Out-of-Africa Phylogeny from Dental Polymorphisms

I discovered the solution to the problem I was working on. Rather I discovered that Hanihara was struggling with the same problem and that he had figured out the solution by 2008. The population history signal is contained in the neutral drift component of systematic variation in human phenotypes. By neutral, I mean derivation due not to diversifying selection but rather to founder effects and isolation-by-distance. There is definitely a population history signal in cranial and craniofacial characters. A weaker signal is found in phonemic data and postcranial skeletal morphology. All of the above, with the exception of linguistic data, are to a greater or lesser extant confounded by natural selection. And linguistic data has a weak population history signal, certainly at great time depths.

So I have been looking at craniodental data. Metric craniodental data (teeth diameter etc) contains as strong a population history signal as craniometrics. But as it turns out we can do much better. The real gold is in dental polymorphisms. Polymorphisms are discrete variants of phenotypes, eg, blood groups (called RBC polymorphisms), hair color (we get the ginger from Neanderthals), eye color, and so on. Such polymorphisms can usually be found in all populations. The differences in their frequency contains some information on population history. This is usually badly confounded by, say, sexual selection. Think of hair color or eye color.

In the case of dental polymorphisms, there are good reasons to believe that the population history signal is not confounded at all so that frequency differences between populations reflect shared inheritance. The basic operating logic is that of Mayr’s founder effect. All founder populations carry with them only a subset of the phenotypic diversity. Moreover, their smaller effective population size means even greater loss of diversity over time as lineages die out randomly.


Figure 1. Founder effect. Source: Wikipedia Commons.

In a recent paper in Nature, Hanihara and others showed that dental morphology contains an extraordinarily strong population history signal that is highly correlated with the best estimates from molecular anthropology. So this is a very promising line of investigation.

Let me illustrate with some frequency distributions for some of these dental polymorphisms. Figure 2 displays the incidence of shoveling of the first Upper Incisor (UI1) by macro region. Why has this morphotype reached near fixation in northeast Asia and the Americas? Why do the western old worlds and eastern old worlds cluster together and on opposite sides of Africa which is right in the middle? The right answer lies in the details of the Out-of-Africa dispersals. Phylogeographer Stephen Oppenheimer argued forcefully in Out of Eden for a thick version of the southern route or beachcomber hypothesis whereby there a single exit from Africa to India, and it was from the subcontinent that Homo sapiens heading west to Europe, south and east to Sahul, a branch of whom instead went up the coast to China. The morphotype achieved near fixation in northeastern Asia before the founder populations migrated to the Americas across Alaska. Moreover, Oppenheimer argues that the big freeze of the Last Glacial Maximum about 20ka decimated northern populations and reduced their genetic diversity. The actual fixation of this morphotype in northern America may reflect this natural history.


Figure 2. UI1 shoveling incidence rate by macroregion.

In the case of the awkward-to-pronounce Hypoconulid polymorphism, we can see the eastern and western branches going out from India both became derived; in the opposite direction. The morphotype became more frequent in the west and less frequent in the east, with a particularly severe decline in frequency in Australia and northern America. LGM strikes again?


Figure 3. Frequency of the Hypoconulid morphotype.

The premolar accessory cusp morphotype is rather rare outside Australia and Melanesia, suggesting that this is a derivation in these populations after they split from the others. This ancient clade also stands out in metric craniodental traits. The derivation is largely a function of the great time-depth of the split. It is no surprise that geographic isolates display marked derivation in the human species. It holds across the animal kingdom.


Figure. Frequency of premolar accessory cusp morphotype.

Almost simultaneously as their paper in Nature, Hanihara and gang published an extraordinary study in Current Anthropology, where they showed that of all the competing Out-of-Africa dispersal scenarios, Oppenheimer’s beachcomber single-dispersal Out-of-Africa via India scenario (“BSD” in the next figure) is the most consistent with the phylogeny obtained from dental polymorphism frequencies.


Figure 4. Out-of-Africa dispersal scenarios. BSD = beachcomber arc single dispersal; EE = eastward expansion single dispersal; MD = multiple dispersals; MDI = multiple dispersals and Australo-Melanesian isolation.

I like to check things with my own two hands. I used the Hanihara (2008) data on dental polymorphisms to extract a phylogram that reflects our population history. In order to compare the frequency data from different polymorphisms, I converted them to percentile scores. Then I used the Euclidean metric to obtain pairwise phenotypic distances, from which I obtained the phylogram using the standard neighbor-joining algorithm. The accuracy of the resulting phylogram is simply astonishing. I hit pay-dirt alright.


Figure 5. Phylogeny from dental polymorphisms.

One can read off our population history from this diagram. Homo sapiens left Africa via the Red Sea route to India 100-80ka. While many founders stayed behind on the subcontinent, some beachcombers kept going further east. These populations reached Sahul by 60ka. The Negrito populations of the Andaman Islands and Melanesia, and the Australian aboriginal population are actual relict populations from that original dispersal. (BTW, the Negritos aren’t black and fizzy haired because they are close to Africans. Their dark skins reflect directional diversifying selection under similar tropical conditions; like the different origins of depigmentation on either extremity of Eurasia, it’s the classic case of a homoplasy.) On their way to Sahul, some of these founders stayed back in southeast Asia which at time was a large continent called Sunda. Some of them would later go on to become Polynesians and the greatest sea-farers the world has ever seen. Others went up north to northeast Asia and later from there on to the Americas soon after the Last Glacial Maximum. Before all these developments in the far east, pioneer Sapiens ventured forth north and west from India, taking the trans-Caucasus route to Europe where they would go on to “replace” Neanderthals. The move back into Africa (the Ethiopians are the descendents of this reverse migration) followed after southwest Asia was finally peopled (it had been a forbidding desert before).

So here’s the kicker. When Dravidians (9ka) and later Indo-Europeans (4ka) reached India, it was a reunion long in the making. What is astonishing is how this turn table between India and Europe has been turning throughout deep history. In a sense, the paleogeographic logic of this pattern puts the flesh on the bone on Diamond’s insight — that the east-west Eurasian axis was more advantageous than the north-south axes of Africa and the Americas for the simple reason that people, ideas, and technology (and germs) could move faster against the latitude gradient than along it. What is amazing is the emerging picture of how Diamond’s logic worked in practice — at least where pots were moving with people (one can hardly demand more of physical anthropology).

