus COGENT score correlates 0.943 with EA3 and 0.909 with EA4. These are strong agreements, although they use a smaller population set than the earlier figures.

Figure 5. The first three education hits versus LD-weighted intelligence

Table 6A. LD-weighted intelligence scores meet the older scores

Does the weighting improve the fit to IQ? Mostly, it lowers the correlations; FI plus COGENT against BRGHT is the small exception. Figure 6 shows correlations of 0.740 for FI and 0.824 for FI plus COGENT with the original population IQ compilation. Table 6B puts the weighted and original versions side by side, using exactly the same samples for each comparison.

Figure 6. LD-weighted intelligence scores versus population IQ

Table 6B. Does LD weighting improve the IQ correlations

The corrected IQ_P&K comparison is less flattering to the weighting: FI falls from 0.815 to 0.771, and FI plus COGENT from 0.814 to 0.804. That comparison keeps the U.S. demographic composite and counts it once. Mexico_overall has no matching phased reference in this subset, so Mexico contributes no national point here. MXL remains a component of the U.S. composite.

So LD weighting leaves the strong agreement with EA1 intact, especially for FI plus COGENT. It does not give us a general improvement across the IQ measures. The original and weighted versions belong side by side; the weighted score still needs validation as a method of improving prediction.

What stands out to me is how well those first three hits have held up. EA3, EA4 and the Norway educational attainment GWAS agree with them. The new UKBB intelligence GWAS gives us a strong result from an independent discovery cohort. Thirteen years and vastly larger samples later, the original population pattern is still there. That is a satisfying result to come back to.

Methods appendix

The collection contains 51 named population samples. EA1 is calculated as twice the sum of the increasing-allele frequencies multiplied by the published combined-sample EduYears effects: A at rs9320913, +0.101; G at rs11584700, +0.095 after reversing the reported allele; and T at rs4851266, +0.082. These use the three-decimal precision printed in Rietveld’s supplement. The three SNPs reached genome-wide significance across the EduYears and College analyses, rather than all three for EduYears alone. [2]

In the 2013 paper, Table 4 examined the three significant SNPs. The broader ten-variant analysis produced a first rotated principal component correlated 0.90 with IQ across 14 populations. The present weighted three-SNP score and its 39-population IQ comparison use a different calculation and larger population set. [1]

EA1 is complete for 46 samples. CHINA HAN, HONG KONG, KOREAN, MONGOL and TURKISH lack a recoverable frequency for rs9320913 after searches of the available original releases and coordinate checks. They remain in the data with a missing score. Partial two-SNP scores and imputed frequency values were not substituted.

The frequency refresh replaces the whole Japanese population block with ToMMo 61KJPN v20250616 (61,424 people, GRCh38, autosomes 1–22). For Taiwan, all requested frequencies are replaced by the Taiwan View Illumina WGS release (1,492 people, GRCh38): 7,155 exact marker allele pairs were recovered across the project panels; 835 remain unavailable from the public pages. Missing current-release frequencies are not filled with older Taiwan values. Other population frequency data, the IQ compilations and the GWAS effect estimates are retained. Marker intersections are recalculated separately for each score, so expanding or changing a panel can change scores even in populations whose allele frequencies were unchanged.

The EA3 MTAG panel combines education, cognitive performance and mathematical phenotypes. The refreshed 51-population frequency intersection retains 2,589 of its 3,269 candidate variants and 2,933 of 3,925 EA4 candidates. Norway uses 42 MoBa EduYears lead variants, accession GCST90832184; Pan-UKBB FI uses 69 SNPs. These are fixed restricted panels, rather than full genome-wide prediction scores.

The authors reported 550 FI and 620 FI plus COGENT lead SNPs using FUMA. The resources checked here—the supplementary workbook, public summary-statistics deposit and code repository—did not contain separate FUMA lead lists. I reconstructed panels from the summary statistics using P < 5 × 10⁻⁸, PLINK clumping with a European reference at r² < 0.1 within 1,000 kb, and the complete 51-population frequency intersection. This retained 454 FI and 502 FI plus COGENT SNPs. None of the exact three EA1 identifiers appears in EA3, EA4 or either new FI panel. Nearby correlated variants may still tag the same regions.

