So is mixing between members of the extreme arms of the clusters expected to be any more problematic than between closer distances?

What is "race" in the human species comparable to, in terms of the way we talk about other species?

A subspecies? A "strain"? A population?

The intensity of this argument comes mostly from disagreements about the level of significance of race, and even more, from concerns about the practical political implications of race.

For example, if the practical debate were only about tissue-transplants or disease-diagnosis, it would have been settled by now, if only by some sort of compromise or kludge. In themselves, these aren't hot-button talk-show issues.

So is mixing between members of the extreme arms of the clusters expected to be any more problematic than between closer distances?

why would either be "problematic"?

What is "race" in the human species comparable to, in terms of the way we talk about other species?

different people will tell you different things. to me, though, it's kind of irrelevant. you could probably put a number on the divergence between human populations and compare it to the divergence between mouse strains, but why? If someone says the human divergence under model A is equivalent to subspecies and someone else says it's equivalent under model B to two populations, what does that change? The differences don't go away.

The question of level of significance doesn't go away either, though.

I know that in biology there are often disputes about the last branch of speciation, and even more so at the levels below species. It seems to be a very nominalistic area of science.

As I've said, for political reasons some of the people who are most willing to speciate endangered species are least willing to recognize race among humans, and conversely. If the scientific procedures are uncertain in this area (they sometimes seem to be) these problems are pretty intractable.

Going back to the level of significance question, it would presumably be possible to set your thresholds such that the Scandinavians and the British are different races, and certainly the Italians and the Norwegians.

The most important reason for this, in my mind, is that African-Americans are not a random sample from African populations, nor are Americans of European ancestry a random sample from all European populations, and the non-random sampling accentuates the genetic differences between the two groups.

This might be easy to confuse with a related idea, so let me offer a rephrasing.

The ancestors of European-Americans and African-Americans are non-randomly sampled from the globe, with a bias towards points on the globe that are quite distant.

Even if counter-factually the results from section 1 did not obtain (for the globe), clustering would have been found in the U.S. population because of its non-random immigration and subsequent assortative mating.

I've posted on the right a Venn diagram from the paper showing that most of the loci identified as under selection are detected in only one of the three groups, indicating that selection is causing people in different parts of the globe to become more distinct.

the difference could be caused by a high false positive rate. i believe nancy cox suggested that the current tests do have a high false positive rate. however, i would bet that even with more sensitive test, differences of the type (if not the magnitude) presented would be found.

the exact mix of genetic and environmental causes of group differences in any phenotype should be an empirical question. in the end, traits will probably be found that represent every conceivable mix of genetic and environmental causes.

to refine a point, a trait can have a "biological" basis without necessarily having a genetic basis. (not sure if the converse is true.)

John: Going back to the level of significance question, it would presumably be possible to set your thresholds such that the Scandinavians and the British are different races, and certainly the Italians and the Norwegians.

I think that's probably right, but there's an important caveat. If you split Italians and the Norwegians, the Scandinavians and the British are (AFAIK) going to be in the "race" with the Norwegians. Moreover, you'll never get African populations in either of those "races".

The number of loci sampled will place a floor on how many groups we can discern, but genotyping is quickly becoming unlimited. Rhetorically: how long before LLCS can get full genome sequences on his worldwide samples?

Going back to the level of significance question, it would presumably be possible to set your thresholds such that the Scandinavians and the British are different races, and certainly the Italians and the Norwegians.

sure, it's a hill vs. mountain question. in the studies I cited in part I, they looked globably and found 5 clusters that correspond to the continents (and explain ~5% of neutral genetic variation. hill or mountain?). If you want to go finer scale, of course you can--you could probably distinguish different villages in Italy by genetic means if you so pleased.

I'm not sure what you mean by "level of significance"? Statistical significance (not an issue)? Phenotypic significance (see part III, it's in progress)? Or what?

The ancestors of European-Americans and African-Americans are non-randomly sampled from the globe, with a bias towards points on the globe that are quite distant.

yeah, that's what I meant, I'll change that. thanks.

the difference could be caused by a high false positive rate. i believe nancy cox suggested that the current tests do have a high false positive rate.

molly przeworski, not nancy. and she claimed they have a high false negative rate. that could also presumably give a similar looking venn diagram, though, but she didn't specifically consider the statistic (iHS) used in that paper, so it's tough to tell. it would be pretty remarkable, though, if all the loci identified as under selection in only one population were actually under selection in all of them.

p-ter, I think one of the problems people have with this is that it somehow reinforces the old repugnant idea that there are 5 basic distinguishable human races with distinct boundaries, when that is obviously not the case, and I came to the conclusion that is what the 'social construct' comment applied to. The reality is that you can distinguish clustering at a continental level, or whatever level you like, and end up defining as a distinguishable race a population cluster (John Hawks predicted this would become the acceptable term for race in 2007 and here we are.....) at a level that most people intuitively would not recognize as a 'distinct' race phenotypically.

In other words, what is the significance of the 5? Actually, none, particularly. It is seeing some significance in it that is the problem.

end up defining as a distinguishable race a population cluster (John Hawks predicted this would become the acceptable term for race in 2007 and here we are.....) at a level that most people intuitively would not recognize as a 'distinct' race phenotypically.

In other words, what is the significance of the 5? Actually, none, particularly. It is seeing some significance in it that is the problem.

what about section II?

and again with the word significance. what do people mean when they say that this? Seriously, I think I've been doing statistics too much and I've lost all understanding of what most people mean when they say it.

what is the significance of the 5?

iirc, even when these groups added many more loci, they had trouble getting more than 6 cluster reliably (see the 2nd paper).

certain geographical regions seem to have had a strong impact on human genetic variation (thru migration?) and these stand out strongly in the data.

only more data will tell us how hard it is to distinguish further groups reliably.

certain geographical regions seem to have had a strong impact on human genetic variation (thru migration?) and these stand out strongly in the data

yeah, they suggest migration--oceans, large mountain ranges, etc. are pretty serious barriers.

Significance: Do we say "There are five races, and this explains everythinh" or "We have five races, BFD". Those are the extremes. Truth aside, how much power does racial differentiation have.

