Out of Africa or Out of Babel?

"Thus, no origin outside of Africa had the explanatory power of an origin anywhere in Africa." — Sohini Ramachandran and colleagues, 2005¹

"There is no obvious outlier and we could not detect any effect of the historical census size of the populations." — Franck Prugnolle, Andrea Manica and François Balloux, 2005²

"Ask yourself, 'Did Ham settle Africa?'.  From figure 3, it is clear that multiple people groups did." — Robert Carter, 2024³

Summary

Human genetic diversity falls in a straight line as you move away from Africa.  The further a population lives from Addis Ababa along a walkable route, the fewer of its gene copies differ from each other.  The mainstream reads this as a chain of founder events out of Africa about 60,000 years ago.  Creationists read the same map as a dispersal out of Mesopotamia after Babel, about 4,000 years ago, with Africa settled early and holding the most diversity because it held the most people.

The gradient can be refitted from any starting point on Earth.  That test was run here on the raw public genotypes.  It comes out against Babel, and weaker than either side usually claims.

What the refit found.

1. Mesopotamian origins fit badly. Ramachandran's regression, recomputed here from the 783-microsatellite genotypes of 1,027 people, gives R-squared 0.859 from Addis Ababa, 0.865 from Nairobi and 0.869 at the global optimum in western equatorial Africa, against 0.647 from Babylon, 0.644 from the plain of Shinar, 0.635 from Ur and 0.612 from Mount Ararat. Addis Ababa beats Babylon in 99.97% of 20,000 bootstrap resamples over populations and in 100% over loci. The size of the gap depends on three tiny African samples; the sign of it does not, under any routing convention tried.
2. The whole difference comes from where the African populations sit, not from the shape of the Eurasian cline. Drop the eight African populations and refit on the 45 others, and the discrimination vanishes: Ararat 0.854, Babylon 0.853, Addis Ababa 0.850. A Mesopotamian origin fits the non-African world marginally better than an African one. Anyone who says the gradient proves an East African birthplace is claiming more than the data give.
3. Under a Babylon origin, all eight African populations lie above the line. Their mean residual is +0.045, or 1.89 times the whole fit's RMSE. Bedouin, Druze and Palestinians live 885 to 904 km from Babylon and carry 0.7247 to 0.7364. The south-east Bantu live 7,807 km away and carry 0.7696. A founder expansion makes the people nearest the start the most diverse. The measurement reverses that.

The statistic that is harder for a young dispersal than heterozygosity is.

4. Africa's private allelic richness is five to ten times everyone else's. Rarefied to 20 gene copies per region per locus so sample size cannot drive it: Africa 0.168 private alleles per locus against 0.017 to 0.031 in every other region. Under a Babel model every private allele anywhere is a post-Babel mutation, so this is a claim that African mutation rates ran five to ten times faster than the rest of the world's.
5. Africa combines high between-population differentiation with the highest within-population diversity, and drift cannot produce that. Mean pairwise F_ST within Africa is 0.0479 with within-population heterozygosity 0.7573. The Americas, the textbook case of small isolated founder groups, show 0.0600 with 0.6060. Drift raises F_ST by removing variation. Africa has both.
6. One measurement bypasses every calendar. Sankararaman 2012 dates Neanderthal gene flow into the European line at 1,805 to 2,043 generations, from recombination maps including one built on living Icelandic pedigrees.⁴ That counts meioses, not years, so rejecting radiometric dating and molecular clocks does not touch it. Four thousand years at 25 years per generation is 160 generations. No creationist author found in this project's search has addressed the gap.

What the creationist side gets right, and it is not nothing.

7. The published creationist literature has never engaged this measurement. Only three creationist sources address African genetic diversity: Carter 2009, Jeanson 2016 and an ICR news piece of 2012. None of them cites Ramachandran 2005, Prugnolle 2005, Handley 2007, Manica 2007 or Li 2008. Carter 2009 lists Liu 2006 in his references and mentions it in one clause, without touching its result. Tomkins's 2020 Journal of Creation paper "Out of Babel — not Africa" never uses the word heterozygosity.
8. The two mechanisms offered both fail on data, not on principle. Carter's small-isolated-tribes mechanism predicts the American pattern, the opposite of what Africa shows. Jeanson's shorter-African-generation mechanism gets within a factor of two on the arithmetic, and a direct measurement of the reverse kills it: Wang 2023 reads generation intervals out of the mutation spectra of 1,247 trios and gets 26.9 ± 3.5 years for Africans against 20.1 to 21.0 years for East Asians, Europeans and South Asians.⁵ Africans have the longest measured interval.
9. The "Africa grew biggest" rescue was tested in print in 2005 and failed. Prugnolle, Manica and Balloux looked for a census-size effect in these data and found none.² No creationist source cites the test.
10. The mainstream disputes are real. Origin estimates inside Africa range over several thousand kilometres. Amos and Hoffman 2010, on the exact dataset Ramachandran used, find diversity was lost in two discrete bottlenecks rather than a smooth serial chain.⁶ Mallick 2016 and Pagani 2016 appeared in one issue of Nature and disagreed about whether any living population carries ancestry from an earlier dispersal. None of these disputes moves a published position within a factor of ten of a post-Babel chronology.