UKBB was absent from the EA1 cohort roster. COGENT adds 35,298 participants and has at least three shared named cohorts with EA1: Helsinki, Lothian 1936 and Minnesota. The cohort counts do not establish the exact number of shared individuals. EA3 reused 59 cohort result files from the preceding education GWAS and added 12 new or updated files; EA4 reused EA3 statistics for 69 cohorts, alongside larger 23andMe data and a revised UKBB analysis. [2, 3, 4, 10, 11]

The new FI study uses education among the measurements informing estimated intelligence scores. Its discovery cohort is independent of EA1, while its phenotype incorporates some education-related information. The population comparisons also reuse the existing reference samples and IQ compilations. [5]

The IQ_P&K analysis pairs U.S. national IQ with the previously specified demographic composite of ASW, MXL and CEU scores, rather than assigning that national estimate separately to those samples. Mexico_overall represents Mexico; MXL is a Mexican-American sample. The Ashkenazi value of 110 is an estimate based on historical Wisconsin data, rather than a representative modern survey of all Ashkenazi Jews. [12, 13]

The scatterplots and matrices use Pearson correlations of population means. Plotted scores are standardized within each comparison; fitted lines summarize the association. Eighteen boxed labels are used consistently in Figures 1–4, selected to cover the seven population groups and the spread of standardized score profiles. Larger standalone plots label all 46 samples in each PGS comparison and all 39 in the IQ plot. The label selection leaves the plotted points and correlations unchanged. Figures 5 and 6 label all 25 comparison populations using the standard 1000 Genomes codes.

Table 5 subtracts the mean of each measure within the existing groups: African, European, East Asian, South Asian, admixed American, Middle Eastern and North African, and Other. It retains all paired rows, including small groups. This removes group means but leaves finer ancestry, linkage disequilibrium and environmental differences unresolved.

Conventional confidence intervals use Fisher’s transformation, and p-values assume independent population rows. Shared ancestry and repeated country IQ estimates can violate that assumption. The accompanying CSV files retain full precision, pairwise sample counts, rank correlations, panel information and missing-data reasons. The larger supplementary matrix contains the 34 named panels from the previous subset analyses plus EA1 and IQ, for 36 measures; the 20,000 random resamples are not separate named scores.

The LD sensitivity uses the original 430- and 476-SNP FI panels. Query SNPs require MAF ≥ 0.01 in GBR and the comparator. For each query, tags within ±500 kb with phased r² > 0.01 in both populations are matched by variant identity, and the query itself is excluded. Pearson r across the two tag-r² profiles is calculated when at least three common tags are available. LD uses the authoritative Phase 3 set of 2,504 people; frequencies use the 3,202-person release. Each score is 2Σβᵢrₚᵢ(fₚᵢ − fGBR,ᵢ), a contrast with Britain using the identical retained markers and multipliers. The matched GBR baseline is 2ΣβᵢrₚᵢfGBR,ᵢ, using the comparator population’s rₚᵢ rather than GBR self-comparison weights of one. The reported value is a weighted deviation from GBR, with no division by SNP count, sum of weights or standard deviation before plotting. Undefined profile correlations are excluded. Signed negative profile correlations are retained and reverse a marker’s multiplier under this experimental rule. This is an anticorrelation between tag-r² profiles, not evidence for a reversal of signed tag–causal LD or a causal effect. Profile similarity is also insensitive to a uniform rescaling of all r² values, so a high profile correlation does not establish equally strong tagging. The primary weighted sets contain 390–430 FI markers and 413–476 FI plus COGENT markers per population. Britain supplies the zero reference and is excluded from correlations. The original LD marker panels are retained after the frequency refresh. Table 6A matches population labels to the refreshed scores, including subgroup proxies, while Table 6B recalculates the original comparison from the identical frequency source. Alternative minimum-tag thresholds, full-precision correlations, conventional confidence intervals and dependent-correlation tests are saved in the analysis files. LD weighting does not establish causal effects or individual predictive validity.