As I understand, there are at least two statistically insignificant groups which are biologically very distinct -- African aborigines and the peoples once called Bushmen in Africa.

why would either be "problematic"?

this is more likely to happen i assume. if different locales developed different genetic strategies then there might be problems when alleles are introduced into a very different background. i suspect that the magnitude of this is minimal though, as i don't think human populations have been isolated long enough for contingent coadapted complexes to be very common. e.g., when euros cross with east asians they don't turn black because their alternative skin color strategies conflict.

p-ter, sorry, I'm just using the word significant to mean to have some (practical) meaning. Yes, oceans and mountains, but where there are no oceans or mountains there are no distinct boundaries.

rikurzhen, if you go and sample people on the Andaman Islands you'll find another cluster, a pretty distinct one. The number of clusters is a function of, inter alia, where the sampling is done. Were Tasmanian aboriginals a 'race'? Distinct from Australian aboriginals on the mainland?

Is Razib part of the Asian 'race'? What does 'Asian' describe? Phenotypically I'm betting a Japanese or Korean would not identify him phenotypically as being of the same 'race' as themselves, let alone someone from Turkey or Palestine.

It's a meaningless term.

if we stay instrumentalist these problems go away. we aren't talking about platonic truths, let alone "races" being fundamental like quarks, or even species separated by interfertility barriers. rather, we're talking clusters of correlated traits. sequence of traits 1-5 can give you a sense of the rest of the genome. e.g., dark skin, kinky hair and a broad nose is a pretty good predictor for being - on the duffy antigen.

yeah, they suggest migration--oceans, large mountain ranges, etc. are pretty serious barriers.

assume you take an allele and track its frequency change over 5,000 miles. if you assume that it is at 0% at one antipode and 100% at the other and you have a 1% increment from location 0 to location 5,000 miles, you have a perfect clinal trait, right? but what about if the mean population density along this straight line is 100, and for 500 miles the density is 10 between miles 2250 and 2750. in other words, if you don't take into account population density you see a smooth gradient, but if you weight for population you'd note a definite discontinuity in the demes along frequency 45 to 55. 90% is an arbitary break between a grade B and a grade A, but there is a definitely difference between a B- student and an A+ one.

razib, fair enough, but what do we use it for? Medical application? No, that cannot be done reliably at this level. So what is it instrumental for?

Sorry to use you as an example, but you were the nearest good looking brown guy around.

The number of clusters is a function of, inter alia, where the sampling is done.

Density of sampling was a subject of dispute that was examined in the 2nd paper. IIRC, they report no effect. I would guess that the Andamanese would cluster in one ("Oceania") of the 5/6 groups if added to the existing samples, rather than forming a new cluster.

rikurzhen, no they don't.

razib, fair enough, but what do we use it for? Medical application? No, that cannot be done reliably at this level. So what is it instrumental for?

the probability of a match with someone of the same geographic race is far, far, higher, for tissue transplants (the likelihood of match is directly proportionate to population distance since the MHC loci are hyperpolymorphic). additionally, there is the strong likelihood that the skew toward whites in medical trials is probably deleterious to the health of non-whites who exhibit different genetic backgrounds (i.e., drug x which is effective for whites might not be effective for non-whites and vice versa).

Significance: Do we say "There are five races, and this explains everythinh" or "We have five races, BFD". Those are the extremes. Truth aside, how much power does racial differentiation have.

um. somewhere between those two.

island groups throw monkey wrenches into these clusters from what i recall. i mean, the sardinians always end up as outliers because being on an island means also sorts of weird genetic drift has accumulated without deme-deme migration equilibrating the random walks in allele space.

razib, but surely that just illustrates that the boundaries between noticeable differences in clusters are set arbitrarily. Given a big enough sample, there is a gradation within B that is continuous with the gradation in A, but they are divided into Grades at an arbitrary boundary for a practical purpose.

if you go and sample people on the Andaman Islands you'll find another cluster, a pretty distinct one.

like I wrote, there are always exceptions in biology.

razib, yes that's why I quoted the Andamanese. They were the biggest monkey wrench I could find.

...i think the AA have been sampled on the Y & mtDNA a bit....

razib, but surely that just illustrates that the boundaries between noticeable differences in clusters are set arbitrarily. Given a big enough sample, there is a gradation within B that is continuous with the gradation in A, but they are divided into Grades at an arbitrary boundary for a practical purpose.

sure, but that's like saying that a p-value of 0.05 is arbitrary. it is, but it is different from a p-value of 0.25, and that's the information we're really trying to communicate, not some true mystical "Significance." human populations are simply vehicles for segregating alleles. when we classify by population we're losing some information and simplifying the genetic reality. but as humans we have a difficult time keeping in our minds a gestalt conception of the alleles frequencies across all loci as a function of geography. marking off lines which separate distinct population clusters is easier, and is good enough for government work.

Given a big enough sample, there is a gradation within B that is continuous with the gradation in A, but they are divided into Grades at an arbitrary boundary for a practical purpose.

this can/will happen for single traits/loci, but the cluster analysis is looking at the correlation structure across traits/loci. these reveal discontinuities of genetic distance in the case of the microsatellite studies.

razib, yes, but if the drug prescription is done indiscriminately on the basis of self-identified race alone it can also be deleterious.

I accept the donor matching thing, but in real cases I know of from my part of the world, it has to be on a much more specific basis than just '1 of 5'. That can be nowhere near close enough.

P-ter, I agree, but where on the scale between "BFD" and "explains everything"?

The Andaman Islanders, Bushmen, and aborigines just mentioned point to a different way of dealing with race. Perhaps all three groups are more different than each other, and the average human from the five major groups, than the average members of the five major groups are from each other. Or perhaps not.

This kind pof thing comes up with language groups and also with animal species (e.g. the very rare egg-laying animals and the relatively rare marsupials.)

"Significance" isn't mystical. It means "what is the explanatory power of race?"

but if the drug prescription is done indiscriminately on the basis of self-identified race alone it can also be deleterious.


because of the nature of american "races" (i.e., they are from very distinct parts of the world, europe and africa and asia, for example) this is not as much of a problem as you'd think. latinos present a major issue, but they're only 15% of the population right now.

where on the scale between "BFD" and "explains everything"

this will vary from trait to trait, and depends at least in part of (Fst?) within/between group distribution of the phenotype

for example, the explanatory power of race for skin color is huge.

as an indication for bidil, it's middling, but better than not using it.

"Significance" isn't mystical. It means "what is the explanatory power of race?"

i was talking p-values and statistical significance. i think you are referring to a different use of the word above?