Verdict.  The geography of the non-African world is compatible with a Near Eastern staging post, which is what Out of Africa says too.  What the young model cannot absorb is the position of Africa itself, and the generation counts.  The distances are not the problem.  The depth is.

Part 1: The measurement both sides are arguing about

Take one population and one stretch of DNA that varies.  Count how often two gene copies drawn at random from it differ.  That number is expected heterozygosity, written H_e.  It runs from 0 where everyone carries the same allele to nearly 1 where almost every copy is different.  It describes a sample.  It is not a theory about the past.

In 2005 Sohini Ramachandran and five colleagues measured it in 1,027 people from 53 populations at 783 autosomal microsatellites, then plotted each population's H_e against how far it lives from Addis Ababa.  Straight-line distances gave a decent fit.  Routing the distances through five obligatory land waypoints, at Anadyr, Cairo, Istanbul, Phnom Penh and Prince Rupert, gave a better one:  their genetic-distance regression rose from R-squared 0.5882 to 0.7834.¹  The heterozygosity regression they printed reads "heterozygosity = 0.7682 − (6.52 × 10⁻⁶) × (distance from Addis Ababa)."¹

They then repeated the whole regression using each of 4,210 lattice points on the globe as the hypothetical origin, and mapped how well each one worked.  That is what makes the paper a test rather than an illustration.

"The 936 locations in Africa used as origins resulted in R² values ranging from 0.757 to 0.870 (the SD of R² within Africa was 0.017), whereas R² using the 3,274 non-African locations as origins ranged from 1.67 × 10⁻⁷ to 0.744 (the SD of R² outside of Africa was 0.245).  Thus, no origin outside of Africa had the explanatory power of an origin anywhere in Africa."¹

That is the shape of the argument.  It is not an inference from Africa having the most diversity.  It is a fitted surface with a maximum, and the maximum is not near Mesopotamia.

The model being fitted is not equilibrium isolation by distance, and the authors say so.  Both sides misread this graph as a statement about slow gene flow reaching balance.  The Discussion rejects that reading:  "There clearly has not been time to reach equilibrium between the extremes of man's inhabited range, or even within continents, in the very short evolutionary history of modern humans."¹  What is being fitted is a chain of founder events, each one drawing a small group out of the last.  Their simulation used 100 bottlenecks of 250 founders each and produced a line of the observed shape.¹

It is not an artefact of microsatellites.  Franck Prugnolle, Andrea Manica and François Balloux had reported the same relationship earlier in 2005 on 377 Marshfield-panel microsatellites in 51 populations, with distances computed by graph search over land:  "geographic distance … from East Africa along likely colonisation routes is an excellent predictor for genetic diversity of human populations (R² = 85%)."²  Three years later Jun Li and ten colleagues repeated it on about 650,000 SNPs in 938 people from the same panel:  "The mean heterozygosity across autosomal haplotypes … is negatively correlated with distance from Addis Ababa, Ethiopia, with a correlation coefficient r of −0.91."⁷  An r of −0.91 is R-squared 0.83, essentially the microsatellite figure on a different marker system.

A skeptical reader should raise SNP ascertainment bias, since the SNPs on that array were chosen for varying in European and Asian medical cohorts.  The bias is real and it runs the wrong way to help the Babel model:  choosing SNPs that vary in Europeans inflates European diversity relative to African, and the African excess survives anyway.  Li's own caveat is about sampling density, not direction:  "HGDP-CEPH is not a random sample of the world's populations:  Some parts of the world (e.g., China and Pakistan) are more densely covered than others (Africa, the Americas, and Oceania)."⁷

Sarah Tishkoff's 2009 survey attacked the sampling gap directly, genotyping 1,327 markers in 2,432 Africans from 121 populations, and found the same thing:  "African and African American populations, with the exception of the Dogon of Mali, show the highest levels of within-population genetic diversity … In addition, genetic diversity declines with distance from Africa … In addition, more private alleles exist in Africa than in other regions."⁸

Part 2: Refitting the gradient from Babel

The Rosenberg lab posted Ramachandran's genotypes and population coordinates in the open.  So the origin scan can be re-run with Babylon, Ur, the plain of Shinar and Mount Ararat among the candidate origins.  Ramachandran's lattice covered those points, since it covered all land, but the paper reports no value for any of them.

What was computed, and how it was checked.  Expected heterozygosity was recomputed per population from the raw genotype file as the unbiased H_e, averaged over the 783 loci with missing calls dropped.  Distances are great-circle, routed through Ramachandran's five waypoints.  Then the same regression, at each candidate origin.  The code is at research/code/origin-scan.ts.