The eight ALFRED matches are between population groups; shared individuals have not been verified. The present EA1–IQ comparison includes 39 population samples.

For MoBa, the authors’ 54 lead SNPs become 42 because 12 lack a usable allele-matched frequency in at least one of the 51 population samples. Every population is scored using the same retained markers within each panel.

The frequency refresh also shows how sensitive a small panel can be to the markers available in every population. On the same 39 IQ samples, Pan-UKBB’s correlation falls from 0.586 with the earlier 63-SNP set to 0.313 with the refreshed 69-SNP set. If I hold the 61 SNPs shared by both versions fixed, it barely moves: 0.537 with the old frequencies and 0.536 with the new ones. The change comes mainly from which markers enter the score. That is another reason to keep track of the exact panel, rather than treating its name as a guarantee of stability.

The Dutch EA1 allele frequency was checked against the original source. Removing the Dutch sample raises the EA1–IQ correlation from 0.810 to 0.833.

The LD profiles were calculated locally from phased genomes with PLINK. [15]

Figures 5 and 6 standardize the GBR deviations across the displayed populations. Zero on those plotted axes marks the comparison-set average, rather than GBR. In the Yoruba comparison, both Yoruba and British allele frequencies use the Yoruba–GBR multipliers; GBR self-comparison correlations of one are not used for this matched reference.

References

[1] Piffer, D. (2013). Factor Analysis of Population Allele Frequencies as a Simple, Novel Method of Detecting Signals of Recent Polygenic Selection: The Example of Educational Attainment and IQ. Mankind Quarterly, 54(2), 168–200. Study link

[2] Rietveld, C. A., Medland, S. E., Derringer, J., et al. (2013). GWAS of 126,559 individuals identifies genetic variants associated with educational attainment. Science, 340(6139), 1467–1471. Study link

[3] Lee, J. J., Wedow, R., Okbay, A., et al. (2018). Gene discovery and polygenic prediction from a genome-wide association study of educational attainment in 1.1 million individuals. Nature Genetics, 50, 1112–1121. Study link

[4] Okbay, A., Wu, Y., Wang, N., et al. (2022). Polygenic prediction of educational attainment within and between families from genome-wide association analyses in 3 million individuals. Nature Genetics, 54, 437–449. Study link

[5] van den Berg, D. M., Malawsky, D. S., Huang, W., et al. (2026). Imputation of fluid intelligence scores reduces ascertainment bias and increases power for analyses of common and rare variants. Nature Genetics. Study link

[6] Kvalvik, E. H., Wang, Y., Walhovd, K. B., Lyngstad, T. H., & Røgeberg, O. (2026). Beyond years of schooling: Genetic associations across educational milestones in two Norwegian cohorts. PLOS Genetics, 22(9), e1012310. Study link

[7] Karczewski, K. J., Gupta, R., Kanai, M., et al. (2025). Pan-UK Biobank genome-wide association analyses enhance discovery and resolution of ancestry-enriched effects. Nature Genetics, 57, 2408–2417. Study link

[8] Okbay, A., Beauchamp, J. P., Fontana, M. A., et al. (2016). Genome-wide association study identifies 74 loci associated with educational attainment. Nature, 533, 539–542. Study link

[9] Davies, G., Marioni, R. E., Liewald, D. C., et al. (2016). Genome-wide association study of cognitive functions and educational attainment in UK Biobank (N = 112,151). Molecular Psychiatry, 21, 758–767. Study link

[10] Trampush, J. W., Yang, M. L. Z., Yu, J., et al. (2017). GWAS meta-analysis reveals novel loci and genetic correlates for general cognitive function: a report from the COGENT consortium. Molecular Psychiatry, 22, 336–345. Study link