Oppenheimer argues that the Upper Paleolithic revolution in Europe (it is to paleoanthropologists what the Industrial Revolution is to modern economic historians and the Neolithic revolution is to prehistorians) was the handiwork of founders from the Indian subcontinent. He traces both the people who authored the Aurignacian culture c. 43ka as well as the later arrivals who authored the Gravettian c. 35ka, the great mammoth hunters of north-central Eurasia, to the subcontinent. But that deepens the paradox of the Upper Paleolithic. If the authors of the Aurignacian came from the subcontinent and so did those of the Gravettian, they why does the Upper Paleolithic not reach India until tens of thousands of years later? Was it combined and uneven development structured by our Heliocentric geometry? Or was something even deeper at play. That’s the big open question of paleoanthropology.



Indo-European Phylogeny

In the previous dispatch I tried to extract a Pleistocene population history signal from phonemic data for subaltern languages. Does this approach actually work? One way to check is to run the exact same algorithm for a well-known phylum. In what follows we’ll extract a phylogram for Indo-European. As we shall see, there is a very strong population history signal in phonemic distance metrics.

The Indo-European family is roughly only five thousand years old. It spread across Western Eurasia shortly after the ethnogenesis of the Yamnaya — triggered by the introduction of advanced Sumerian technology, in particular, the wagon and probably Brewer’s yeast, that made the economic exploitation of the steppe possible for the first time. (See David Anthony’s excellent detective work.) We know this not only from massive whole genome studies but also from overwhelming archaeological evidence. The Yamnaya were a rank society of warrior-pastoralists much given to feasting, drinking, and all-round boisterous male bonding rituals. (“Will you fight with me brother?”) Their diagnostic signature in Europe is that male warrior elites are buried with their weapons and ornaments signifying their social status. At home in the steppe and in the more advanced Near East and later India, they can identified by their chariots. Extraordinarily, there were major chariot manufacturing centers in the steppe that supplied the chariot civilizations of the Near East in the second millennium BCE, piggy-backing on the brisk horse trade. More on that later. Let’s not get tied down by the extraordinary discoveries of prehistorians.

We begin by testing that Indo-European phonemic diversity is characterized by isolation-by-distance. The Mantel test statistic for pairwise phonemic and geographic distances equals 0.274. The probability of seeing this by chance is less than one in 100,000 (p<10^5). So there is very good reason to think that phonemic distances contain a strong population history signal.

Figure 1 displays a reduced form version of the full phylogram. What I have done is fold the subtrees corresponding to the main subfamilies. For example, the “North Indian” branch at the very top of the phylogram is a clade with 17 languages that are closely related and geographically centered in north India (Hindi, Gujarati, Marathi, and so on). The “NW Indian” branch below it is a clade with 13 languages that cluster around Pakistan and the Indian northwest (Punjabi, Baluchi, Sindhi, and so on). That the algorithm correctly classifies these languages (and other subfamilies) together is assuring.


Figure 1. Top level breakdown of the Indo-European family.

What the phylogram shows is that the split with the greatest time-depth is the East-West split between the Indian and Western branches. This is as it should be since we know the Yamnaya migration pulses east and west were near contemporaneous and took place almost immediately after the ethnogenesis of the Yamnaya.


Figure 2. Yamnaya expansion 5ka. Source: Reich (2018).

So the phylogram is successful at discerning population history and branching of the Indo-European family at both fine-scale and the top-level splits with the greatest time depth. Where it fails badly is in assigning Latin and Sri Lankan together with the Swedish-Norwegian clade. The suggested close relationship between Breton and ancient Zoroastrian is less than persuasive. Modulo these anomalies, the phylogram is extremely compelling. Figure 3 displays the beast in all its glory. Study it at length and it quickly becomes apparent that the accuracy of the phylogram is simply astonishing.


Figure 3. Indo-European phylogram based on phonemic distance.

So, yes, there is a very strong population history signal in phonemic distances. We are not deluding ourselves that we can recover Pleistocene population history from subaltern languages. The issue is whether the phylograms thus derived are consistent with those obtained from physical anthropology and paleontology. As I mentioned in the previous dispatch, the disconnect between the two is the most important outstanding puzzle of modern research into our origins.

On a separate note, what emerges from this analysis is the extraordinary position of shatter-belts. Language isolates packed into New Guinea (as we saw in the previous post), Iran, Afghanistan etc attest to the history of subaltern populations that were once widespread but have since been pushed into marginal zones. Scott made a compelling case in The Art of Not Being Governed that the ethnolinguistic fragmentation of mainland southeast Asia reflects centuries of attempts at state formation and resistance from below. That is plausible for Iran and Afghanistan (and certainly the Balkans). But it is not a plausible explanation for the extraordinary ethnolinguistic fragmentation found in New Guinea. What is?



Phonemic Phylograms for Subaltern Languages

The picture of population history that emerges from physical anthropology (molecular, craniometric, craniodental, etc) is consistent with an Out-of-Africa model of the ethnographic and historic present. All non-African populations descend from anatomically modern (Homo sapiens sensu stricto) founder populations that dispersed from Africa 130-40ka. The evidence from paleontology (fossil molecular, craniometric, craniodental, etc) complicates this picture. There were evidently successful interspecific families (if Sapiens, Neanderthals, Denisovans, etc, are regarded as species) or interracial families (if they are regarded as allopatric subspecies) — the DNA of non-African people can only contain sequences acquired from them if their issue survived to procreate. Put another way, all lineages of the other taxa encountered by Sapiens could not have vanished without issue for otherwise their signature could not still be found in contemporary populations. [So these were at best allospecies.]

The picture is complicated further when we look at the archaeological evidence. The evidence is inconsistent with Out-of-Africa in a thick sense — that Sapiens replaced Neanderthals sensu lato because they were behaviorally modern and the latter were not. There is no evidence of modern behavior in the assemblages associated with anatomically modern humans within and without Africa for tens of thousands of years after the “speciation” of Sapiens and the subsequent Out-of-Africa dispersals. The onset of modern behavior is late, staggered and impossible to reconcile with the simple Human Revolution story laid by Klein, Mellars, Stringer, Gamble, Tattersall and others from the late 1980s onwards. (Although Gamble seems to have since changed his mind.)