You see, the explanatory power for skin color was already known to be huge, and if that was all we were talking about, there would be no big debate about race.

for example, the explanatory power of race for skin color is huge.

rik, do you think this is really a good example? people will often point out that east asians and europeans are both very light, that south asians, africans and papuans are very dark, etc. perahps a better way is just to keep adding traits and show how different clusters exist. e.g., south asians = dark skin + straight hair, africans = dark skin + kinky hair, etc.

Yeah, that's always the problem with the word significance. I meant the non-statistical meaning, "importance".

rik, do you think this is really a good example?

you're right, i had the figure for euro-african skin color Fst in my head when i suggested it was huge. i don't know what the ANOVA works out to for worldwide groups.

you're right, i had the figure for euro-african skin color Fst in my head when i suggested it was huge. i don't know what the ANOVA works out to for worldwide groups.

dark-skinned populations generally exhibit the same 'consensus' sequence for their MC1R loci. i suspect there's only one way to get dark (gain function). this is independent of evolutionary history on the other loci as well. in contrast with light-skinned pops, who can 'lose function' in a host of ways....

rikurzhen, yes probably. I'm a fraud/imposter - I don't actually understand this subject at all, I'm just an interested but very ignorant layman and come to read this BLOG to try to learn from razib et al.

I also have personal reasons for not liking the concept of race, I'm a Euro that jumped the species barrier and mated with an East Asian. And no, our daughter didn't come out black, she has freckles.

I started reading about genetics to try to help my daughter reinforce her sense of self-identity in a world where people want to put racial labels on others and use words like 'mongrel' etc, and I find this BLOG very helpful in trying to get a basic understanding of it, for which I am grateful to razib and the others for their efforts. It is an endlessly fascinating subject, and we got past the 'mongrel' thing a long time ago.

I also have personal reasons for not liking the concept of race, I'm a Euro that jumped the species barrier and mated with an East Asian. And no, our daughter didn't come out black, she has freckles.

my own children will be biracial, so be aware that you are amongst fellow travellers.

I also have personal reasons for not liking the concept of race, I'm a Euro that jumped the species barrier and mated with an East Asian.

that's nothing like "jumping the species barrier"! no one is arguing anything like that here.

mixed race kids obviously add an order of magnitude of complexity to the discussion, especially 1st generation where they are going to have two complete haploid genomes from different clusters.

re: race. the key here is that the public, as in jo-schmo, won't understand the genetic details about correlation structure. they aren't interested in the nitty-gritty of evolutionary genetics. so, the question is, do you just tell them that race doesn't exist? i don't think that's a tenable viewpoint in the near future because the media will start reporting on a lot of evolutionarily significant genes which are very, very, very different (i.e., one is 0% for allele x and another is 100% for allele x). jo-schmo already has a platonist/essentialist bias, so i don't think you'll be able to say race doesn't exist. it does, "for government work," on the genetic level. so we need to figure out the most intelligible way to make the concepts clear without debasing the science.

so we need to figure out the most intelligible way to make the concepts clear without debasing the science.

true. i think i've previously said that i'm only bothered by intentional obfuscation. being slow and careful when it comes to public presentation (but certainly not in the scholarly literature, see Gottfredson in press), isn't a terribly bad idea.

is the raw data for these genetic distances available? The grouping into 5 may not be arbitrary in the sense of a p value but more defensible.. eg.. the partition that maximizes a fit statistic.

I am not a geneticist but sure would like to run the 3000x3000 matrix on this algorithm.

http://genetics.plosjournals.org...en-0010070- b001

The data used in this study are a subset of the genotypes available at http://research.marshfieldclinic...ic.org/ genetics, and the exact data employed in our analysis are available at http:// rosenberglab.bioinformati...s.med.umich.edu.

Most of you guys will be familiar with Steve Sailer's formulation of 'race'. I can only think that blogger's jealousy explains your refusal to accept his eminently sensible attempt at defining this term. Nothing in this thread is near as convincing.

Most of you guys will be familiar with Steve Sailer's formulation of 'race'. I can only think that blogger's jealousy explains your refusal to accept his eminently sensible attempt at defining this term. Nothing in this thread is near as convincing.

seeing as how i read steve's blog everyday, and have been following his work since 2000, yeah, i know about it. but 'race as an extended family' explains one aspect, that of ancestry. when it comes to selection it isn't as relevant since selection works on a subset of loci...and well, we're a bit more interested in selection here on this thread. selection tracks phylogeny to some extent, but not in many of the "monkey wrench" cases (e.g., dark-skinned peoples are not that closely related, but their color is the same because of selective constraint). steve's formulation has the most utility for the american case.

razib, thank you, that's appreciated brother.

The 'public understanding' problem is a very difficult one, because to a layman this is a very difficult subject to grasp, even for one like me who is interested in it (now) for its own sake. I honestly don't know how you solve that one. It's not a unique problem though, it applies to other fields where you have the difficulty of avoiding gross misunderstanding without debasing the science.

I think giving analagous examples that people are more familiar with can help a lot, if you can find some relevant ones. I find that helpful for people to understand risk concepts - quote sources of risk they feel familiar with.

rikurzhen, thank you, my daughter is going to get a real kick out of finding out she has two complete haploid genomes :)

The grouping into 5 may not be arbitrary in the sense of a p value but more defensible.. eg.. the partition that maximizes a fit statistic.

the program used is STRUCTURE; you set the number of clusters and it runs Markov chain monte carlo to estimate the posterior probabilites of membership in each cluster. They used a bunch of different numbers of clusters and found that 5 maximized a "clusteredness" statistic.

p-ter, I was joking about the species barrier.

http://content.nejm.org/cgi/cont.../full/354/4/ 408



Denying the existence of racial or ethnic differences in gene frequencies, some of which may contribute to disease or treatment response, is unlikely to benefit minority populations.8 A good example is provided by the recommendation that patients with colon cancer who are receiving irinotecan undergo genetic testing for homozygosity for deficiency alleles of the enzyme uridine diphosphate glucuronosyltransferase isoform 1A1.18 Homozygotes have severe side effects and a reduction in the starting dose by at least one level of irinotecan is recommended.19 Racial and ethnic variations in the frequency of deficiency alleles should be taken into account when evaluating patients. The *28 deficiency allele is homozygous in approximately 20 percent of African Americans, 15 percent of whites, 1 percent of East Asians, and [less than]0.1 percent of Pacific Islanders.20 In contrast, homozygotes for the *6 deficiency allele occur in approximately 2.5 percent of East Asians but are not found in African Americans or whites.20