[11] Trampush, J. W., Lencz, T., Knowles, E., et al. (2015). Independent evidence for an association between general cognitive ability and a genetic locus for educational attainment. American Journal of Medical Genetics Part B: Neuropsychiatric Genetics, 168B(5), 363–373. Study link

[12] Parra, L., & Kirkegaard, E. O. W. (2025). National IQs: measurement and defense. OpenPsych. Table A1 supplies the national estimates used here. Study link

[13] Dunkel, C. S., Woodley of Menie, M. A., Pallesen, J., & Kirkegaard, E. O. W. (2019). Polygenic Scores Mediate the Jewish Phenotypic Advantage in Educational Attainment and Cognitive Ability Compared With Catholics and Lutherans. Evolutionary Behavioral Sciences, 13(4), 366–375. Study link

[14] Wang, Y., Guo, J., Ni, G., Yang, J., Visscher, P. M., & Yengo, L. (2020). Theoretical and empirical quantification of the accuracy of polygenic scores in ancestry divergent populations. Nature Communications, 11, 3865. Study link

[15] Chang, C.C., Chow, C.C., Tellier, L.C.A.M., Vattikuti, S., Purcell, S.M. & Lee, J.J. (2015). Second-generation PLINK: rising to the challenge of larger and richer datasets. GigaScience, 4, 7. Phased LD calculations follow the PLINK 2 LD documentation. Study link · PLINK LD documentation

In modern societies, IQ predicts important dimensions of socioeconomic status, including educational attainment, occupational standing and income. Longitudinal research summarized by Strenze shows that intelligence measured earlier in life predicts later socioeconomic success. Family background and opportunities also matter, but the association gives us a historical question: did intelligence predict social standing in the past?

Ancient societies offered different routes to wealth and influence, from farming and trade to warfare and political leadership. Inheritance could also preserve advantage across generations. These differences make the intelligence–status relationship a question to investigate across historical settings.

Ancient people left no IQ tests or income surveys, but their graves sometimes preserve clues to their position. Weapons, jewellery, vessels and other goods represent resources committed to burial and can signal prestige or family standing. Burial richness can therefore serve as a proxy for socioeconomic status, interpreted within the relevant funerary tradition. Its meaning also depends on ritual, age, sex and preservation.

Figure 1 illustrates the contrast between lavish royal furnishing and an unfurnished grave.

Figure 1. Royal furnishing and an unfurnished burial

Ancient DNA supplies a second piece of the comparison. A polygenic score (PGS) combines variants associated with intelligence in modern research, allowing us to test their association with burial treatment. To my knowledge, this is the first reported association between a genetic score specifically for intelligence and quantified burial richness in ancient individuals. It opens an archaeological test of the intelligence–status relationship.

I will also consider why cognition and lighter-skin scores might be associated with richer burials, including inherited family status, assortative mating and preferential treatment of lighter-skinned individuals.

Read more

I found a mapping error in the IQ_P&K analysis: the U.S. national IQ was assigned separately to the African-American and Utah White samples, and MXL was treated as a Mexican national proxy. Mexico_overall, from our Mexican whole-genome dataset, was retained. The row removed as a Mexico proxy was the separate MEXICAN/MXL sample. I have replaced the two U.S. rows and that MEXICAN/MXL row with one U.S. score: 13.35% ASW + 22.72% MXL + 63.93% CEU, using demographic weights normalized over the three represented groups. Those groups cover 88% of U.S. residents.

I also added the separately sourced historical Ashkenazi estimate of 110. The main comparison therefore changes from 44 to 43 observations. Pearson r changes from 0.495 to 0.497 for Pan-UKBB, 0.740 to 0.780 for combined FI, and 0.739 to 0.790 for FI + COGENT (Table 1 and Figure 7). Both new FI correlations are stronger, and their advantage over Pan-UKBB widens. The national-only check, without the Ashkenazi subgroup (N = 42), also improves: 0.522/0.771/0.793 in the same score order (Table 10).