A further source of evidence information is linguistic. Cavalli-Sforza and others began to show in the 1990s that linguistic phylograms resembled those derived from physical anthropology. Atkinson remade the case in the last decade. It has since been debunked, among others by Crenza et al. (2015).  Phylograms extracted from linguistic data are not consistent with phylograms obtained from physical anthropology. Why should that be? Something very interesting is going on with this disconnect.

The problem with phonemic data is not that the population history signal is confounded by unstable rates of innovation. As we shall see, this is not the problem. The problem is rather that the Holocene Filter confounds the Pleistocene population history signal. Most people on the planet today speak languages in a small handful of families (Indo-European, Sino-Tibetan, Bantu, Nilo-Saharan, Dravidian, Austronesian etc) that underwent massive geographic and population size expansions during the Holocene. Their ethnogenesis of these families was trigged by the agricultural (9ka) and pastoral (5ka) revolutions. Contemporary populations in Eurasia, Africa and elsewhere as well as those of the ethnographic present are descended from very recent migrants (the men more so than the women since the migrations were always sex-biased) superimposed over yet older strata of Holocene expanders. Underneath these massive boulders are ancient populations of hunter-gatherers like the San, the Andamanese and thousands of others who survive as isolates in deserts, on islands, in dense forests and mountain redoubts, and suchlike. If we are interested in Pleistocene history, we need to isolate and study the substratum of hunter-gatherer populations of the ethnographic present. Here we make such an attempt.

In order to recover Pleistocene population history we have to figure a way of controlling for the Holocene filter. I think there is a simple method that can work if we have a large enough phonemic database. Populations (roughly identified as the ancestors of the speakers of language families) that underwent Holocene expansions can be expected to be large today for that very reason. This means that we throw out all the big language families our sample will become more representative of the ancient substrata.

We examine the phonemic data collated by Crenza et al. (2015) from the Ethnologue Database. After throwing out language families with speaker populations larger than those of the Hmong-Mein family (who were largely overrun and driven out of China and into the shatter-belt of Indochina by the Han) and New World populations (who are known to have reached the New World after the Last Glacial Maximum) we obtain a sample of 103 languages that have been classified into 19 languages by linguists. These have a good claim to be direct descendents of languages spoken by Pleistocene populations in the Old World and Sahul before the Mesolithic (New World) and Holocene expansions. Figure 1 displays the latitude and longitude of these languages.


The dataset consists of Boolean variables denoting the presence or absence of 728 phonemes (vowels and consonants). Phonemic distance is computed from the number of shared phonemes (the Hamming metric).

We begin by testing that isolation-by-distance explains pairwise phonemic distance. The appropriate test is a Mantel test comparing geographic distance (computed by the Haversine formula using known waypoints) and phonemic distance. We obtain a robust test statistic equal to 0.585. The probability of observing this value of chance is less than one in a million (p<10^-6). At the level of language phyla (Crenza et al. report the highest language phylum for each language, not family per se), the Mantel test statistic is a still robust 0.350 and statistically significant (p=0.041). So phonemic distance displays the same isolation-by-distance as physical anthropology distances. We can thus be confident that phonemic distance contains a population history signal.

Figure 2 presents the phylogram (lineage tree) obtained at the level of language phyla as the languages are classified today. We derive the phylogenetic tree from sequential neighbor joining algorithm.


Figure 2. Phylogram based on phonemic distance. Source: Creanza et al. (2015), Ethnologue Database, author’s computations.

What stands out is the isolation of the Hmong, the Khoisan and the Indian Aborigines (whose location places them in Jim Corbett national park in the foothills of the Himalayas). Interestingly, the Australians, with their unusual languages (they don’t have word order) are placed next to the northeastern Siberians, the Chukotko-Kamchatkan language family. Andamanese is closest to West Papuan suggesting that this subaltern population was once widespread across the southern dispersal route.

Can we recover the language classification (ie, the “families”) by looking at phonemic distance at the level of languages? Figure 2 displays the phylogram for families. For each language, we display the family, the ISO code for the language, and the latitude and longitude in parentheses.


Figure 3. Phylogram based on phonemic distance. Source: Creanza et al. (2015), Ethnologue Database, author’s computations.

The first, and most reassuring thing to note is that the language families as identified by linguists are largely placed together. The Australians have a complicated structure but they are classified together (the bottom half of the phylogram above the Hmong). Ditto the Hmong and the Khoisan. The fact that recognized phyla are generally classified together is very strong evidence of a population history signal in phonemic data.

The big anomaly is the classification of New Guinea languages. Unlike Australian, Khoisan, and Hmong languages, the Trans-New Guinea phylum does not cluster together. Rather there seem to be meaningful multiple clusters of languages that have been classified as falling within the Trans-New Guinean family. This is quite possibly due to the fact that New Guinea served as the shatter-belt par excellence in our deep history. What this phylogram suggests is that scholars may have misclassified New Guinea languages; in particular by not recognizing enough language families on this extraordinarily diverse island.

In Figure 4 we have folded up the tree to reveal the underlying big relationship together with the great anomalies. The collapsed subtrees and branch nodes are labeled in blue. Those that are not labeled “Branch #” are subtrees with many languages in the same family underneath. The anomalies are interesting. Why is the language of Indian aborigines close to Khoisan languages in Africa? Why is one Andamanese language classified with Sahul languages and another with Basque in Europe and the Hmong in China? Even more intriguing is the phylogenetic affinity (surely spurious) between the Hmong and Australian families. So some of these anomalies are probably random. But others correspond to actual population history. Recall that the speakers of the predecessors of these languages probably occupied much larger areas than they do at present. The most striking feature of course is the phonemic incoherence of languages folded into the Trans-New Guinea. Above all else, what it attests to is the sheer variation in New Guinea. Could the island have served as a shatter-belt in the Pleistocene?


Figure 4. Manually folded version of Figure 3. Only subtrees with languages from the same phylum have been folded.

This is a work in progress. My goal is to cointegrate the information from physical anthropology, paleontology, archaeology and linguistics to tell a more compelling history of the Pleistocene than has so far been on offer. Bear with me.