It reproduces the paper's published lattice-scan statistics closely.  R-squared across African origins runs 0.757 to 0.870, matching the published range.  The standard deviation within Africa is 0.018 against 0.017, and outside Africa 0.254 against 0.245.  The strongest negative correlation falls at exactly the published coordinates of 4.3N, 12.8E, at r = −0.9324 against the published −0.933.¹  Two differences should be named.  Ramachandran scored 4,210 land points from an equal-area lattice, 936 of them in Africa;  this scan uses a plain 2.5-degree grid with box-defined continents and no coastline mask, 913 points of which fall in the Africa box.  And their best non-African origins are in South America, where the correlation runs the other way, while this scan's best non-African point is 30N, 35E at 0.727 against their 0.744.

One published number does not reproduce, and chasing it down mattered.  Ramachandran's Figure 4A regression from Addis Ababa prints R-squared 0.7630 with a slope of −6.52 × 10⁻⁶.  The routing above gets 0.8593 with a slope of −7.567 × 10⁻⁶, so the paper's Addis distances are about 16% longer.  The paper names its five waypoints but never says which waypoints join which regions;  it points at a supplementary figure for that.  A sweep of 2,304 possible assignments finds the one that fits.  Route western and southern Asia out of Africa through Cairo and then Istanbul, route eastern Asia and Oceania through Cairo and then Phnom Penh, and route the Americas on through Anadyr and Prince Rupert, and the regression from Addis Ababa gives R-squared 0.7762 with a slope of −6.611 × 10⁻⁶.  That is within 2% of both published figures.

Neither convention reproduces everything.  The one used for the tables below matches the lattice scan and misses Figure 4A.  The one just described matches Figure 4A and misses the lattice scan, putting African origins in a range of 0.718 to 0.778 where the paper prints 0.757 to 0.870.  Addis Ababa beats Babylon under both, 0.859 to 0.647 and 0.776 to 0.566, and with the waypoints switched off entirely, 0.662 to 0.533.⁹

Result 1.  African origins fit;  Mesopotamian origins do not.

A bootstrap over the 53 populations, 20,000 resamples, puts Addis Ababa ahead of Babylon in 99.97% of them.  Resampling the 783 loci instead puts Addis ahead in 100%.  The sign of the gap is not a sampling accident.

Its size is more fragile, and the reason is sample size.  The two most diverse populations in the panel are the south-west Bantu, sampled from three people, and the south-east Bantu, sampled from five.  The sample-size correction that expected heterozygosity requires lifts the south-west Bantu from an uncorrected 0.6284 to 0.7607.  Drop every population with fewer than eight sampled people and Babylon climbs from 0.647 to 0.754 while Addis Ababa holds at 0.861.  The gap survives at half its headline width.⁹

Result 2.  Under a Babylon origin the African populations are all on the wrong side of the line.  All eight African populations fall above the line fitted from Babylon, and they are seven of the eight largest positive residuals in the world.  The mean African residual is +0.0448 against an RMSE of 0.0237, so the African block sits 1.89 RMSE units high.  This result does not rest on the small samples:  drop them and it strengthens to 2.09.⁹

Set the populations side by side at their distances from Babylon and the problem needs no statistics at all.  The Druze and Palestinians live 885 km away and carry H_e of 0.7247 and 0.7357.  The Bedouin at 904 km carry 0.7364.  Under a serial founder expansion out of Mesopotamia the people who never left should be the most diverse group on Earth.  They are less diverse than every African population in the sample.

Part 3: What the gradient does not show

This section is the concession, and it is not a small one.

Drop the eight African populations out of the dataset and refit the whole scan on the 45 non-African ones.  The discrimination disappears.

Among non-Africans, a Mesopotamian origin fits marginally better than an African one:  0.854 from Ararat and 0.853 from Babylon against 0.850 from Addis Ababa.  The serial founder effect therefore does not locate an origin inside Africa by the shape of the Eurasian and American cline.  Every bit of its discriminating power comes from one fact, that the African populations sit above the line drawn to the rest of the world.

The ordering of that last column is not even stable.  Switch off the waypoints and use plain great-circle distance and the Mesopotamian lead grows:  Ararat 0.733, Ur 0.723, Babylon 0.718, Addis Ababa 0.623.  Use the routing that reproduces Ramachandran's Figure 4A and it reverses:  Addis Ababa 0.731, Ararat 0.650, Babylon 0.645.⁹  A ranking that flips with the routing rule is a ranking that carries no information.  The reading kept above is the one least favourable to this article's own verdict.

So the honest statement of this measurement is "Africa is the high-diversity end of the chain," not "the chain started at 9N, 38E."  Anyone who tells a skeptic the gradient proves an East African birthplace has overstated it.

The sharpest statement of this limit came from the people who built the second origin scan.  Hua Liu, Franck Prugnolle, Andrea Manica and François Balloux fitted a stepping-stone colonisation model to the same 783 microsatellites in 2006.  Their expected diversity does not fall in a straight line, because populations in the middle of a chain receive migrants from both sides.  Their conclusion, footnoted to Ramachandran 2005:

"An important corollary stemming from this nonlinearity is that use of linear regressions on gene diversities might not allow correct inference of the geographic origin of modern humans."¹⁰

That warning cuts against the fit from Addis Ababa exactly as far as it cuts against the fit from Babylon.