The Ashkenazi input in the original IQ column changes from 108 to 110. Its correlations barely move: 0.619/0.814/0.846 become 0.615/0.814/0.844 (Table 1). GDP, log GDP, and the EA3, EA4 and Norway EA comparisons are unchanged.

After years of waiting, the drought is finally over: we have a new intelligence GWAS! I downloaded the results and tested its scores against the older Pan-UKBB fluid-intelligence score on the same 51 populations I used before. [1] The improvement is large: the correlation with population IQ rises from 0.615 to 0.814, and reaches 0.844 when the new study is combined with COGENT.

The paper, by David van den Berg, Daniel Malawsky, Wei Huang and colleagues, with Abdel Abdellaoui among the senior authors, is called Imputation of fluid intelligence scores reduces ascertainment bias and increases power for analyses of common and rare variants. [2] Its main innovation is to make better use of the UK Biobank participants who have incomplete intelligence-test data.

The new scores also correlate more strongly with GDP and with three education scores. That is promising. I wanted to know how much of the improvement came from having more SNPs, and how much came from the SNPs that intelligence and education share. These are comparisons of population averages; testing prediction in individuals would require a separate analysis.

Read more

The epitaphs raise a question about biology. If institutions that rewarded scholarship endured for centuries, and if those rewards affected who left descendants, could ancient DNA carry any trace of the change?

The earlier article on this question found an intriguing upward shift in a genetic score around dates marking major Confucian institutional milestones. But was that pattern robust—or partly an artifact of the original data and dating choices? Here I rerun the test with more ancient individuals, a revised genetic score, corrected sample dates, and an independent check based on an East Asian genetic study. Then I apply the very same milestone dates to ancient Europeans as a placebo test. The result tells us considerably more about how seriously the original finding should be taken.

Read more

In my earlier post Was the ancient elite tall, blonde, and light-skinned?, I explored whether genetic traits associated with appearance and education clustered in ancient populations. Here I bring burial furnishing directly into the comparison. The strongest current finding is that a genetic score pointing toward lighter untanned skin is associated with more richly furnished burials within the recorded cemeteries, even when an education-related genetic score enters the same model.

The education-score association is positive, smaller after that adjustment, and sensitive to influential burials. This shifts the emphasis of the story toward pigmentation-associated variation. The explanation could still involve ancestry, inherited family position or burial customs.

Was colourism already present in ancient Europe? Did the association reflect differences in ancestry, or could generations of elites spending more time indoors have favoured pigmentation adapted to their way of life?

Read more

The goal of this study is to test the hypothesis that lower local UVB exposure and a longer local history of farming are associated with lighter-pigmentation genetic scores (after accounting for sample dates). These associations leave room for migration and ancestry changes as well as dietary adaptation. After the paywall I will also answer the questions raised by the readers on X about the previous post.

Ultraviolet B radiation (UVB) starts vitamin D production in the skin. Melanin absorbs some of that radiation. Where UVB is scarce, especially at higher latitudes, less melanin allows more UVB to reach the skin’s vitamin D precursor at a given exposure. If this improves survival or reproduction, natural selection can favour light-associated alleles over generations. This is the vitamin D explanation for why low UVB could favour lighter skin. A farming diet could add to that pressure if it supplies less vitamin D than the foods it replaces. Jablonski and Chaplin 2010

The development of civilisation may also have increased time spent indoors. Less outdoor time would mean less UVB reaching the skin, even at the same ambient UVB intensity, potentially adding to the pressure favouring lighter skin. Our data contain no measure of time spent indoors, so we cannot test this possibility directly or determine how much of the farming association it might explain.

I use the continuous genetic score from the same 587 variants discussed in Can we predict population skin color with polygenic scores?. I call it Skin587 and reverse its original direction here so that higher values point towards lighter pigmentation.

(BP means years before 1950; kyr BP expresses dates in thousands of years. Score changes use a fixed ancient-cohort standard deviation (SD). Standardized regression coefficients rescale each predictor and the outcome by their variation among the same people, allowing the associations to be compared.)

