Thursday, July 31, 2008

Lewontin's Fallacy?

Turning to race, we must begin with the fraught question of whether it even exists, or whether it is instead a social construct. The Harvard geneticist Richard Lewontin originated the idea of race as a social construct in 1972, arguing that the genetic differences across races were so trivial that no scientist working exclusively with genetic data would sort people into blacks, whites, or Asians. In his words, "racial classification is now seen to be of virtually no genetic or taxonomic significance."

Lewontin's position, which quickly became a tenet of political correctness, carried with it a potential means of being falsified. If he was correct, then a statistical analysis of genetic markers would not produce clusters corresponding to common racial labels.

In the last few years, that test has become feasible, and now we know that Lewontin was wrong. Several analyses have confirmed the genetic reality of group identities going under the label of race or ethnicity. In the most recent, published this year, all but five of the 3,636 subjects fell into the cluster of genetic markers corresponding to their self-identified ethnic group. When a statistical procedure, blind to physical characteristics and working exclusively with genetic information, classifies 99.9 percent of the individuals in a large sample in the same way they classify themselves, it is hard to argue that race is imaginary.

The above is from an article by Charles Murray in Commentary, “The Inequality Taboo.” It apparently refers to an earlier article on ‘Lewontin’s Fallacy’ by Edwards (2003).

Murray is right in believing that human genetic variation does cluster geographically and that these clusters are adaptively significant —be they ‘races’, ‘geographical populations’ or whatever.

But is this point proven by the above line of reasoning? Lewontin never argued that human genetic variation is random. He simply affirmed that human races, however they may be defined, account for only a small percentage of total variation. Hence, there is far more variability within than between human populations. Murray counters that this conclusion is false because Lewontin looked at only one genetic trait at a time.

Clearly, if two groups overlap, they are more easily told apart with several criteria than with just one. If we use enough criteria, the overlap will shrink to zero: individuals will be assignable to either group with no ambiguity. But none of this means that within-group variability has decreased. In fact, it has actually increased. The only difference now is that this variability consists of combinations of genes that are unique to each group. How does this fact invalidate Lewontin’s contention that “the largest part by far of human variation [is] accounted for by the differences between individuals.”?

One might object that Charles Murray was talking about genes that contribute to intelligence and that such a contribution is almost certainly polygenic. Yes, but we’re not looking at several genes that display one form in one group and another form in the other. The two groups are still very heterogeneous whether you’re looking at any one gene or at gene combinations.

To find the flaw in Lewontin’s argument, we must examine his initial assumption: a random sample of genes should tell us how important race differences are. True, a large enough sample of genes will tell us whether a species has begun to differentiate into identifiable subpopulations. It will also tell us, roughly, when these subpopulations began to differentiate from each other.

But it won’t tell us how important between-population differences are in relation to within-population differences. It’s an apples and oranges comparison. The two groups of genes are qualitatively different.

First, when genes vary between populations, it’s usually because these populations inhabit different environments with different sets of selection pressures. Genes that differ across this environmental boundary are necessarily genes that make a difference, i.e., that have selective value.

In contrast, when genes vary within a population, despite similar selection pressures, it’s usually because they have little or no selective value (or because they form a balanced polymorphism, but that’s another topic!).

Second, the genetic markers used by population geneticists (blood groups, enzymes, mtDNA, etc.) tend to be selectively less important. This is partly because that is how population geneticists want them to be. Researchers will often choose markers that are close to selective neutrality because such markers change at a predictable rate (through random mutations) and can thus provide a time clock of a population’s history.

Such markers are also chosen because their protein products are easier to find and measure in body tissues. These ‘structural proteins’ are usually similar when we compare different species or even different genera. Humans and chimps, for instance, look very much alike when it comes to the protein building blocks that make up their body tissues. We have diverged from other apes largely through evolutionary changes at a higher level. i.e., regulatory genes that control development and other higher-order processes.

This point was grasped by Stephen J. Gould (1977, 406). He explained how we distort our understanding of genetic variation by relying on data from structural genes:

The most important event in evolutionary biology during the past decade has been the development of electrophoretic techniques for the routine measurement of genetic variation in natural populations. Yet this imposing edifice of new data and interpretation rests upon the shaky foundation of its concentration on structural genes alone (faute de mieux, to be sure; it is notoriously difficult to measure differences in genes that vary only in the timing and amount of their products in ontogeny, while genes that code for stable proteins are easily assessed).


Edwards, A.W.F. (2003). Human genetic diversity: Lewontin’s fallacy. BioEssays, 25, 798-801.

Gould, S.J. (1977). Ontogeny and Phylogeny. Belknap Press: Cambridge (Mass.)

Lewontin, R.C. (1972). The apportionment of human diversity. Evolutionary Biology, 6,381-398.

Murray, C. (2005). The inequality taboo. Commentary, September.


Anonymous said...

Surely the problem here is that we are related to all other humans via a branched tree, so we expect nested relationships.