Notes on the Anthropology Wars: Evidence from Prestige School PhD Placements

Physical anthropology began as racial anthropology—as the science of race. Although Boas and his students had been challenging scientific racialism since the turn of the century, the Boasian critique did not become politically significant until the 1930s when Ruth Benedict and other students of Boas at Columbia University began to mount a frontal challenge. Even after Auschwitz, racialism refused to relinquish its hold on the scientific mind. It was not until after the antisystemic turn of the 1960s that Boasian antiracism got the upper hand on racial anthropologists in the larger discourse. That turning point can be dated quite precisely. It was when Coon was pushed out as the President of the AAPA in 1963. Still, it is clear that the scientific status of race remained contested territory since otherwise Lewontin’s famous intervention in 1972 (showing that racial taxonomy could explain no more than a negligible fraction of human genetic variation) would not have been necessary. By 1994 certainly, Boasian antiracism had become hegemonic in the larger discourse, as became evident in the controversy surrounding the publication of The Bell Curve.

Having narrowed down the rise of Boasian antiracism to between 1972 and 1994, I have tried to pin down the periodization further. My working hypothesis, based on readings in paleoanthropology, is that the turning point — the onset of hegemonic Boasian antiracism — can be dated very precisely to 1987 when novel radiometric and molecular evidence resolved the long-standing debate between partisans of Multiregionalism and Out-of-Africa theories of the ethnographic present. Put simply, it became impossible to deny that all present-day non-African populations were descended from populations that had dispersed recently from Africa. The upshot was that not enough time had elapsed since the Out-of-Africa dispersals for allopatric subspecies or continental races to develop isolating mechanisms — understood certainly after Mayr (1963) as the dominant process by which continental races emerge in the animal kingdom. Boasian antiracists could then shout down racial anthropologists by pointing to the increasingly unambiguous evidence for the recent Out-of-Africa hypothesis. It thereby became the go-to weapon in the antiracist arsenal.

Jeremy Kessler suggested that I narrow down the scope of the enquiry so as to be able to craft a tighter narrative of this intellectual history. That’s very good advice. There is a hell of lot going on in the Western imaginary in the race question, especially in the United States. It is simply too ambitious to attempt a broader history of everyday racial anthropology so to speak. While it may have broader implications, I will confine my attention to the scholarly discourse within anthropology.

So what happened in anthropology between the 1960s and the 1990s?

Part of what happened is that anthropologists began to realize that racial taxonomy provided a poor handle on human biological variability. In this process, Lewontin’s intervention and subsequent analyzes by other physical anthropologists were crucial. As important was the rise of population thinking. It became more and more obvious that the right unit of analysis to interrogate human variability is not continental races, subraces, or any such essentialist taxa, but rather naturally-occurring situated breeding populations or demes. With the possible exception of geographic isolates, there is always gene flow between demes largely as a function of distance, so you get diagnostic clines for polymorphic traits reflecting isolation-by-distance and diversifying selection. That’s the point of departure to understand human biological variability.

What this means is that phylogenetic trees that suggest racial classification to Nicholas Wade and other neoracialists, do not in fact reflect the existence of races since there is always population structure for any taxonomic level above that of the deme. That is, the interpopulation variation within these taxa (the hypothesized races) is more often than not greater than that between the taxa. For instance, there is no such thing as the Australian race because there is more interdemic variation within this taxon than without. So the taxa in the phylograms I constructed from craniofacial and cranial data are not allopatric subspecies or races since populations within them differ as or more dramatically from each other as they do as a group from others. The best way to see this is to “zoom in”. The point is brought home forcefully in the following figure. Panel A illustrates the population structure of Eurasia using the first two principal components, while panel B zooms into Nepal’s neighborhood. Forget about there being a South Asian race (or North Indian and South Indian or whatever). Even among the Nepalese the Magar deme is closer to the East Asian cline, while the Chhetri deme is closer to Indian Brahmins. This kind of population structure is characteristic of human biological variation. It makes mincemeat out of racial taxonomy.


But what happened in anthropology was not simply scholars changing their mind about the usefulness of racial taxonomy. No sir. What actually happened was more like a Thirty-Year War. Simply put, physical anthropology was gutted. Boasian cultural anthropologists mounted relentless attacks on their colleagues in physical anthropology, who they regarded as overwhelmingly racist whatever their politico-ethical stance. Relatively fewer and fewer were given tenure. Many were driven out of the Ivory Tower. Ever fewer were hired to replace retiring professors. This one-sided warfare unfolded between the 1960s and the 1990s, until Boasian anthropologists established unchallenged supremacy in anthropology departments.

The gutting of physical anthropology would later come to bite when incoming results from molecular anthropology (that tellingly emerged in biology not physical anthropology where it actually belongs) would prompt a revival of racialism. Anthropologists — having virtually forbidden the scientific study of human biological variability — would find themselves without sufficient resources to meet the challenge. With postracial physical anthropology stunted by the decades of neglect and scorn, the interpretation of the results in molecular anthropology would be left to geneticists caught up in the self-congratulation of DNA supremacism and amateur neoracialists with little understanding of human variability. But I am getting ahead of the story.

I have been looking for ways to test these working hypotheses. Recently I found a study looking at PhD placements in anthropology. The big story there is that only one in five PhDs in anthropology as a whole finds academic employment in degree granting institutions. Prestige schools obviously do much much better than their less prestigious counterparts. (Prestige schools here refers to a dozen top schools that account for a large share of PhD production and placement in anthropology: Chicago, Berkeley, Harvard, Michigan–Ann Arbor, Columbia, Stanford, UCLA, U Penn, NYU, Yale, Cornell, and Princeton.) With the expansion of PhD programs elsewhere, as the next figure shows, their share of placements have been declining since the 1970s. Prestige school PhDs secured almost half the academic positions in 1920-1972. Their share has since dropped to a third. That’s still a lot.


The next figure shows the share of PhD placements in prestige schools by discipline. Note that Boasian/cultural anthropology placements are on the right axis. We can see that the share Boasian anthropology increased sharply in the 1990s and 2000s. By the time we get to the 1990s, more than two-thirds of all anthropology PhDs placed by prestige schools were Boasian anthropologists. By contrast, the share of physical anthropology collapsed from a fifth to an eighth. But the two movements are not congruent. The collapse in the physical anthropology share took place in the 1970s and 1980s. What’s going on?