Farming, ultraviolet exposure and pigmentation

To test the proposed connection between farming and pigmentation, I assigned a date for the appearance of farming around each sample locality, using archaeological domestication and crop evidence. Farming duration is the interval from that assigned date to the person’s date, with zero for people who lived before local farming. Figure 1 shows the localities and their assigned chronologies. Betti et al. 2020

The expanded analysis retains 6,643 European genotype records from 92 verified studies, dated from the present back to 18,000 BP. It includes 527 records predating assigned local farming. Years BP enters the model alongside farming duration, so the farming coefficient is estimated after accounting for the general chronological trend.

Figure 1. Sample locations and local farming history

The main model uses local UVB, the part of sunlight that starts vitamin D synthesis in the skin. UVB and latitude are strongly related, so I fit latitude in a separate sensitivity model. Both specifications include sample date, farming duration, longitude and study effects. The local glUV dataset measures erythemally weighted radiation: it is a geographical UVB proxy, rather than a direct measure of vitamin D production. Modern exposure at an ancient sample’s recorded location also carries uncertainty about prehistoric conditions.

I report two ways of accounting for shared background. Study-clustered regression allows samples from the same publication to share unexplained influences. A genetic relationship model also accounts for genome-wide similarity between samples, although it can absorb part of the ancestry turnover we want to understand. Its uncertainty calculation assumes the fitted relationship model is adequate.

In the UVB specification, farming duration has a standardized beta of +0.126 in study-clustered regression, with a 95% interval from +0.023 to +0.229 and p=0.022. The genetic relationship model gives +0.114, with an interval from +0.058 to +0.169 and p<0.001. A standardized beta describes the score change associated with one standard deviation of the predictor. Raw coefficients are about +0.058 and +0.053 score SD per additional thousand years of local farming (Table 1).

Table 1. Panel A: Study clustered regression with UVB

Table 1. Panel B: Genetic relationship model with UVB

The UVB association is negative: a standardized beta of -0.238 in clustered regression (95% interval -0.322 to -0.154) and -0.153 in the relationship model (-0.196 to -0.110), with p<0.001 in both. Higher UVB is associated with a darker-directed score after the other adjustments; lower UVB is associated with a lighter-directed score (Table 1; Figure 2). This pattern is consistent with the vitamin D hypothesis described above. Vitamin D levels and individual dietary intake were not measured.

Figure 2. Farming duration and UVB associations

The farming association is stronger in the latitude-only sensitivity, with standardized betas of +0.167 and +0.143 in the clustered and relationship models (Table 2), compared with +0.126 and +0.114 under UVB control (Table 1). Farming duration remains strongly correlated with chronological age: its variance inflation factor is 11.18 with latitude and 9.86 with UVB. The data distinguish those effects imperfectly.

Table 2 provides every substantive predictor in the latitude sensitivity, including chronological age and longitude. Table 1 gives the corresponding UVB model. The supporting coefficient CSV retains the full raw and standardized estimates and their uncertainty.

Table 2. Panel A: Latitude sensitivity in clustered regression

Table 2. Panel B: Latitude sensitivity in the relationship model

Longer local farming history is therefore associated with lighter-directed pigmentation scores under several specifications. The precise dietary interpretation remains uncertain. Migration, ancestry turnover, latitude, regional history and dating error can all influence that association. The genetic relationship model adjusts for genome-wide similarity, which can remove part of the historical ancestry signal as well as confounding. Its narrower intervals deserve comparison with the study-clustered results.

The farming and UV associations leave another part of the story to explain: which populations carried the light-associated alleles, and how those alleles spread. Below, I turn to the questions readers raised about the previous post:

• Did Corded Ware get its light skin alleles from earlier European farmers, particularly the Globular Amphora culture?

• Were those alleles already present in the eastern and Caucasus hunter-gatherer ancestors of the Yamnaya?