On the other hand, for lung cancer, from the perspective of public health it is less clear that genetic analysis offers a useful pathway to a largely socioeconomic problem.21 As Table 1 indicates, eliminating smoking would largely reduce and equalize the rates of lung cancer across the reported racial and ethnic groups.

for which I am grateful to razib and the others for their efforts

oh, and straight up, i haven't blogged much science at this domain since doing duty at SB. so much respect to the rest of the bloggers here who have taken up the torch. no need to include me anymore, let the past rest.

update

I last worked on classification trees in the early 90s and wrote a couple of papers then on a technique called clique optimization.. I havent done anything in that area since or kept up. I did write the code in ansi c (last mod in 1993) and was able to compile to windows (so it should compile on *ix)

I randomly picked 500 cases from one of the data sets and computed a city block similarity metric based on the four vars named Information something or the other in the dataset with 8400 odd cases. I suspected this was the filtered info.

I wasnt sure what would happen as I hadnt the faintest idea if what I did made any sense (from genetics pov).. Surprise.. the data is a beautiful parsimonious ultrametric hierarchy (i.e. a rooted tree) with 4 or 5 levels and accounts for most of the variance in the similarity metric.

The first level had about 7 groups.. maybe a few of these are singletons.. so the number of real groups is probably less and this level alone accounts for more than 50% of the variance..

and the overall correlation with just 4 levels in the tree is an astounding 0.93..

I can email the zip (it is small) someplace for those who want to interpret the data.. If you use the DOS edit.exe you can see the ASCII high chars used to draw the tree. I am busy with other stuff right now and will take a closer look later.

If you want to see why researchers always come up with three to seven "continental-scale racial groups," I showed two world maps, one with prehistoric migration routes on it, the other with all the possible continent to continent migrations routes _not_ taken -- e.g., until 1492, nobody ever went directly from Africa to South America, even though they are only 1600 miles apart. You can find them here:

http://www.vdare.com/sailer/ 0503...50320_leroi.htm

I do think that for the question of how to explain what a racial group is to the public, my definition that it's a "partly inbred extended family" has a lot more potential than discussions of N-space.

I suppose there could be situations in which your top down defintion and my bottom up definition would clash, but they sound pretty contrived. I'd be interested to hear how selection would render my definition irrelevant.

Steve Sailor: your definition is easier for the average person to understand, but it doesn't imply the existence of subpopulations with significant gene frequency variations, which is most relevant to group differences in phenotype. Similarly, as Razib noted, it doesn't say anything about whether the distant branches of the family have been subjected to different selection pressures. The clustering observation implies this (that's likely how the clustering came about on so many loci - although I suppose it might also be due to founders or other small population effects).

That is, a "partly inbred extended family" might or might not exhibit the clustering found. It doesn't resolve the present situation, for example, from one in which clustering is only on a few genes (e.g., for pigmentation and hair color), and in which it would be fair to say there is little content to the term "race". In the latter case "race" only tells you obvious things you already knew from just looking at the person. In the clustering case, it (statistically) implies much more about genes whose effect can't be seen superficially.

There is much more information in the fact that clustering occurs on a large fraction of all genes. For someone who is a bit more sophisticated about math or genetics, it immediately tells them there might be significant group differences between the sub-populations which go beyond a few superficial features. For example, a population that satisfies your definition might or might not have differences in disease resistance, but it's highly likely from the cluster definition.

Admittedly, none of this is easy for joe shmo to understand. You need some basic idea about conditional probability or statistics. "If Jane has the gene for straight black hair, she is more likely to have the disease-related gene X..." note this kind of statement is impied by clustering but not necessarily by your definition - it might be that distant branches of the family tree *only* differ in hair and skin color.

The clustering observation implies this (that's likely how the clustering came about on so many loci - although I suppose it might also be due to founders or other small population effects).

the loci used in the clustering studies were ostensibly neutral--i.e. they have not been subject to selection.

But (correct me if I'm wrong), isn't the more interesting question whether we'll eventually find clustering on other loci as well? Is there a plausible scenario where the observed clustering isn't accompanied by additional clustering on genes subject to selection?

John, I don't see why "seeing a difference" between races is a problem when human races are different in a whole host of ways, both socially and biologically.

But perhaps your concern is closer to home: how do you deal with your daughter.

On that score, I wish you good luck. I am white and my daughter is black. As long she identifies as black, (America's currently preferred race) she can probably go to an Ivy League school on a close-to-full-boat scholarship.

Unfortunately, if she identifies as white, her grades (and my cash supplies) are not enough to get her in.

It is a tough choice: black race consciousness and Harvard or race-doesn't-matter egalitarianism and the Big Ten.

But (correct me if I'm wrong), isn't the more interesting question whether we'll eventually find clustering on other loci as well?

populations also cluster if you look at copy-number variation, which is more likely to be functional. the interesting question to me is determining which loci are functional and finding what they do.

John, if your trouble is with the word "mongrel", you might care to know that in my British Primary School in the 50s, we were taught that we British were "mongrels". The list wasn't very long - Ancient Brits, Romans, Gaels, Anglo-Saxons, Vikings, Normans, plus, as a friend said "Huge Knots and whatnots", but mongrelism it was.

It's neither of those. I just used the word mongrel as a real-life example that I heard applied to someone else, in Australia as it happens, and when the word is used to describe 'biracial' people, it is undoubtedly offensive and intended to be, because it derives from notions of racial purity. And I did say we had gone past that - the person who used the word was instantly hounded out of political office. But I don't want to disrupt the thread by getting off the topic onto politics.

The topic is the biological basis for the concept of race.

I don't have a problem with seeing differences, I love diversity, my difficulty is in seeing the relevance of the grouping, or the purpose of the grouping; how those differences are lumped and what they are used for.

The issues that apply in the USA don't apply where I am. Hadley, I am sympathetic to problem you mention. Here the issues are maybe more subtle, but they exist.

My trouble is with classifying people by race rather than as individuals, particularly when it seems such a crude classification. As steve has said, there are group differences between sub-populations.

I am not trying to deny that this has some medical application; it obviously does. But in rik's example,once I know about how the deficiency alelle varies between groups, in treating an individual patient I don't see how it helps me to know whether he is African American or white - he still needs individual genetic testing to tell me whether he is one of the 15 or 20%.