Two more mainstream sources concede the same thing.  Ramachandran's own paper warns that "Because sampling was not very dense in Africa, especially in Eastern and Northern Africa, a larger sample might enable this approach to further localize the specific origin of the expansion."¹  And in 2021 a group including Carina Schlebusch, Chris Stringer, Mattias Jakobsson, Andrea Manica, Sarah Tishkoff and Mark Thomas rejected a Nature claim that mitochondrial DNA localises human origins to a southern African wetland.  Their title is "Human origins in Southern African palaeo-wetlands?  Strong claims from weak evidence," and their objection is methodological:  one locus has one genealogy, and present-day geographic distributions are poor guides to past ones.¹¹  That objection applies in weaker form to every origin scan in this article, including the one computed above.

The mainstream disagrees about where in Africa, and by thousands of kilometres

Four published origin estimates, four different places.  Ramachandran 2005 puts the best origin near 4.3N, 12.8E, in western equatorial Africa.¹  Tishkoff 2009 optimises both an origin and a waypoint and lands on "southwestern Africa, at 12.5°E and 17.5°S, near the coastal border of Namibia and Angola."⁸  Brenna Henn's 2011 study repeated the scan on two statistics that admixture cannot inflate, linkage disequilibrium decay and F_ST, and got 14S, 12E from the first and 20S, 22E from the second, with a P value of 0.06 on the second after permutation.¹²¹²  The reproduction above puts its own optimum near 2.5N, 20E.

These points are 2,000 to 4,000 km apart.  A creationist writer who says the geneticists cannot agree where humans came from is right about the location inside Africa.  That does not close the distance between Africa and Mesopotamia, or shift one of these estimates off the continent.

Two more real disputes belong here for the same reason.  Amos and Hoffman 2010 analysed the same 783 microsatellites in the same 53 populations with a method that looks for a transient excess of heterozygosity relative to allele number, which appears right after a sharp decline.  They found the loss of diversity concentrated in two events rather than spread over a hundred:  "We find evidence of two primary events, one 'out of Africa' and one placed around the Bering Strait … but contrast with the apparently smooth gradient of variability that is revealed when current heterozygosity is plotted against distance from Africa."⁶  A dispersal with a small number of large founder events is closer in shape to a Babel story than the smooth-cline picture is, and this is a mainstream paper on the mainstream dataset saying so.

Read where the two events sit before using it.  "A clear pattern is seen, with Africans showing least evidence of a bottleneck and the Middle East on average the strongest evidence."⁶  The Middle Eastern populations look like descendants of a founder group and the African populations look like the source.  A Babel model needs that backwards.

And Prugnolle's reading of his own data contradicts Amos:  "Our results point to a history of colonisation of the world characterised by a very large number of small bottlenecks and limited subsequent gene flow."²  Two mainstream teams, similar data, opposite answers about the number of founder events.  Neither answer is within a factor of ten of 4,000 years.

The third dispute is about how many dispersals left Africa.  Mallick 2016 tested 300 high-coverage genomes for ancestry from an earlier exit and found none:  "no position or proportion of the deep lineage improves the fit."¹³  Pagani 2016 appeared in the same issue of Nature and concluded the opposite:  "we estimate that at least 2% of their genome sequence reflects an earlier, otherwise extinct, dispersal," dated near 120,000 years ago.¹⁴  The gap between them is 2% of one population's genome, and 120,000 years against 60,000.  Both agree that non-Africans descend overwhelmingly from one African source, and neither is within four orders of magnitude of a post-Babel dispersal.

Part 4: Private alleles, and the one pattern drift cannot make

Heterozygosity is the statistic the creationist side has the best chance against, because heterozygosity is stubborn.  Nei's 1975 result, worked through in population bottlenecks, is that a population which crashes and then grows fast keeps most of its heterozygosity and loses its rare alleles permanently.  Counts of distinct alleles are the harder test.  They were computed here on the same genotypes, rarefied to 20 gene copies per region per locus so unequal sample sizes cannot drive the comparison.

Africa carries five to ten times the private allelic richness of any other region at the same sample size.  Under a Babel model every population descends from the same eight people on the Ark, so every private allele anywhere is a mutation that arose after Babel.  Producing 0.168 private alleles per locus in Africa and 0.017 in Europe from the same stock in the same 4,000 years requires African mutation rates roughly ten times the European rate.  Nobody has measured anything like that, and the one direct measurement of a related quantity runs the other way.

The second table is the one that decides between the two models.  Mean pairwise F_ST among populations of the same region, set against mean within-population heterozygosity, on the same 783 loci:

Read the American row first, because it is the control.  Small groups, isolated, drifting apart:  the highest F_ST in the table and the lowest heterozygosity.  That is what drift does.  It separates populations by throwing variation away.

Now read the African row.  The second-highest F_ST and the highest heterozygosity.  Drift cannot produce that combination, because the mechanism that raises F_ST is the same one that removes heterozygosity from the subpopulations.  What produces it is deep divergence among populations that were each individually large for a long time.