• How did farmers in Spain compare with farmers in Sweden?

• What do “intermediate” and “pale” mean relative to modern Europeans?

• Can changes within steppe-derived chromosome segments tell us where light skin alleles came from?

• How do light-associated variants in Sintashta and Andronovo fit a farming explanation?

Read more

Descendants inherit recombined chromosomes from steppe migrants, earlier European farmers and hunter-gatherers. Pigmentation alleles can change frequency within each inherited background. Irving-Pease and colleagues traced allele-frequency histories in several ancient source lineages; they did not estimate a single WSH-specific trajectory. I ask where change occurred using the Skin587 score from my earlier skin-colour analysis, blue-eye alleles and a blonde-hair association score.

Figure 1 shows why chromosome-level ancestry matters. A person can be 40% steppe-related overall while a pigmentation allele lies on a farmer-like or hunter-gatherer-like copy. Testing alleles on their own labelled copies offers a more direct comparison than pairing whole-genome ancestry with a whole-genome score. The inference still depends on the labels and the observed sites.

Figure 1. Ancestry labels can differ between chromosome copies

Start when a proto-Yamnaya source existed

Nikitin and colleagues date the formation of Serednii Stih to around 4500 BCE. Another study explicitly describes this group as Proto-Yamna. The primary analysis begins at 6,500 years before present (BP), approximately 4550 BCE on a 1950 reference; a 6,000 BP check begins near the estimated formation of the Core Yamnaya mixture. These are rounded boundaries around historical processes, rather than exact first births of an ancestry.

This cap removes much older Europeans whom a closed-set classifier had labelled partly “steppe” despite predating the source. The forced three-way model still assigns a known label even when a chromosome has another history. All tract percentages and scores below refer to classifier-labelled sequence.

Where the samples come from

The main Skin587 cohort contains 3,328 European sample records dated ≤6,500 BP, with coordinates, paired WSH/non-WSH scores and a stored genotype-probability (GP) quality proxy above 0.9. A

Figure 2. Reported localities of 3,328 eligible European sample records

Did WSH-labelled chromosomes become lighter?

The 587-weight Skin587 score points toward darker pigmentation as it rises. Of its 587 positions, 579 occur in the modelled chromosomes. I averaged the observed effect-allele contributions separately on WSH-labelled copies and across all valid chromosome copies in the same people. The whole-genome comparison therefore uses exactly the same usable allele opportunities as the local scores. Both slopes share one fixed score scale.

Older samples have a higher score on WSH-labelled copies: +0.1640 fixed score standard deviations (SD) per 1,000 years older (study-clustered 95% interval +0.0955 to +0.2324; p=.001). The matched whole-genome slope is +0.1495 (+0.0928 to +0.2063). Thus the WSH-labelled score moved in the lighter-associated direction toward the present, as did the genome-wide score (Figure 3; Table 1, panel A).

Figure 3. Skin-score change within WSH-labelled ancestry and across the genome

The WSH estimate is slightly larger. The direct WSH-minus-genome slope is +0.014 fixed SD per 1,000 years older (−0.039 to +0.068; p=.526; Figure 3; Table 1, panel A).

A score trend among WSH-labelled tracts is consistent with lightening on that ancestry background. Testing natural selection requires comparing the frequency trajectories of its pigmentation alleles with suitable matched neutral variants on well-validated ancestry tracts. Ju and Mathieson found evidence of skin-pigmentation selection in steppe-assigned people using whole-genome scores and 10,000 frequency-matched control panels. Their whole-genome test leaves the WSH-specific case open.

Figure 4 shows the two Skin587 age trends in Table 1 against each sample’s actual date. Older samples are on the left in every panel. I keep each date in Years BP so the horizontal distance still means elapsed time, rather than a deviation from the date predicted by study and location. The plotted scores account for study, location and the two imputation-variance measures; the lines have the adjusted age slopes in Table 1. The dots do not show the unadjusted score levels of WSH and non-WSH ancestry.