Then consider naso-pharyngeal cancer, which is a rare condition in the USA, but is quite prevalent in southern China. In this case it doesn't help at all to know someone is Asian, or East Asian, or Chinese, or even southern Chinese - you need to know that he is Cantonese. Not Chiu Chow or Hakka or one of the other subgroups that inhabit Guandong Province, but Cantonese. Classifying him as Asian won't help you at all.

Then consider naso-pharyngeal cancer, which is a rare condition in the USA, but is quite prevalent in southern China. In this case it doesn't help at all to know someone is Asian, or East Asian, or Chinese, or even southern Chinese - you need to know that he is Cantonese. Not Chiu Chow or Hakka or one of the other subgroups that inhabit Guandong Province, but Cantonese. Classifying him as Asian won't help you at all.


I think you are focussing on the wrong thing here.

These are linguistic subgroups, not genetic subgroups. I suspect that the issue is more likely an environmental association with a genetic susceptibility that all southern Han have. I note also that there is said to be a high prevelance among the Hmong.

It seem kinda like claiming that among Aboriginal Australians Walpiri speakers are more likely to get cancer than the Arunta. Then if you knew that the Walpiri lived in areas where nuclear tests were conducted in the '50s ... (I am making the associations up! I don't actually know which groups live where in Australia, but the idea is sound.)

But in rik's example,once I know about how the deficiency alelle varies between groups, in treating an individual patient I don't see how it helps me to know whether he is African American or white - he still needs individual genetic testing to tell me whether he is one of the 15 or 20%.

this is actually a claim commonly made by geneticists. at first i also thought it sounded reasonable, but some other work as led me to think it won't quite work that way.

say there's an association between race and a disease phenotype, and that that association is ascribed to variation in frequencies of a particular allele. in most cases, it will turn out that the allele doesn't fully account for the racial difference (or even much of it at all). if the group differences have any environmental component, then genotyping will never be able to account for all of the predictive power of simple racial labels. the maximal information is probably achieved by having race *and* the genotype. thus, it would be very hard to ever replace racial labels.

mix-race people present a complication i haven't thought through, but using ancestry estimation might be useful for those with cryptic ancestry.

say there's an association between race and a disease phenotype, and that that association is ascribed to variation in frequencies of a particular allele. in most cases, it will turn out that the allele doesn't fully account for the racial difference (or even much of it at all). if the group differences have any environmental component, then genotyping will never be able to account for all of the predictive power of simple racial labels. the maximal information is probably achieved by having race *and* the genotype. thus, it would be very hard to ever replace racial labels.

Really? It seems to me that it's the combination of alleles (and the environment as well) and full genotype information should tell us whether or not the individual has a combination of alleles that make them more or less susceptible (within that environmen, ie, for the particular regimen the doctor was thinking of).

Richard, no, Cantonese are not just a linguistic subgroup among people now living in southern China. Hakka people derive geographically originally from northern China. The Hmong originated in the Yellow River Valley 5,000 years ago and migrated south over a long period of time - who knows what happened along the way. But I don't want to get into a debate about that, it isn't the point.

If you look at the incidence of naso-pharyngeal cancer in the USA you might well conclude that diet is a risk factor, but in a geographic population where diet and environmental factors are perhaps a lot more homogeneous, medical specialists practising in southern China who have been treating the condition for many years are adamant that it is genetic to a sub-group level - maybe in the context of southern China that appears to be the most significant risk factor. One very experienced specialist told me that every single case of naso-phryngeal cancer he has ever seen was in a person of Cantonese descent. But then he has probably not practised in Indochina.

My question is really at what level the racial labelling becomes useful. Maybe in the USA "Asian" is helpful in this case because they are already a minority group. In China it's not particularly, but there is at least a strong belief in the medical profession here that "Cantonese" as a label is - they're not a minority in Guangdong Province, but maybe it helps narrow it down somewhat.

rik, yes, I see your point. Maybe mixed race people need to report two labels. For first generation biracial people, self-identification as having two complete haploid genomes from different clusters is not too difficult, if they are that confident of the two clusters.

"environment" is going to be harder than genotype to quantify and thus understand. moreover, it's unlikely that group differences on most traits will come down to genetic differences at a single locus. it's really just a practical matter. getting 100% of the between-group heritability accounted for may simply not be feasible in the short or medium term. figuring out what parts of the environment are salient is similarly going to be hard.

So ... I'll show my ignorance by asking a question here.

What does Lewinton, for example, think of this? Does he provide a refutation?

Cheers,
--Bob

steve s., re: 'extended family' & race, steve h. addressed some of it, but i will clarify. a gestalt perception of race arises from a finite number of traits. e.g., skin color, hair form, facial features, etc. the variation on these is controlled by a finite number of genes. insofar as total ancestry is a good correlate for the nature of these finite genes, then race as an extended family makes eminent sense. in the american context it makes total sense since the two traditional races, blacks and whites, are generally disjoint on phenotype (very few blacks are assumed to be white, and those that are often majority white ancestrally, like walter white of the NAACP, who was 7/8 white in terms of ancestry). on the other hand, in places like latin america this makes less sense since people within the same family can be different phenotypic races. similarly, re: selection, since selection can operate on a finite slice of the genome, that specific set can code for the phenotypes which we assume to be "race." for example, in total genome the ainu of japan tended to cluster with the "extended family" eastern asians, but in terms of phenotype their (pre-admixture status) assignment might be more ambiguous. why? it seems likely that over the last 10,000 years east asians qua east asians have evolved in situ into a distinctive morph. in other words, ancestry can tell us some things, but when you move the geographic scale to the world, and the time scale to thousands of years, genealogy might be less relevant than selection.

i don't know if this is relevant for the typical person, but it definitely gets confusing in the scientific media since they use will talk of amerindians being "proto-asian," and most people have a sense of what "asian" means, but that sense might have evolved relatively recently. in this case ancestral information which is correct is imparting a false conception.

lewontin on race, written in 2005:
http://raceandgenomics.ssrc.org/...c.org/Lewontin/

about 6% to 10% of the total human variation is between the classically defined geographical races that we think of in an everyday sense as identified by skin color, hair form, and nose shape. This imprecision in assigning the proportion of variation assigned to differences among population within ”races” as compared to variation among “races,” arises precisely because there is no objective way to assign the various human populations to clear-cut races. Into which “race” do the Hindi and Urdu speakers of the Indian sub-continent fall? Should they be grouped with Europeans or with Asians or should a separate race be assigned to them? Are the Lapps of Finland and the Hazari of Afghanistan really Europeans or Asians? What about Indonesians and Melanesians?