Tishkoff's independent African sample reports the same asymmetry in continental units:  "The proportion of genetic variation among all studied African populations was 1.71% … In comparison, Native American and Oceanic populations showed the greatest proportion of genetic variation among populations (8.36% and 4.59%, respectively), most likely due to genetic drift."⁸

Part 5: What the creationist literature has said about this

Very little, and that is a finding rather than an insult.  A search of creation.com, answersingenesis.org, answersresearchjournal.org and icr.org for the serial founder effect, the heterozygosity gradient, and the authors who published it turned up three sources that address African genetic diversity at all:  Carter 2009, Jeanson 2016 and an ICR news piece of 2012.  None of them cites Ramachandran 2005, Prugnolle 2005, Handley 2007, Manica 2007 or Li 2008.  The one contact with this literature is a single clause in Carter 2009:  "A highly complex model may not be needed however (e.g. Liu et al. 2006), but there are always dangers involved in oversimplification."¹⁵  Liu 2006 is the paper that fits an explicit colonisation model to these same 783 microsatellites.  Carter cites it for the proposition that simple models can work, and says nothing about what it found.

Jeffrey Tomkins's 2020 Journal of Creation paper is the flagship statement of the position, titled "Out of Babel — not Africa."  It opens by calling recent human evolution "a quagmire of ever-changing speculations that vary and conflict in their dates and geographical specifics with each new paleoanthropological or archaic DNA sequence discovery."¹⁶  Its four arguments are that the mainstream is not agreed on which African sub-model is right, that archaic admixture broke the replacement version of Recent African Origins, that pedigree mitochondrial rates give young dates, and that a language study supports a Babel dispersal.  The word heterozygosity does not appear in it.  Neither does any origin scan, any of the six papers above, or any account of why African populations sit above a line fitted from anywhere else.

Its archaic-admixture argument overstates its own source.  Tomkins writes that a 2020 study "has thrown yet another wrench into the evolutionary machine of the RAO paradigm which shows that genomic regions of Neandertal ancestry are present in individuals of African descent at much higher levels than previously believed."¹⁶  The study is Lu Chen and colleagues, and its finding is real:  17 Mb of Neanderthal sequence per African individual where earlier methods found under 0.5 Mb.¹⁷  Its own explanation sits in its section heading:  "Back-Migration with Non-Africans and Pre-out-of-Africa Human-to-Neanderthal Gene Flow Contribute to Apparent Neanderthal Ancestry in Africans."¹⁷  Africans carry about 33% of the non-African level, and the authors attribute most of that to Eurasian ancestry entering Africa later.  Tomkins does report the explanation, in one sentence:  "The RAO advocates have attempted to explain this anomaly by claiming that humans with Neandertal ancestors migrated back down into Africa from Europe on multiple occasions, spreading their genes among African populations."¹⁶  He then changes the subject.  Stating an explanation and never answering it is not the same as concealing it, and the charge here is the milder one.  The useful lesson in the paper is methodological:  a decade of studies reported no Neanderthal ancestry in Africans because the method used Africans as the unadmixed reference, which set the answer to zero by construction.

Mechanism one: small isolated tribes

Robert Carter's 2009 Journal of Creation paper is the only place a creationist author offers a mechanism for the African excess.

"This is a critical issue, for small and isolated populations experience inbreeding, rapid drift and the rapid accumulation of new mutations.  The situation violates the fundamental assumptions behind the SNM while providing potentially excellent material to support the biblical model:  for if several of the sub-Saharan African populations existed in such a condition after the Flood, this might go a long way in explaining why there is more genetic diversity among people of African descent."¹⁵

The mechanism predicts the American row of the table in Part 4, not the African row.  Small isolated populations lose within-population heterozygosity;  that is what drift is.  Subdivision raises diversity for the whole species by slowing coalescence, but every number in this article is measured within populations, and there drift subtracts.  "Rapid accumulation of new mutations" does no work either:  per-individual mutation rates do not depend on population size, and what small populations do faster is fix mutations, which removes heterozygosity.

Carter's third route is the only one that raises heterozygosity quickly, and he states it correctly:  "When previously separated populations come back together, heterozygosity increases."¹⁵  Admixture is real and fast.  It is also visible:  populations that mix show reduced differentiation from the groups they mixed with, and African F_ST is the second-highest in the table rather than the lowest.  Ramachandran's own figure flags the Mozabite and the Uygur as sitting above the line for that reason, and neither is African.