The next figure further illustrates the total number of PhDs placed by discipline. The absolute number of placements in Boasian anthropology skyrocketed in the 1970s and 1980s from less than 50 to above 200 (Cf. previous figure). The dramatic academic expansion of the 1970s lifted all boats to some extent. But the rise of Boasian anthropology is manifest. Not only did Boasian anthropology corner all the massive gains in the 1970s, it continued to expand through the 1980s and 1990s. And even when tide went out with the onset of the Great Recession, Boasian anthropology lost relatively little compared to physical anthropology.


Of course, this sort of numbers game cannot prove my case. For that I need to interrogate the anthropology discourse at length. But I think these numbers do provide some empirical foundation for the story of the rise of Boasian antiracism in the intellectual history of American anthropology.


Effective Constraints on Sovereign Borrowers

Krugman traces the idea masquerading as theory (“Modern Monetary Theory”) to Abba Lerner’s “functional finance” doctrine from 1943:

His argument was that countries that (a) rely on fiat money they control and (b) don’t borrow in someone else’s currency don’t face any debt constraints, because they can always print money to service their debt. What they face, instead, is an inflation constraint: too much fiscal stimulus will cause an overheating economy. So their budget policies should be entirely focused on getting the level of aggregate demand right: the budget deficit should be big enough to produce full employment, but no so big as to produce inflationary overheating.

Simply put, the idea is that sovereigns with obligations in their own fiat currency are constrained only by inflation in how much debt they can pile up. Krugman points to the potential problem of snowballing debt whereby debt servicing claims a larger and larger portion of the public purse. But this is a function of interest rates:

If r<g, which is true now and has mostly been true in the past, the level of debt really isn’t too much of an issue. But if r>g you do have the possibility of a debt snowball: the higher the ratio of debt to GDP the faster, other things equal, that ratio will grow. And debt can’t go to infinity — it can’t exceed total wealth, and in fact as debt gets ever higher people will demand ever-increasing returns to hold it.

So here we have another effective constraint besides inflation. How much debt sovereigns may pile up is a function of the compensation demanded by investors. Now this compensation is not uniform across borrowers. Far from it. Figure 1 displays spreads against the German bund for selected sovereigns.


Figure 1. Selected sovereign bond spreads.

The United States is much further along in the monetary cycle than the eurozone. But why do Italy and Portugal have to pay so much higher to access capital markets than Germany? The next figure shows that there is no relationship between debt-to-GDP ratios and sovereign bond yields. The rank correlation coefficient is not only insignificant but bears the wrong sign (r=-0.15, p=0.40). Restricting the sample to so-called emerging markets does not affect the result (r=-0.04, p=0.82). Note that we have already excluded Argentina (debt ratio of 57% and bond yields at 26%) and Japan (debt ratio 253% and yield 0%) since they are clear outliers. So there is simply no evidence than bond markets pay much attention to the debt burden of sovereigns.


Figure 2. The data is from 2017.

So how do bond markets judge sovereign borrowers? The short answer is that yields compensate bondholders for a number of perceived risks. Sovereigns may default, inflation may erode the value of the bond, exchange rate movements may impose losses on bondholders, interest rates may rise and thereby reduce the value of their bond. Moreover, bond yields reflect compensation for not just the expected value of the bond but also for the risk that the value may deteriorate, if for no other reason than that markets can be fickle (so that tomorrow you may not be able to sell your bond for the price you paid for it even if the price you paid was considered by all to be fair today). All these risks are constantly reevaluated by markets. The diachronic pattern is controlled by the market price of risk, itself a function of risk appetite in global markets. The synchronic pattern on the other hand is controlled by the status of sovereigns.

Some sovereigns are regarded by bond markets as safe asset providers. I have identified some safe-asset providers in Figure 2. The main one missing is Japan which would be far out to the right and bottom. Because the sovereign debt of safe asset providers is perceived to be credit default-remote and relatively protected against inflation and exchange rate movements, these assets can serve as collateral in the wholesale funding flywheel, the core of global financial intermediation. It is the practices of institutional players in this ecosystem that determines who is and who is not a safe asset provider. Safe assets can be identified by what happens to yields when the market as a whole tanks. The diagnostic pattern is that when shit hits the ceiling safe assets go up in value as investors flee to safety.

Hélène Rey has identified the curse of the regional safe asset providers. These are small countries whose debt is regarded as safe in wholesale banking practice. Even if their central banks would like to push up yields (say to defend their currency or fight inflation), adverse market developments may send them tumbling down. This is what happened to the Swiss central bank in 2015. Regional safe asset providers

… face a variant of the old ‘Triffin dilemma’: faced with a surge in the demand for their (safe) assets, regional safe asset providers must choose between increasing their external exposure, or letting their currency appreciate. In the former case, the increased exposure can generate potentially large valuation losses in the event of a global crisis…. In the limit, as the exposure grows, it could even threaten the fiscal capacity of the regional safe asset provider, or the loss absorbing capacity of its central bank, leading to a run equilibrium. Alternatively, a regional safe asset provider may choose to limit its exposure, i.e. the supply of its safe assets. The surge in demand then translates into an appreciation of the domestic currency which may adversely impact the real economy, especially the tradable sector. The smaller the  regional safe asset provider is, the less palatable either of these alternatives is likely to be, a phenomenon we dub the ‘curse of the regional safe asset provider.’

Large safe asset providers on the other hand are not so cursed precisely because of their size. But the more important point for our purposes is that big safe asset providers (Germany, Japan, and above all, the United States) are not, and in fact, cannot, be punished by bond markets for fiscal profligacy. The reason for that is the structural shortage of safe assets in the global financial system.

Screen Shot 2019-02-13 at 2.39.40 PM.png

Figure 3. The consumption-to-wealth ratio, a measure of the financial cycle, predicts global real rates. Both are computed as US-UK wealth-weighted averages. Source: Farooqui (2016).

The issue is not whether “MMT” holds in some toy model. The issue is to what degree sovereigns are disciplined by the bond market. The answer to that question depends on their structural position that is in turn determined by wholesale banking practice. The big safe asset providers — the United States, Germany, and Japan — face considerable slack in bond market discipline because the world can’t get enough of their debt. Sovereigns not thus privileged are more exposed to bond market discipline. They may indeed have to worry about market perceptions of their public finances.