...

A clustering of populations that does correspond to classical continental "races" can be acheived by using a special class of non-functional DNA, microsatellites. By selecting among microsatellites, it is possible to find a set that will cluster together African populations, European populations, and Asian populations, etc. These selected microsatellite DNA markers are not typical of genes, however, but have been chosen precisely because they are "maximally informative" about group differences.

...

Two of my incontrovertibly WASP grandchildren have a single Ashenazi Jewish great-grandparent and so have a one in eight chance of inheriting a Tay-Sachs abnormality carried by that ancestor. For purposes of medical testing we do not want to know whether a person is “Hispanic” but rather whether that person’s family came from a Caribbean country such as Cuba, that had a large influx of West African slaves, or one in which there was a great deal of intermixture with native American tribes as in Chile and Mexico, or one in which there was only a negligible population of non-Europeans. Racial identification simply does not do the work needed. What we ought to ask on medical questionnaires is not racial identification, but ancestry

Lewontin is an intelligent man-- he's reading the same papers I cite here, but wants to make a distinction between "race" and "ancestry"-- the first is entirely a social construct (despite the fact that you can see it if you look at genetic clusters), the second biological. or something like that.

Thank you razib for this concise explanation of race and what race means in terms of genetic variation.

What I don't understand about Lewontin's viewpoint is that it is seems essential that the microsatellite markers he refers to be atypical. If I could simply choose some genes at random and find differences in allele distribution in a significant fraction of them between the different groups, he'd be harder pressed to trivialize it. I'm not an expert, but it seems to me the second scenario is more likely than what Lewontin suggests. At least, some of the language in the Risch papers suggests you don't have to look very hard to find robust clustering. That's also why the Chicago paper cited in the part III of the post above is interesting - it finds disjoint sets of genes under recent selection in different groups. Could an expert here comment on this?

Regarding the difficulties in telling smaller sub-populations apart, this doesn't invalidate the overall concept. Just because pluto is problematic doesn't imply that there's no value in the categories "planet" vs "asteroid" or "comet". A category can fail in some difficult or extreme case while still capturing something useful in the more typical case. It's just poor logic to claim otherwise.

Razib:

Let's take the case of Sardinians, whom you mention above as showing up as quite distinct on population genetics metrics because they've largely been inbreeding on the same island for thousands of years. So, due to founders' effects and drift, they show up as a fairly distinct little racial group within larger groups like Causasians. Yet, the selection pressures on them are little different than on, say, Corsicans, Italians, or Spaniards.

What this suggests is that selection is just a subset of the several forces that can cause statistical genetic differences among partly inbred extended families. In contrast, my definition works fine for whatever the reason for genetic differences.

In contrast, Congolese, Melanesians, and Amazonians have been under similar tropical forest selection pressures, and indeed the first two racial groups ended up looking a lot alike, but the South Americans look highly different. Under my genealogical approach, they are all obviously different racial groups -- which is what every population geneticist would call them -- but under a selection-centric approach, the results are uncertain and unstable.

May I suggest that not only is my approach easier for the layman to understand, but it also offers advantages for the cognoscenti as well?

May I suggest that not only is my approach easier for the layman to understand, but it also offers advantages for the cognoscenti as well?

steve, i'm not in a either|or mode here. as i said, i'm talking about an instrumentalist take. the method of classification which is preferred based on the utility of the context. race is a social construct since it is a simplification of the real nature of human genetic variation, it just isn't an arbitrary one. in more technical lingo i think your preference is to view race as phylogeny. this is probably the common sense way to go, and most taxonomists would likely agree with you. nevertheless, focusing on selection might be considered race as phenetics. in the united states the "one drop" rule of black ancestry was a peculiar form of race as phylogeny, insofar as no matter how white someone looked, one confirmed african ancestor could define one as black. on the other hand, south africa during the apartheid years used phenetics because it seems highly likely that method of classification similar to the american one would have rendered huge swaths of the afrikaner speaking white population non-white (the median non-white ancestral might be around 1 out of 20 in blood quanta). the contrast with the american model is that a small number of "coloreds" and even indians "became white" purely on phenetic grounds every year (there were controversies, like sandra lang, with children of white parents who looked colored because of non-white ancestry from both parents, but i believe the law was changed so that no child of white parents could be reclassified).

Lewontin: What we ought to ask on medical questionnaires is not racial identification, but ancestry

if they could differ, then you ought to ask both.

Steve S. --

Sorry to repeat, but "partly inbred extended family" just doesn't imply the same things at the gene level as clustering. There's more information in the latter. Imagine you land on an alien planet, and are told about the inhabitants that they form a "partly inbred extended family". That doesn't tell you as much as if I tell you how their genetic information is clustered, that correlations exist between, e.g., the color of their fur and the genes determining night vision, etc. (In a partly inbred extended family, there might or might not be any such correlations.)

Also, a minor point: selection pressures are affected by culture as well as the local environment. Identical physical environments but differing societal structures might lead to selection for different traits. For example, in the same rainforest you might have one group practicing agriculture while the other sticks to hunting. The resulting selection on genes might be very different.

What I don't understand about Lewontin's viewpoint is that it is seems essential that the microsatellite markers he refers to be atypical. If I could simply choose some genes at random and find differences in allele distribution in a significant fraction of them between the different groups, he'd be harder pressed to trivialize it.

True. I imagine if you only used markers in genes (and used enough of them), you'd find a similar clustering as if you used ostensibly neutral microsatelites.

p-ter: Thanks for the response and the link. Interesting. I'd gotten the impression from this blog that was was a really smart guy, and wondered what his take on all this might be.

Razib:

race is a social construct since it is a simplification of the real nature of human genetic variation, it just isn't an arbitrary one

This comment was a real advance in my understanding, I think. I don't think this is what the Social Scientists mean when they say, "race is a social construct." Indeed, I think I started getting interested in the topic because that statement didn't seem to be, really, very credible (although quite PC). It seemed like this was one of the "Ten Impossible Things You Must Believe" to be a sociologist.

Was I being unfair to the social science types? Is your statement above a statement they'd generally agree with?