Moreover, on some characteristics (genes) we expect no variance within and between subtrees (groups) because that characteristic is highly constrained (conserved), while on others, there can be random differences because there are no constraints (and the amount of variance that has accumulated will depend on branching time, but there should be less variance between people closer to us on the tree), while some characters are the sine qua non of our branch and are totally unshared with any other branch ...

However, on the MHC genes, we expect enormous amounts of between individual variance because of its function.

Thus, in my mind, Lewontin was dishonest, but then all ideologues are dishonest.

Anonymous said...

Gould had a vested interest:
Tooby on Gould
"In Goulds veiw, most evolutionary change takes place when closely related biological lineages compete, with one suviving and spreading through the others' ranges while the others go extinct...there is not much difference between a incipient species and a 'race' and in Goulds world of sudden genetic revolutions there is not necessarily any difference at all... Gould does intimate that competitive ability between sibling species is often the deciding force"

But Tooby says "the only way to prevent the destructive scrambling of our complex adaptations every generation is for all of the genes necessary for coding for each of our complex adaptation to be universal or near universal and hence reliably supplied by each parent... a profoundly important fact derivable from adaptationalist principles
... Favourable mutations are random and probabilty theory also makes it highly unlikely that favourable mutations will be disproportionatly concentrated in any one population...To restrict interbreeding is to cut off ones population from the slow influx of spreading favourable mutations being harvested across the species range (hence) Goulds theories are highly unlikely to apply to a long lived, slow-reproducing species such as humans".

If I read your post correctly, a gene that varies between populations is usually the result of rapid change due to its high selective value in one of those populations.

An awful lot of people (Lewontin, Tooby et al) seem to have grasped this and - being wide awake to the implications - muddied the waters with comparisons to no selective value gene variation.

Anonymous said...

Let's say we choose the genes that code for lactase persistence in adult humans.

They are different for different populations, and pretty much non-existent in some populations.

In this case, I would say, variance between populations is larger than variance within ...

Anonymous said...

anon 1 should remember that we are descended from two individuals, and mixing can tend to change the relationships up somewhat.

However, intermarriage is not as frequent as some would have us believe. On the other hand, though, advantageous genes will tend to spread through a population very quickly.

Yet, on the other hand, neutral genes will tend to get displaced by advantageous genes but can remain in the population, while deleterious genes will be driven to low levels quickly.

Anonymous said...


Tooby's argument is that complex genetic traits, such as those underlying behavior, cannot evolve rapidly. This is because any significant change would require simultaneous adjustments to a large number of interacting genes. By the same reasoning, a complex genetic trait cannot easily vary within a single species, since gene flow would break it up and create maladaptive variants.

The problem with this reasoning is that genes can control the output of other genes. A single 'regulatory gene' can control an entire gene complex. We see this with many mammals, such as voles, where a single genetic 'switch' can determine whether the animal will be monogamous or polygamous.

As Harpending and Cochran (2002) point out:

"Even if 40 or 50 thousand years were too short a time for the evolutionary development of a truly new and highly complex mental adaptation, which is by no means certain, it is certainly long enough for some groups to lose such an adaptation, for some groups to develop a highly exaggerated version of an adaptation, or for changes in the triggers or timing of that adaptation to evolve. That is what we see in domesticated dogs, for example, who have entirely lost certain key behavioral adaptations of wolves such as paternal investment. Other wolf behaviors have been exaggerated or distorted. A border collie's herding is recognizably derived from wolf behaviors, as is a terrier's aggressiveness, but this hardly means that collies, wolves, and terriers are all the same. Paternal investment may be particularly fragile and easily lost in mammals, because parental investment via internal gestation and lactation is engineered into females but not males."

Harpending, H. and G. Cochran. 2002. "In our genes", Proceedings of the National Academy of Sciences 99(1):10-12.

Anonymous said...

Since language is a social construct, it stands to reason that race is also a social construct. But that doesn't mean that it doesn't exist.
An analogy would be to say that there are no different breeds of dogs because breed is a social construct. Dogs don't seem to recognize their differences, but people do!
Recently I have read several essays by black writers who reinforce the fact that the concept of race has multiple meanings for them that it may not carry for white people. The meanings may be social constructs, but they are no less real.
I wonder how much Lewontin's reasoning or that of others who agree with him is influenced by the realization that white folks are very much in the minority in the global society.

Meng Hu said...

"Second, the genetic markers used by population geneticists (blood groups, enzymes, mtDNA, etc.) tend to be selectively less important. This is partly because that is how population geneticists want them to be. Researchers will often choose markers that are close to selective neutrality because such markers change at a predictable rate (through random mutations) and can thus provide a time clock of a population’s history."

I believe this part is important. Do you know if this topic is usually discussed among geneticists ? Thanks.

Meng Hu said...

I should have said :

"Do you know if this topic is usually discussed among geneticists when it comes to genetic variation within vs between population ?"

I ask because I don't remember someone else pointed this out to Lewontin and co. I am somewhat surprised.

HEINER said...

This is what happens when you allow jews in academia, they don't search for truth but for what is good for the jews.