With inflation still pretty much dead and policy rates pretty much still on the floor, there is simply no case to be made for fiscal discipline for the big safe asset providers. In effect, big safe asset providers are not debt constrained. And this state of affairs will continue until the global financial system is transformed beyond recognition. It is in this sense I believe that Adam Tooze champions “MMT.”

I am not suggesting that President Warren should go on a debt binge. But worrying about bond market discipline for fiscal profligacy is to worry about precisely the wrong problem. The United States can afford to double its outstanding debt-to-GDP ratio from 100 to 200 percent and will still be perceived as less risky than Japan.



Phylogeny from Craniometrics

As we saw in the previous dispatch, different craniofacial characters are variously under the control of neutral drift, sexual selection, and thermoregulatory adaptation to the paleoclimate. I suggested that kosher inference of phylogeny (ie lineage) is difficult because the population history signal is confounded by natural selection. One way to go about this would be to control for dimorphism and absolute latitude. That doesn’t seem to work. One gets nonsense trees. Might not there be another way?

The law of large numbers dictates that if we average over a large number of characters, second-order factors will get averaged away thereby revealing the dominant, first-order term. There is good reason to believe that the controlling factor for cranial morphology is neutral drift. As we shall, this turns out to be right. In what follows we shall obtain phylogenetic trees from Hanihara’s and Howells’ craniometric datasets. The idea is to standardize all characters to have mean 0 and variance 1, and use distance measures between populations in this space to back out the underlying phylogenetic relationships.


Figure 1. Phylogeny from craniofacial characters. Source: Hanihara (2000), author’s computations.

Figure 1 displays the phylogenetic tree obtained from the Hanihara sample. See from the top-right, at the base of the tree. The first cluster of 7 correctly identifies western Eurasian phylogeny: eastern and western Europe are closest to western Asia; Europe, north Africa, and western Asia split last from northern and southern India. The second cluster of Sahul, Negrito and southern Africa identifies the really ancient populations. The split between this group and others is correctly identified as having great time-depth. The bottom supercluster correctly identifies the eastern world, although the tree places the East-West split as having as great a time-depth as the San-Sahul split (which is known to be older). But the subdivisions within the eastern supercluster are correctly identified with the possible exception of the close phylogenetic relationship between circumpolar peoples and Polynesians. All in all, not bad for just half-a-dozen craniofacial characters.

The Howells dataset has 82 linear measurements each for 2,524 from thirty populations. The sample is thus wide enough for the averaging method to really work. Figure 2 displays the longitude and latitude for the populations in the sample. This will help us identify whether the inferred tree makes sense.


Figure 2. Locations of demes in the Howells sample.


Figure 3. Phylogeny from craniometrics. Source: Howells Craniometric Dataset, author’s computations.

The accuracy of the inferred phylogenetic tree is simply astonishing. Read from bottom-right. The Sahul peoples (Australia, Tasmania, and Tolai) are correctly identified as having split last from Bushman and Zulu; the great-time depth of the Andamanese and its close relationship to the ancient clade likewise. (Perhaps it is best to mentally pull the ancient cluster to the left and place them at the root of the tree.) The precise pattern of the eastern cluster (the next 13 above the Andamanese) is exactly right. The phylogenetic relationships in the Austronesian cluster from New Zealand to Easter Islands (a result of recent Holocene expansions) are correctly identified in detail, as is the cluster’s close phylogenetic relationship with the eastern cluster. Further up, the Americans (Peru, Santa Cruz, Arikara) are mixed up with medieval Austro-Hungarians (Berg, Zalavar). Similarly, while the circumpolar peoples are identified as having recently split, as are Dogon and Tieta, they are placed next to each other and to the ancient Egyptians and the medieval Norse.  This may be because of the diachronic patterns in morphology, such as those we identified in the European case. In any case, every recent split in the phylogram (the last of the forks) is accurate. The great time-depth of the ancient cluster is spot-on.

So the algorithm does an excellent job in predicting phylogeny. Although second and third-level branches are sometimes confounded (particularly for ancient and medieval populations so this is presumably due to diachronic patterns). But what is clear is that craniometric distance is an excellent signal of phylogeny, suggesting both that cranial characters are under tight genetic control and that neutral drift is the controlling factor in craniometric variation.

Finally, we check that postcranial osteometric data does not predict phylogeny as well. Presumably this is because either postcranial skeletal morphology is not under tight genetic control, or it is so but neutral drift is not the factor controlling osteometric variation. Whatever the case may be, as shall see, the population history signal in the postcranial skeleton is relatively weak and easily confounded by selection.


Figure 4. Phylogeny from osteometrics. Source: Goldman Data Set, author’s computations.

We look at the Goldman osteometric dataset. We restrict the sample to the Old World and apply the same algorithm as before. Figure 4 displays the phylogram thus obtained. Although most recent splits are not too far off the mark, the prediction is unimpressive. Malaysia ends up with the Europeans, Australia with Madagascar, Tasmania and South Africa with China and the Philippines, and the Andamanese with the Congolese and the Indonesians! The algorithm thus fails catastrophically in predicting phylogenetic relationships between populations.

In sum, kosher inference of phylogeny is possible from craniometrics but not from osteometrics. This suggests that the former is under tighter genetic control than the latter, and/or contains a stronger population history signal that is less confounded by natural selection.

Postscript. The Howells phylogram can be improved by including the second moments. I figured out that the reason Hanihara’s craniofacial data yields such a convincing phylogram is because it contains both means and variances of the measurements. The second moments also contain phylogenetic information since the variance of metric characters and the second moment of the frequency of neutrally-derived discrete polymorphisms is a function of distance from Africa. Indeed, the inclusion solves the problem in our previous estimate.


Figure 5. A more precise phylogram from craniometrics. (Corrected version.)