Mechanism two: a shorter African generation

Nathaniel Jeanson's 2016 Answers Research Journal paper is more specific.  Africans carry more mitochondrial differences, so Jeanson proposes that African generation intervals have historically been much shorter, using 15 years, which fits more generations and therefore more mutations into the same 4,000 years.¹⁸

The arithmetic is not the objection.  Extend the mechanism from mitochondria to the microsatellites used above.  Under the infinite-alleles approximation heterozygosity grows by about 2μ(1 − H) per generation.  At the rate Ramachandran's own simulation used for this panel, 7.567 × 10⁻⁴ per locus per generation, with H about 0.75, that is about 3.8 × 10⁻⁴ per generation.¹  Four thousand years at 15 years per generation is 267 generations;  at 25 years it is 160.  The 107 extra generations buy about 0.04 of heterozygosity, against an observed African excess over the Babylon regression of 0.045.  The mechanism is within a factor of two of producing the gap, so the arithmetic is not where it fails.

It fails on the input.  Richard Wang, Samer Al-Saffar, Jeffrey Rogers and Matthew Hahn measured generation intervals directly in 2023, by reading them out of the mutation spectra of 1,247 sequenced trios and dated variants:

"The dominating pattern across the past 10,000 generations is a significantly shorter sex-averaged generation interval for East Asian, European, and South Asian populations—20.1 ± 3.9, 20.6 ± 3.8, and 21.0 ± 3.7 years—compared to the African population, 26.9 ± 3.5 years (P < 1 × 10⁻¹⁰, t test)."⁵

Africans have the longest measured interval of any continental group, by about six years.  That is the reverse of what the mechanism requires.  Two smaller problems sit on top.  The share of women married at ages 15 to 19, which is where Jeanson's 15-year figure comes from, is not a generation interval;  a generation interval is mean parental age at the birth of offspring across the whole reproductive span.  And a mechanism that works through generations elapsed gives no reason for heterozygosity to fall in a straight line against kilometres travelled.

Mechanism three: Africa was settled first and grew biggest

This one is not published in the creationist literature as a worked argument, but it is the first thing a defender reaches for, and it has already been tested in the primary literature.  Prugnolle, Manica and Balloux went looking for it in these same data:  "There is no obvious outlier and we could not detect any effect of the historical census size of the populations or any obvious macro-geographic pattern, such as a stepwise decrease in genetic diversity corresponding for instance to a severe bottleneck following the colonisation of a continent."²

The reason it does not work is worth stating in one sentence.  Over 4,000 years standing heterozygosity is inherited, not generated:  drift removes it at about 1/(2N_e) per generation, negligible once N_e reaches the thousands, and mutation adds it slowly.  A large population does not build diversity in 160 generations.  It only fails to lose what it started with.  So a downstream population cannot end up with more diversity than its source by growing large, and under the Babel model the Near East is the source.

The four-allele model, checked

Carter's clearest testable claim is about allele counts:  "most variations come in two main flavors (Adam and Eve could have given us up to four) and these are generally found in all populations.  When we find more variation than this, almost always the third, fourth, fifth, etc., versions are restricted to one subpopulation or another.  These are mutations that have occurred since the Tower of Babel."¹⁹

Run against the same genotype file:  the mean number of distinct alleles per locus is 11.93, the maximum at a single locus is 35, and 428 of the 783 loci carry more than ten alleles.  Ten is the post-Flood ceiling at an autosomal locus, since Noah and his wife supply four and the three daughters-in-law supply six, the sons carrying only their parents' copies.  More than half of these loci therefore need at least one post-Babel mutation, and many need a dozen.  That is not by itself fatal, and this article will not pretend otherwise:  microsatellites mutate at roughly 10⁻³ per locus per generation, thousands of times the point-mutation rate.  The claim fails on the geography of the extra alleles.  If they are post-Babel mutations they should be private to subpopulations at comparable rates everywhere.  Africa's rarefied private allelic richness is 0.168 against 0.017 to 0.031 elsewhere.

Carter's opening concession in the same piece is worth quoting on the other side, because it is correct:  "the worldwide human population in particular has a surprisingly low diversity across almost the entire genome.  This is the basis for the belief (among evolutionists) that humanity went through a near-extinction bottleneck event prior to the Out of Africa migration."¹⁹  Human nucleotide diversity is low for a mammal of this range, the mainstream agrees, and it calls the cause a bottleneck.  The dispute is about when and how deep.

The ICR reading, and the number it quotes

The 2012 ICR piece reads a study of African hunter-gatherer genomes as a post-Babel sequence:  free intermarriage at Babel, dispersal, Ice Age isolation in Africa, then recent diversification.²⁰²⁰  Two steps run against the argument.  Ice Age isolation reduces within-population heterozygosity, the quantity that has to come out high.  And the article quotes its source's count approvingly:  "their study found 13,407,517 SNPs that differ from the human genome reference sequence."²⁰  Lachance counted "variants (SNPs, insertions, and deletions …)" across fifteen genomes, not SNPs alone.²¹  Either way the number is a cost, not a support:  under the model being defended, every variant beyond ten per site has to arise in about 160 generations.

Part 6: The clock that needs no calendar

Every dating argument in this field can be refused by refusing its calibration.  Radiometric dates assume decay rates;  molecular clocks assume mutation rates.  A reader who rejects both has rejected the dates in the previous five parts.