Cheers
--Bob

I imagine if you only used markers in genes (and used enough of them), you'd find a similar clustering as if you used ostensibly neutral microsatelites.

to follow this up, it's clear that if you use only the 3 HapMap populations, you could choose a random sample of a few thousand (of the millions) of markers and form well-defined clusters. Once extensive genotyping is done on the individuals from the Human Diversity Panel (from 52 populations), this will be easily tested.

Was I being unfair to the social science types?

no. I was in a sociology lecture a few years back where it was explicitly claimed that if you had someone's DNA sequence, you could not determine their race. And not in a "the probability of correct assignment is not 100%" way, in a "there's no information there" way.

to be fair, I imagine the first people who studied race had something like razib's statement in their heads.

The resulting selection on genes might be very different.


For example, pastoralists like the Mongols were might not have genes for more efficient digestion of the carbs in rice that many Southern Han undoubtedly have.

Razib:

Actually, many U.S. Southern states didn't use a true one drop rule precisely because lots of very important people were a little bit black.

Anyway, too many discussions about defining race get hung up on the Jim Crow South, which is hardly the be-all and end-all of human existence.

The challenge is this:

Does Hsu's approach distinguish any racial groups that my partly inbred extended family approach cannot?

If it does, isn't it vulnerable to the scoffing attack on the whole concept of race made by Jared Diamond in his 1994 "Race Without Color" article where he claimed that you could make up patently absurd racial groups based on any arbitrary trait?

If it doesn't, why not use the approach that actually explains the underlying mechanism for the creation of racial groups?

of course racial identifications in the USA will cluster together-- the populations are taken from opposite ends of the globe

is that not what I wrote in this post?

I believe it is:

The most important reason for this, in my mind, is that the ancestors of European-Americans and African-Americans were not randomly sampled from the globe (there's a bias towards points on the globe that are quite distant), and this non-random sampling accentuates the genetic differences between the two groups

is that not what I wrote in this post?

it is what you said in the post, but since your post was about the "consensus on race", and the section was called "Clusters and race", i took it that you were using the situation in America to argue for the existence of races. afterall, it doesn't make much sense to say that races exist here but not there.

i might have assumed wrong, in which case i apologize. if indeed i did assume wrong though, then im wondering why your article doesn't explicitly make the case against reifying the continental model of race frequently espoused on this blog.

afterall, it doesn't make much sense to say that races exist here but not there.

well, it's complicated. different societies have different concepts of "race", and it's entirely possible that some of them line up with genetic clusters and some of them don't, right? in america, they do. I haven't seen much data from other countries.

My earlier comments might have come off as more aggressive than I was intending. It seemed to me like this article was defending the American racial schema, which was my main point of disagreement.

It seems to me like you're saying the American conception of race works pretty well in America... I think we agree on that. But to me the important issue is that this view of race should be discouraged, because on a global scale it fails on numerous levels...

Also, from a biological standpoint, the actual geographic seperation between these "clusters" in America doesn't exist in a notable way, so, by the phylogeographic definition, they're not biological races, are they?

Also, from a biological standpoint, the actual geographic seperation between these "clusters" in America doesn't exist in a notable way, so, by the phylogeographic definition, they're not biological races, are they?

Irrelevant. The lack of separation is a recent development. The differences originated within immense geographical separation, and as yet interracial mixing has been inadequate to eliminate them.

We 'get' that you want to abolish racial types as generally-accepted categories. You no longer need to grasp at straws and present flimsy excuses for doing so. The horse is dead. Further beating is pointless.

Irrelevant. The lack of separation is a recent development.

How is it irrelevant because it's recent? The phylogeographic definition requires that "members of a subspecies would share a unique, geographic locale." Notice how it's not in past tense.

What's irrelevant is whether racial groups have the status of subspecies or not. That doesn't impact our ability to make fairly confident predictions about relative gene and trait distribution.

Race = subspecies, therefore they aren't "racial groups" in the first place if they're not species.

*subspecies

ben,

I think you're involving yourself in a bottomless, essentialist linguistic analysis of the terms "race" and "subspecies". Whether or not anyone here wants to agree that "race=subspecies" or not has nothing to do with whether or not genetic clustering analysis has predictive value. Your claim

they aren't "racial groups" in the first place if they're not species

simply does not follow unless one accepts the initial premise, which is rather hidden. Even if we were to posit that West Africans and Northeastern Asians (Japanese) were entirely different species (not that anyone here is doing this) it would still have no effect on the predictive validity of cluster-interpreted group testing. As others have pointed out the term "race" is meaningful in a man-on-the-street basis because people using common-sense see differences between people whose ancestors are from wildly different parts of the planet.

Let me give you an example: Let's say you have group A and Group B, where the males in A reach peak testosterone at 16 years old while the males in B reach it at 22. Every man on the street will say "race A" matures more quickly than "race B". Well, actually he'll probably use "faster" as opposed to "more quickly" but that simply demonstrates the poverty of grammar in our education system, erm, moving on. Okay, so what's the relevance of different peak testosterone times? Well, if you place one hundred 16 year olds together and 50 percent each are from A and B the A's will physically dominate the B's (and probably score all the chicks).

So, the "race-denying" scientist will piously proclaim that "race is scientifically meaningless" and the parents of the B's will be saying "WTF, my kid's getting the crap kicked out of him". Whether or not "race=subspecies" the fact is that the above situation would be a result of tens of thousands of years of different evolutionary pressures, and whatever name you choose to distinguish A's from B's is completely beside the point. And the fact of the matter is that genetic clustering analysis will be able to pinpoint the aggressive A's from the less-aggressive B's.

ben,

Let me give you sort of a metaphor here: I've heard even scientifically-literate people refer to Pitbulls and Collies as "different species of dog". Now clearly "dog" is usually considered a species by many biologists BUT, in fact, other biologists consider dogs a subspecies of the canid family. Why? Because dogs readily breed with wolves and have very healthy offspring.

Whether or not you want to call collies and pits different species, subspecies, lines, breeds, etc. it makes no difference in how you analyze genes and their effect on behavior. Collies and Pitts evince clearly disparate phenotypes, and no one could possibly pretend they are identical. The only question is the extent of the differences.

asher,

1) I made a typo in one of my comments and corrected it in the next one by saying "*subspecies.") So don't think I meant to say "Race = species". Rather, "race = subspecies".

2) You point out that the categories of "race" used on the street overlap a great deal with gene clustering. This is correct.