Figure 4. Adjusted Skin587 trends on WSH and other chromosome copies

Was change faster on WSH than on other tracts?

Read more

Figure 1. Two hypothetical burials, different grave furnishings

In my earlier essay, Was the Ancient Elite Tall, Blonde, and Light-Skinned?, I raised a hypothesis: genetic signals associated with lighter skin, higher educational attainment and social standing might have clustered in past populations. Higher-status people may have spent more time in sheltered work than outdoor labourers, exposing their descendants to different selection pressures; ancestry and marriage networks could also have shaped the pattern. A grave inventory offers a more direct observation with which to test part of it.

In earlier work, we traced scores for educational attainment, socioeconomic status and other traits through ancient European genomes. Akbari and colleagues studied how variants associated with modern traits changed through time in West Eurasia. These studies describe temporal genetic patterns. Interpreting those patterns as ancient education or social status assumes that relationships learned from modern genome-wide association studies carry over to very different populations and societies. Neither study checked an individual’s education score against an archaeological measure of that same person’s position.

There is a useful precedent for a physical trait: Marciniak and colleagues compared height scores with stature estimated from the long bones of the same 167 ancient people. For pigmentation, Ju and Mathieson traced skin-colour-associated variants through time, but could not check the original skin colour of those people directly. A skeleton leaves no comparable measure of years of schooling, income or test performance.

Modern evidence still gives a reason to test an education score against social outcomes. A separate income genome-wide association study found a 0.917 genetic correlation with educational attainment: the associated genetic signals overlapped strongly. An education-based score also explained an additional 4.7% of variation in occupational income in a modern validation sample. A study of occupational status found genetic correlations of 0.96–0.97 with educational attainment. These results come from modern cohorts; they do not show that the restricted ancient EA4 score predicts status in past societies.

This study tests that missing link. I compare each ancient person’s educational-attainment score (EA4) with the objects recorded in that person’s own grave. To my knowledge, this is the first cross-cemetery test pairing an ancient education score with the documented burial goods of those same individuals. Earlier ancient DNA work on Bronze Age Europe and Mokrin used kinship to investigate burial differences, but did not make this score-to-inventory comparison. Grave goods directly record burial treatment and can inform questions about social status.

Read more

In Did Cold Climates Select for Cold Personalities?, warming was linked to higher genetic scores for Extraversion, a trait associated with sociability and outgoing behaviour. Did Climate Change Make Humans Shorter or Taller? found higher height scores after cooling. Did Climate Cooling Select for Cognitive Abilities? suggested a cooling association too, but only when warming and cooling were analysed separately. These were clues from DNA, not measurements of ancient personalities, stature or schooling.

A larger collection of ancient European genomes now lets me revisit all three. Looking back three and five thousand years produces a surprise: height and educational-attainment scores point in different climatic directions, while the clearest personality pattern appears between places rather than through time.

Europe’s people also moved and mixed as its climate changed. Taking shared ancestry into account makes some of the associations much smaller, but does not erase every pattern. I show the results both ways, because the difference changes what we can reasonably conclude.

The full analysis below shows where the height–cooling link holds up, how the educational-attainment finding changes direction, and why the personality result depends on which question we ask. Subscribe to read the trait-by-trait findings and all seven charts, including a skin-colour and sunlight check of what these comparisons can reveal.

Read more

Using polygenic scores to predict individual differences within populations is well established. Their use to compare population averages remains controversial. A score that predicts variation within one population does not automatically provide a reliable measure of differences between populations, because its estimated effects can depend on ancestry, environmental conditions and the design of the original study. [8]

For this test, I use a score extracted from a genome-wide association study of untanned skin colour. Pigmentation can be measured objectively on sun-protected skin, helping distinguish baseline colour from short-term changes due to sun exposure. My expectation is that, once those temporary fluctuations are set aside, baseline pigmentation is much less environmentally sensitive than height or educational outcomes. That makes it a useful benchmark for evaluating prediction across populations. [4,9]

Read more