Sriram Sankararaman, Nick Patterson, Heng Li, Svante Pääbo and David Reich published a measurement in 2012 that does not work that way.  When two populations interbreed, the donor DNA arrives in long unbroken chunks, and recombination chops them shorter at every meiosis.  Measure how far apart two Neanderthal-derived alleles can still be and remain linked, and you have counted generations, not years.  One of their two maps is the deCODE map, built by watching crossovers in living Icelandic pedigrees.

Their Table 2 gives the answer twice, from two independent maps:  1,805 to 1,993 generations since Neanderthal gene flow into the European line from the pedigree map, and 1,881 to 2,043 generations from the European linkage-disequilibrium map, as 95% credible intervals.⁴  The years in the abstract are looser than the generation count.  Multiplying the generations by a generation interval drawn uniformly from 25 to 33 years gives 47,000 to 65,000 years;  the wider 37,000 to 86,000 range adds a 15% allowance for a bias their simulations found in the ascertainment.⁴  That 15% applies to the generation count too, so read the floor as about 1,530 generations rather than 1,805.

A post-Babel dispersal of 4,000 years at 25 years per generation is 160 generations.  At 20 years it is 200.  The measurement is nine to thirteen times larger, and about eight times larger at its most generous reading.  Three escapes exist and each has a price.  A generation interval short enough to close the gap would have to be about two years, which is biologically impossible.  A much higher ancient recombination rate is a free parameter nobody has measured, and raising it breaks every genetic map used in medicine.  Denying the admixture happened requires another source for the Neanderthal-matching haplotypes, and this paper's test was designed to separate admixture from ancient population structure:  "The date of 37,000–86,000 years BP is too recent to be consistent with the 'ancient African population structure' scenario."⁴  No creationist author found in this project's search has addressed the generation count at all.

The tree shapes are argued in the companion articles, and their shape alone settles less than their branch lengths.  The mitochondrial root has two children, but three daughters-in-law do not need a three-way root:  their maternal lines all go back to Eve and can part from one another in two steps.  What counts against a young-earth history is depth.  Founders placed at the deepest nodes need 10 to 14 mutations in the 1,656 years before the Flood;  the measured rates supply 0.5 to 3.9, and only Jeanson's own rate, at its most generous, reaches 10.  And the non-African world hangs off M and N, two twigs of L3, and the path from the root to the root of M alone carries 39 substitutions.²²  A Babel history must fit nearly all of them into the century or so between the Flood and the dispersal.  Carter looked at his own figure of the Y tree and wrote:  "Do you see 'three' branches?  I can force my brain to see them, but this does not mean what I am seeing is correct."³  See mitochondrial Eve and Y-chromosome Adam for both in full.

One point here does favour the dispersal shape a Babel model predicts.  Karmin 2015 reports "a cluster of major non-African founder haplogroups in a narrow time interval at 47–52 kya."²³  A burst of founder lineages at one moment is the shape of a single dispersal from a single place.  The disagreement is about what sits beneath it.

Objections and responses

"Africa has the most diversity because it was settled first and grew largest."  Tested and rejected in print in 2005 on these data, by the authors of the second origin scan:  no effect of historical census size was detectable.²  Growing large lets a population keep what it started with;  it cannot lift a population above its own source.

"The origin scan assumes a serial founder model, so it is circular."  Partly fair, and Liu 2006 and Schlebusch 2021 make versions of the same objection.¹⁰¹¹  What the scan establishes is conditional:  if diversity was lost along the route of travel, the route started nearer Africa than Mesopotamia.  Two things limit the damage.  The same relation appears on three marker systems and on statistics that are not heterozygosity at all, including linkage-disequilibrium decay.¹²  And the Babel model predicts diversity loss along the route too, since it is also a dispersal from one place.

"The geneticists cannot even agree where in Africa."  True, by thousands of kilometres, and Henn's own F_ST scan reaches only P = 0.06.¹²  That is why this article claims no birthplace.  It moves no estimate off the continent, or within a factor of ten of a Flood chronology.

"Your own reproduction fails to match Ramachandran's published Figure 4A."  It did, and Part 2 now says which waypoint assignment closes it, to within 2% on both the R-squared and the slope.  No routing convention tried, including no waypoints at all, puts a Mesopotamian origin ahead of an African one across all 53 populations.

"Microsatellites mutate fast, so counting alleles proves nothing about founders."  Granted, and Part 5 grants it.  Twelve alleles per locus from ten founding alleles in 160 generations is not impossible at 10⁻³ per generation.  The argument rests on the distribution of the extras:  Africa 0.168 private alleles per locus, everyone else 0.017 to 0.031, at matched sample size.

"Archaic admixture broke Out of Africa."  It did break the strict replacement version.  But the paper creationist authors cite hardest explains its own result as back-migration of Eurasians into Africa, in its section heading.¹⁷  Tomkins reports that explanation and then leaves it standing without an answer, which is where the argument stops rather than where it goes wrong.