The new estimate is hard to argue with. The Teita and Dogon are correctly identified as closest to Zulu and Bushman. All others are descendents of this ancient cluster of demes around the San. The San-Sahul split occurs at great-time depth, followed by the ancient split with the Andaman Islanders. The medieval Austro-Hungarians (Berg and Zalavar) are accurately placed closest to the medieval Norse and the ancient Egyptians. The Americans (Arikara, Santa Cruz, and Peru) are correctly showed as closely related to eastern Eurasians (Japanese, Atayal, Hainan, and the Philippines). The Austronesians are placed close to the Anyang and the circumpolar peoples (Eskimo and Buriat). Although the algorithm does get the time-depth of the most ancient splits wrong. The ancient subtree including the San and the Andamanese has the greatest time-depth (60ka), the Euro-Asian split has the second greatest time-depth (45ka), then you have the Americans and the circumpolar people splitting off (14ka), and finally the Austronesians split off from Taiwan (5ka). The phylogram, although highly accurate in detail, gets the greatest time-depths catastrophically wrong.


Do Craniofacial Characters Reflect Neutral Drift, Climatic Adaptation, or Sexual Selection?

New York University Primatologist James Higham delivered an extraordinarily interesting talk on primate reproductive ecology at the Natural History museum this week. He pointed out that otherwise morphologically quite similar primate species can coexist sympatrically without interbreeding — some half a dozen of them in one particular rainforest. The vast bulk of the sexual signaling is carried out by facial characters. All primates know who is conspecific — a potential sex partner — and who is not through an exquisitely subtle sensitivity to facial characters. It is an extraordinary fact that no two people who aren’t identical twins look alike; that we can recognize thousands, perhaps millions, of distinct faces; that we never forget a face even if the name of the acquaintance slips us. Across the primate order, we are tuned in to extremely subtle differences in facial characters.

I asked him after the talk what is a good signal for a character to be under sexual selection. He told me something that resolved a longstanding problem for the Policy Tensor. The signal, he told me, is sexual dimorphism. If a character is dimorphic, it’s a good bet that it is under sexual selection. We have established that scale parameters of the human skeleton (pelvic bone width, femur head diameter, skull size) reflect thermoregulatory adaptation to the paleoclimate where the population spent the Upper Pleistocene. We decomposed cranial variation into that due to neutral drift and bioclimatic adaptation by projecting it onto distance from Africa (Khoi-San) and ET (or absolute latitude). If Higham is right, and I do believe he is, we can use dimorphism as the signal for sexual selection. So we can then attempt a three way decomposition. In what follows, we’ll examine a number of craniofacial shape variables or indices that control for scale. This will allow us to test the three causal vectors and identify characters that are neutrally derived, under selection for thermoregulatory control, or under sexual selection.

There are no races. There are demes or situated breeding populations that exhibit derived characters due to a combination of relative isolation (particularly among geographic isolates) and natural selection (adaptive or sexual). I have played Lewontin’s game of apportionment whereby one shows that race dummies explain a negligible portion of the phenotypic or genotypic variation. I am sure you are bored of it. So here’s another proof.

There is systematic (ie interdemic) variation in the frequency of certain alleles, in the frequency of blood groups, in the root structure of molars, and other polymorphisms (hair color, eye color). Many have been claimed to be racial characters. Yet, none of these map onto each other. Other characters covary smoothly—skin pigmentation (Gloger’s Rule), body size (Bergmann’s Rule), relative length of appendages (Allen’s Rule)—because they reflect climatic adaptation. None of this variability can be explained by positing that mankind has allopatric subspecies or continental races.

The malaria/sickle cell anemia connection is a good example that is often cited as a Black/Bantu racial character. Yet that’s not what the frequency distribution shows (see next figure). For instance, the high frequency of the allele in Bengal, Gujarat, and southwestern India is inconsistent with the coding of this trait as a Bantu character. 


Take another so-called racial character. Racial anthropologists claimed for a century that a diagnostic character of the Australian race was their large teeth. It is true that Australian molars show larger crown diameters on average relative to other continents. But there is substantial systematic variation within Australia. Demes in southwestern Australia have massive molars; those in the central, southeast, northwest and northeast regions do not. Positing the existence of the Australian race turns out to be actively misleading in understanding morphological variation. That’s pretty much the case with every so-called racial character. 

Even committed racial anthropologists were compelled to recognize the primacy of what they called subraces. In reality, what we have are thousands of demes that show significant and interesting variation. This variation is the explanandum of postracial physical anthropology. 

Hanihara (2000) looked at variation in a number of craniofacial characters. We start off by looking at his shape indices. Figure 1 shows the infraglabellar index, which captures the relative size of the infraglabellar notch (GOL/NOL in Howell’s labels), the distance between the nose and the brow. We see that it is roughly proportional to distance from Africa, with the geographic isolate Sahul standing out. This suggests that this character may be neutral. As we shall see later, our intuition is right.

A different pattern emerges with the gnathic index (BPL/BNL) that measures prognathism or how much the jaws protrude. We see that this is a derived condition in demes in both Africa and Sahul (making it what’s called a homoplasy). The pattern suggests that it emerged at opposite ends of the earth for different reasons, or at least under the control of different genes (similar to loss of skin pigmentation in north Asians and Europeans which has a different genetic basis).

Figure 3 displays the frontal flatness index (NAS/FMB) that measures the flatness of the face. Again this appears to be a derived condition in Eastern Eurasians. Although there is massive variation in this character within Eastern Eurasia—more than everyone else put together! We shall see this character is under strong sexual selection in some Asian demes.

The Hanihara sample contains only male crania. In order to test our hypotheses, we must turn to the good old Howells Craniometric Dataset. Before we do that, I just want to show that distance from southern Africa (computed using the Haversine formula and known waypoints) gives us good control over the infraglabellar index. See next figure. Our estimated correlation coefficient is very large and significant (r=0.577, p<0.0001) suggesting that this trait is neutral, ie not under natural selection.

We begin by looking at sexual dimorphism in craniofacial characters. Table 1 displays sexual dimorphism indices for a number of craniofacial indices. The variability is astonishing. All characters are dimorphic in some demes, suggesting they may be under sexual selection. The posterior craniofacial index (ASB/ZYB), the transverse craniofacial index (ZYB/XCB) and the Simotic flatness index (SIS/WNB) are dimorphic in the vast majority of populations in the Howells sample. These three characters may well be under sexual selection very broadly across the human race. 