My objection is not that they don't cluster, but rather, that they are arbitrary, and in the end based on history and appearance more than science.

Let me explain. The system of race used to classify people could just as well be more broad or more specific. We could make it more broad by thinking of all sub-Saharan Africans as a race, and all non-sub-Saharan Africans as the other race. We could make it more specific by identifying Somalians and Ethipoians based on their "white" skull shapes, and group other sub-Saharan Africans differently based on theirs, but we don't do that either.

Both of the above possible systems of "race" are not adopted. Why? Because, the system of "race" employed in the street has its roots in New World conflicts between the indigenous peoples, slaves brought from Africa, and Europeans from West Europe.

To say that it overlaps with genetic clusters to a certain degree is not to make much of a point. Thousands of other categorization schemes could fulfill this criteria. In the end, defenders of "race" are being more politically correct than it's opponents-- they're reifying an arbitrary scheme created by history and appearance rather than science.

To say that it overlaps with genetic clusters to a certain degree is not to make much of a point.

To people concerned with genetic clusters, that's the only point that's relevant! And the pop culture conception of races does in fact correlate with gene clusters fairly well in the contexts from which those conceptions were derived.

We grasp that you want to abolish these social categories; we grok your arguments. Now stop beating the dead horse!

We will not cease scientifically useful investigation merely because it is sociopolitically inconvenient.

ben g

My objection is not that they don't cluster, but rather, that they are arbitrary, and in the end based on history and appearance more than science.

Do you hold similar objections to the arbitrary, fuzzy definitions that are associated with the concepts of family and color?

Where are the exact demarcation lines that categorize people into a family grouping and what is the exact spectral boundary of the color blue? Further, would you argue that a color that is a few wavelengths off the boundary is not blue, even though most everyone would classify it as being within the fuzzy boundary of blue?

ben g,

I gave you a long and detailed example of how a man-on-the-street understanding of race might have real-world consequences. The posters above me, I believe, gave you good reasons why putative arbitrary distinctions do not undermine the validity of phenotypically grouped analysis.

I really only am interested if you address my specific example instead of fixating on an essentialist definition of "race, a term which I really don't use.

We will not cease scientifically useful investigation merely because it is sociopolitically inconvenient.

This is a straw man. I never said any type of research should cease.

Do you hold similar objections to the arbitrary, fuzzy definitions that are associated with the concepts of family and color?

You said this in response to what I said earlier: "My objection is not that they don't cluster, but rather, that they are arbitrary, and in the end based on history and appearance more than science."

Before I respond to your comment, let me note that what I'm criticizing in that above sentence are attempts at trying to reify any sort of division of the human species on a global level. So let me be clear that as far as the conventional/American model, not only do I think its arbitrary, I also think it doesn't cluster.

With that noted, let me state why I think arbitrariness is a problem for any attempt to divide humanity into categories at a global level. It's a problem because the arbitrary categories that are defended tend to be the ones that most resemble the conventional model of race. To me (and many others), this makes it look like people are trying to "slice the pie" of human variation in the way that most resembles their prejudices. The "arbitrary" category that's defended usually ends up being the one that's most in line with the history/culture of the time. So, to answer your comparison about the grouping of colors-- if a certain way of group colors completely dominated everyday discussion of color variation for purely historical/cultural reasons, then yes, I would oppose that categorization.

Asher:

I really only am interested if you address my specific example instead of fixating on an essentialist definition of "race, a term which I really don't use.

I ignored your example earlier because it argued against something I wasn't claiming. But I suppose I should've made that clear... Your example assumed, like Caledonian just did, that I'm against the research of differences between populations. This is not the case. What I'm against here are two things:

1) The claim that human "races" exist. This is not the case, by the phylogeographic definition of race/subspecies (the one most often employed).

2) Scientific discourse that defends and gives special attention to an arbitrary categorization based on historical political/social realities.

2) Scientific discourse that defends and gives special attention to an arbitrary categorization based on historical political/social realities.

A quick note on that #2... if scientists are researching the nature and history of those political/social realities themselves, then I do support science giving the conventional ways of thinking about race "special attention."

So let me be clear that as far as the conventional/American model, not only do I think its arbitrary, I also think it doesn't cluster.

With that noted, let me state why I think arbitrariness is a problem for any attempt to divide humanity into categories at a global level. It's a problem because the arbitrary categories that are defended tend to be the ones that most resemble the conventional model of race. To me (and many others), this makes it look like people are trying to "slice the pie" of human variation in the way that most resembles their prejudices.

That's a tight little tautology you've constructed there.

You initially define a racial framework to be arbitrary and then you reach your conclusion that such a framework should be rejected because it's arbitrary. Bravo.

Secondarily, your rejection is thoroughly immersed in your idiosyncratic view on the motivation of others who find utility in using race in their analysis.

I see no way of piercing your position, and I find the rigor of your reasoning to be unconvincing, so I'll take a pass on being convinced.

ben g,

Your last claim:

A quick note on that #2... if scientists are researching the nature and history of those political/social realities themselves, then I do support science giving the conventional ways of thinking about race "special attention."

is pure deconstructionism. I'm sure that you think all motives are suspect and that every new round of sociological analysis will lead to another round and another round, ad nauseum.

As for your ignoring my example, I beg to differ that it very much touches upon what you're say. Look, there are two different possible uses of the term "race", one nominal and the other essential. Average, non-scientist, human beings tend to use terms in a more essentialist manner than people from higher educational backgrounds. When a scientist uses race he's using it in a strictly nominalist sense. You may not be aware of this but data samples can be grouped together using cluster analysis that is available to all sorts of data that have nothing to do with "race". In fact, the historical usage of "race", by cluster, is actually an intuitive approximation of what we can now ascertain using mathematical cluster sampling and by minimizing each data point to a particular cluster (okay, i'm not a math guy so my terminology is fuzzy but i'm quite familiar with the concept). Not only are the locations variable but the number of clusters are variable.

So, the history of race is a history of people using intuitive and discursive reasoning and coming up with the same results we now model. Certainly, some of the specific claims made about various races/populations clusters in the past was ridiculous. However, the groupings themselves were attempts to categorize the experiences and data that people encountered. The attempts are similar but the analysis is much more thorough and rigorous than before.

Look, I'm quite familiar with deconstructionism and have debated, if you can even call it that, radical post-modernists in the past. I know the drill. If you can come up with something that is not deconstructionist/po-mo then fine, otherwise, i'm done.