"So the data prove humans originated in East Africa."  No, and this is the objection to the verdict of this article.  Drop the African populations and the gradient stops distinguishing Ethiopia from Ararat at three decimal places.  What the measurement supports is a direction, with Africa at the diverse end, plus a depth of divergence among African populations that a 4,000-year model has no mechanism for.  A birthplace is more than the data carry.

Issues and unanswered questions

Two sources recorded as unobtainable in the first draft were obtained on the second pass.  Liu, Prugnolle, Manica and Balloux's 2006 AJHG paper is free at PMC1559480, and Lachance et al. 2012 is free at PMC3426505.  Both are now cited above.  Liu 2006 turns out not to be a third origin scan:  it assumes an East African origin and fits demographic parameters to it, giving an initial expansion 56,063 ± 5,678 years ago from a founding effective population of 1,064 ± 93.¹⁰  Its value here is the warning quoted in Part 3, not an independent location of the origin.

The private-allele asymmetry is the sharpest unanswered question, on either side.  Under a Babel model, Africa's five-to-tenfold excess of private alleles is a claim about relative mutation rates by continent, and no measurement of the required size and sign exists.  A young-earth answer needs one.  The nearest direct measurement available, Wang 2023 on generation intervals, supplies the opposite sign.

What a young-earth response would have to compute.  Nobody on the creationist side has published what a Babel dispersal predicts for the heterozygosity-distance relation.  The work is cheap:  the genotype file is 6.4 MB and public, and the fitting code used here is in this article's research/code/ directory.  A spatially explicit post-Babel colonisation model, with founder sizes and growth rates, would have to end with African H_e above Levantine H_e starting from ten founding alleles per locus.  No serial founder process does that, so the model would have to add a mechanism, and the mechanism then has to survive Wang 2023 and the F_ST table.

Where this article's own case is thinnest.  Three of the eight African populations were sampled from three, five and seven people, and the first two carry the largest residuals in the Babylon fit.  Dropping them halves the Addis-against-Babylon gap.  The HGDP-CEPH panel has eight African populations against nineteen East Asian ones, and Ramachandran says in print that denser African sampling might localise the origin differently.¹  The origin scan uses box-defined landmasses rather than a coastline mask, and no routing convention tried reproduces both of the paper's own published figures.  Liu 2006 says in print that linear regressions on gene diversities may not locate an origin at all.  The residual and private-allele results do not depend on the regression;  the R-squared table does.

Verdict

The dispersal did not start in Mesopotamia.  Refitting the gradient from every candidate origin gives 0.859 from Addis Ababa and 0.647 from Babylon, with Addis ahead in 99.97% of 20,000 bootstrap resamples, and under a Babylon origin all eight African populations sit above the line as a block 1.89 RMSE units high.  Africa's private allelic richness is five to ten times everyone else's at matched sample size, and Africa is the one region that combines high between-population differentiation with the highest within-population diversity, which drift cannot make.  The single measurement that no recalibration touches is Sankararaman's 1,805 to 2,043 generations of recombination since Neanderthal admixture, against 160 generations in 4,000 years.

What the young-earth side has right.  The cline among non-Africans alone does not distinguish Mesopotamia from Ethiopia, and this article's own computation shows it.  Liu, Prugnolle, Manica and Balloux warned in 2006 that a linear regression on gene diversity may not locate an origin, and that warning has never been answered.  Amos and Hoffman find two discrete bottlenecks rather than a hundred small ones.  Karmin finds a burst of non-African founder lineages in a narrow interval.  The mainstream cannot agree where in Africa, and the candidates are thousands of kilometres apart.  Chen 2020 showed that a decade of results finding no Neanderthal ancestry in Africans came from a method that assumed the answer.  Every one of these is a real point.  None of them touches the timescale.

What an honest young-earth position looks like here.  Say that a Near Eastern staging post for the non-African world fits the data, because it does, and note that Out of Africa says the same thing.  Then engage the two things that actually cost the model:  the position of the African populations relative to any Mesopotamian origin, and the generation counts.  Do not repeat the census-size rescue, which was tested and failed in 2005.  Do not repeat the shorter-African-generation rescue, which was measured in the opposite direction in 2023.  And do not report a paper's own explanation of its result and then walk away from it, as the archaic-admixture argument does.

Companion pages

• The literature — every source consulted, what it measured, and which way it cuts.
• Quarantine — the claims held back, and why.
• Mitochondrial Eve — the maternal clock, and why the mitochondrial root has two children rather than three.
• Y-Chromosome Adam and Noah — the paternal clock, and where the tree's one near-trifurcation actually sits.
• Neanderthals and Denisovans — who the archaic populations were; Archaic Admixture — what admixture does and does not show.
• Population bottlenecks — why allelic richness and heterozygosity behave differently after a crash.
• Babel, Language, and Human Dispersal: the hub for this article and its siblings.
• The Tower of Babel: the text of Genesis 11, its Mesopotamian setting, and where and when the genealogies put the dispersion.
• The Table of Nations: Genesis 10 name by name, and its test against ancient DNA.
• Language Families and the Tower of Babel: how many families there are, how old they are, and whether there is time since Babel for their diversity.

Sources