Table 1. Dimorphism indices. 
 DimorphismNumber of demes dimorphicPercentage of demes dimorphic
Gnathic Index0.991519%
Posterior Index0.9682492%
Transverse Index1.0392596%
Upperfacial Index0.992519%
Nasal Index0.978727%
Orbital Index0.9801038%
Frontal Flatness Index0.980519%
Orbital Flatness Index0.980312%
Maxillary Index0.98014%
Nasodacryal Index0.980831%
Simotic Index1.1921973%
Source: Howells Craniometric Dataset, author’s computations.

Instead of using dimorphism indices directly, we shall use the t-Statistic for the test of equality of means between the sexes as the population level as the signal. And instead of deluging you with a barrage of estimates, I’ll present my final estimates from the Howells cross-section. Basically, the idea is that if we project the variation in these characters onto distance from southern Africa, absolute latitude, and our measure of dimorphism (the t-Statistic for gender equality in the character at the population level) along the cross-section, we should be able to get a handle on which of the three variables controls which character. 

Table 2. Standardized coefficients (tStat)
 Distance from AfricaAbsolute latitudeDimorphismR-squared
Gnathic Index0.12-1.86-0.290.15
Posterior Index-3.862.26-0.600.47
Transverse craniofacial Index1.28-1.03-2.450.38
Upperfacial Index0.091.43-2.100.23
Nasal Index-2.74-2.70-2.270.54
Orbital Index1.96-0.74-1.420.19
Frontal Flatness Index-1.381.04-3.560.41
Orbital Flatness Index-1.29-0.09-1.790.21
Maxillary Flatness Index1.61-0.29-2.290.28
Nasodacryal Index0.842.63-2.710.38
Simotic Index2.933.05-1.820.63
Infraglabellar Index2.20-0.57-0.070.21
Source: Howells Craniometric Dataset, author’s computations. Coefficients in bold are significant at the 5 percent level.

We can see from the results reported in Table 2 that the transverse craniofacial index, the upperfacial index, the frontal flatness index, and the maxillary flatness index are evidently exclusively under sexual selection since they are correlated with our measure of dimorphism but not distance from Africa or absolute latitude.

Thermoregulatory adaptations seem implicated in the shape of the back of the head (posterior craniofacial index), the relative width of the nose (nasal index), and the nasodacryal and simotic indices. Neutral drift is implicated too. It controls the infraglabellar index alone, and the simotic and the posterior indices jointly with absolute latitude. 

Recall from Table 1 that both the upperfacial index and the frontal index are dimorphic in only 5 demes each in the 30-deme Howells Craniometric Dataset. Going back to the interpretation of the Hanihara (2000) data, the “Asian” frontal flatness business is thus revealed as a derived character that owes to the definitely cultural phenomena of sexual selection (tStat=-3.56) in some populations; precisely which ones we cannot say because the Hanihara (2000) dataset contains data on only male crania so we cannot compute dimorphism metrics. Dimorphism controls half of the dozen characters examined in the present study; distance from Africa and absolute latitude control a third each. In the Venn diagram of control, the sole character in the intersection of all three is the nasal index, the relative width of the nose.

What the below plot suggests is that craniofacial flatness is under very strong sexual selection, suggesting that this is what is going on in some Asian demes as captured by Figure 2. So we have causal vectors pointing “the wrong way” (in the reductionist paradigm), from Society to Nature. Surely, sexual selection in our species is a cultural phenomena. Sexual selection is lived by situated populations. It is reenacted through the articulation and disarticulation of stable desiderata in the eye of the beholders. The disciplinary force of cultural selection acts directly on the sexual economy by structuring the eye of the beholder at the level of the situated populations. Discourse and Reality cannot but cointegrate.


Finally, Table 3 present our apportionment of craniofacial variation in terms of our three predictors. 

Table 3. Apportionment of craniofacial variation.
 Neutral driftBioclimatic adaptationSexual selection
Gnathic Index0.1%13.5%0.3%
Posterior Craniofacial Index35.2%12.0%0.9%
Transverse Craniofacial Index5.3%3.4%19.6%
Upperfacial Index0.0%7.2%15.5%
Nasal Index17.9%17.4%12.3%
Orbital Index13.5%1.9%7.1%
Frontal Flatness Index5.1%2.9%33.7%
Orbital Flatness Index6.2%0.0%11.9%
Maxillary Index8.6%0.3%17.6%
Nasodacryal Index1.9%18.7%19.9%
Simotic Index19.8%21.6%7.6%
Infraglabellar Index17.9%1.2%0.0%
Source: Howells Craniometric Dataset, author’s computations. Estimates in bold are significant at 5 percent. 

What is clear from Table 3 is that sexual selection is a potent force shaping human craniofacial morphology. The upper face, the nose, and the flatness of the face are all under sexual selection in some demes. The important thing to remember is that sexual selection, like neutral drift and climatic adaptation, works at the level of the situated populations. What this means is that one cannot infer population history directly from phenotypic characters; one must control for sexual selection and climatic adaptation. To wit, if population A looks more similar to population B than C it may not be because C split from A and B first and then A split from B. For it may be that A and C split away from B first but A and B acquired the same characters as a result of adaptation to the macroclimate (as happened with skin pigmentation) or as a result of the same character coming under sexual selection in A and B but not C. Put another way, natural selection (climatic, sexual, or whatever) confounds the population history signal in phenotypic characters. So we must be careful. 

Postscript. It seems that my A,B,C example was not clear. I am banging on this drum because this issue has been ignored for more than a century now; first, as a result of the mental rigidities associated with essentialist racial taxonomy; and later, as a result of DNA supremacism whereby scholars drank the Kool-Aid and resolved to explain the heavens and everything under them through molecular anthropology. For a century now, physical anthropologists have reasoned back from systematic variation in morphology and genetics to phylogeny (who split when from whom — the tree of descent). This approach was correctly used to infer that Americans were closer to Asians than Europeans so that the American-Asian split happened after the European-Asian split. But backing out population history from phenotypic or genotypic distance metrics does not always work, above all because the population history signal is more often than not confounded by natural selection. So if you are using phenotypic characters or genomic sequences to make inferences about population history, you better be careful. In order to make kosher inferences you have to control for stuff that is under selection. In other words, What matters is not overall genetic distance between populations but neutral distance — that’s what contains information about population history and phylogeny. We have seen that dimorphism gives us a good handle on sexual selection in craniofacial morphology.