Any human group will always have "more genetic variation within" than between it and any other human group. So what? It's meaningless.
Lewontin’s infamous paper on the apportioning of human genetic variation is well known, while Edwards’ excellent debunking of certain misinterpretations of Lewontin’s work is unfortunately less well known.
I’d like to analyze “Lewontin’s fallacy” from a different angle, to demonstrate why it is meaningless even if taken at face value. First, some caveats. I’m going to stress apportioning genetic variation within or between “groups” rather than specifically invoke race - race here is simply one type of possible group. I’m also not going to focus on particular numbers, but rather look at the big picture – the greater amount of genetic variation within as compared to between groups.
These caveats are for two reasons. First, to explain the concept as generally as possible. Second, because there are politically-motivated hacks who attempt to defend Lewontin’s meme by making fine distinctions between “race” and “population” and who claim that only 5-10% of the 15% is inter-racial variation, and the other 5-10% is due to variation between populations that compose races. Amazingly enough, these same clueless hypocrites critique Harpending, Sarich, and Miele for dividing the intra-group variation between groups vs. individuals, etc. - we are told this is an obvious error since the variation of each subgroup is simply part of that of the larger group and cannot be separated from it. At the same time, they try to reduce the amount of inter-racial variation by separating from it the inter-population variation (as per Lewontin), populations being a sub-group of races. You can't have it both ways. See this. What stupidity.
Back to my own analysis. As stated, let us generalize the concept. If genetic variation is divided up as per Lewontin, here I assert the following will always hold. Take any human population and divide it into any number of groups. The amount of genetic variation within any of these groups will always be greater than that between any of the groups. This is because most human variation (at the locus by locus level) is distributed in an unstructured manner between all people. It is NOT specifically a feature of ethnic or racial population groups and certainly does not mean – as some have suggested – that members of the same population group are more genetically distant than are members of different groups!
As an example, take the population of the United States and divide it between right-handed and left-handed people, each of those two groups being multiracial. Let’s apportion the genetic variation of the total US population. There will be more variation within each group than between. If you were to compare a multiracial group of right-handers vs. a very homogeneous monoracial group of left-handers, there will still be more variation within each group than between, and the same holds true if the left-handers are multiracial and the right-handers are all of the same race.
Let’s apportion the total genetic variation in some other populations. If you take a single ethnic population, say, Germans, the same principle holds. You can divide them up any way you please – right vs. left-handed, blond vs. brunette, tall vs. short, fat vs. thin, male vs. female, whatever – and there will always be more genetic variation within each group than between groups. You can then mix these groups up and divide them into new groups completely at random and the same Lewontin “finding” will still hold. Imagine the entire human race was composed of Nigerians. Let’s apportion the total genetic variation of that population between, say, Nigerians born on weekdays and Nigerians born on weekends. What will you find? You guessed it, more variation within each group than between. How about apportioning genetic variation in Russians – Russians who have ever had an ingrown toenail and those who have never had one. Surprise! More genetic variation within than between.
The point is that you can take any human population – starting from the entire human race to some fraction thereof – and then create any subdivision of that population, regardless of how arbitrary, and you will always find more genetic variation within the group than between. This is simply a byproduct of the unstructured nature of such variation. Folks with access to various population genetics data can – and should – confirm this empirically, as a fact that emerges from the inherent nature of genetic variation.
Lewontin’s “finding” is therefore not specific to race, although it of course does apply to race, since race is one (non-arbitrary!) way of dividing humanity. However, since there are political points to be made trying to delegitimize race, then Lewontin’s meme is applied to race, not to any other subdivision. For example, there is no vested interests who want to use the apportionment of genetic variation to deconstruct the meaning of the group “Russians who have had an ingrown toenail.” When it comes to race, however, the race-denying crazies do come out in force.
Thus, even above and beyond Edwards’ cogent criticism, Lewontin’s meme is meaningless because it is ubiquitous. It can tell us nothing about the validity of the race concept since it is not specific to race. It is simply a measure of the unstructured nature of most human genetic variation. Of the total amount of genetic variation that exists, the majority of it will always be found within any reasonably sized group of people, regardless of how that group is chosen. In fact, as pointed out by Dr. Harpending in the appendix of On Genetic Interests, a significant portion of the total genetic variation exists within single individuals.
On the other hand, the genetic variation that exists between ethnoracial groups is highly structured, and that fraction of the variation is specific for defined inter-group comparisons and, hence, highly relevant for human classification. Identifying the structured genetic data important for human classification among the more random bulk of human genetic variation is in a sense analogous to trying to identify an important but faint radio message among a large amount of background static. Regardless that the message is fainter, it is nevertheless more important.