Single Loci vs. Combinations: Both Notes Can Be Right

2026-07-21 · AI-generated insight

Read together, these notes stage the most famous quarrel in the biology of race—and reveal that its two sides are not actually contradicting each other on the facts.

Lewontin's arithmetic is sound: for any single randomly chosen gene, roughly 85% of variation sits within populations and only ~6% tracks racial classification. Winegard and Carl are equally correct: skulls can be sorted into groups with ~80% accuracy using just two variables. How can both hold? Because Lewontin measures one locus at a time, while classification exploits the correlations across many loci at once. Small per-trait differences, when they point in the same direction across dozens of features, compound into reliable joint distinctions. This is precisely the objection A.W.F. Edwards would later name "Lewontin's Fallacy"—the leap from low variance-per-locus to "no taxonomic significance" ignores the information carried by combinations.

What makes the pairing sharp is that Lewontin's own text quietly concedes the fragility of his conclusion. Notice how much of his passage is spent worrying that the 6.3% figure is "sensitive to our racial representations"—that splitting Hindi and Urdu speakers, or unlumping Melanesians, would raise the between-race component, and that a purely genetic clustering would produce categories that "lump certain Africans with Lapps." These are not the caveats of a settled measurement; they are admissions that the number depends heavily on which populations you sample and how you draw the boundaries. Filed, appropriately, under "Bias in Scientific Measurement."

Then comes the tell: Lewontin pivots from a contested statistic straight to a moral verdict—classification is "of no social value and is positively destructive." Winegard and Carl mirror this move in the opposite direction, invoking 80% accuracy to rebut the charge that classification is "ipso facto racist." Both are smuggling normative conclusions out of the same underlying data.

The genuine synthesis is uncomfortable for partisans: the empirical dispute was largely resolved by distinguishing per-locus variance from multivariate structure. What remains is not a factual disagreement but a values disagreement wearing a lab coat—each side reaching for whichever framing of the numbers licenses the ethics it already prefers.

Woven from these notes

The primary reason that natural philosophers began to classify humans into different races is that human populations look different from one another. Their skin colors, hair textures, facial structures, and stature all differ, often in predictable ways. Furthermore, these differences reflect their divergent geographical origins. In fact, researchers can classify human variation by continent quite accurately using only data from the human skull. (They are able to correctly classify human skulls into black and white Americans with about 80% accuracy, using only two variables.) Therefore, despite the common charge that racial classification is ipso facto racist, and that most Enlightenment typologies of human variety were motivated by prejudices and a desire to rationalize imperial domination, ordinary people and Enlightenment philosophers also took to classifying human differences for the mundane reason that such differences actually exist.

— Bo Winegard, Noah Carl Superior: The Return of Race Science—A Review 2026-07-21

The results are quite remarkable. The mean” proportion of the total species diversity that is contained within populations is 85.4%, with a maximum of 99.7% for the Xm gene, and a minimum of 63.6% for Duffy. Less than 15% of all human genetic diversity is accounted for by differences between human groups! Moreover, the difference between populations within a race accounts for an additional 8.3%, so that only 6.3% is accounted for by racial classification. This allocation of 85% of human genetic diversity to individual variation within populations is sensitive to the sample of populations considered. As we have several times pointed out, our sample is heavily weighted with “primitive” peoples with small populations, so that their Ho values count much too heavily compared with their proportion in the total human population. Scanning Table 3 we see that, more often than not, the Hpop values are lower for South Asian aborigines, Australian aborigines, Oceanians, and Amerinds than for the three large racial groups. Moreover, the total human diversity, Hspecies, is inflated because of the overweighting of these small groups, which tend to have gene frequencies that deviate from the large races. Thus the fraction of diversity within populations is doubly underestimated since the numerator of that fraction is underestimated and the denominator overestimated. When we consider the remaining diversity, not explained by within-population effects, the allocation to within-race and between-race effects is sensitive to our racial representations. On the one hand the over-representation of aborigines and Oceanians tends to give too much weight to diversity between races. On the other hand, the racial component is underestimated by certain arbitrary lumpings of divergent populations in one race. For example, if the Hindi and Urdu speaking peoples were separated out as a race, and if the Melanesian peoples of the South Asian seas were not lumped with the Oceanians, then the racial component of diversity would be increased. Of course, by assigning each population to separate races we would carry this procedure to the reductio ad absurdum. A post facto assignment, based on gene frequencies, would also increase the racial component, but if this were carried out objectively it would lump certain Africans with Lapps! Clearly, if we are to assess the meaning of racial classifications in genetic terms, we must concern ourselves with the usual racial divisions. All things considered, then, the 6.3% of human diversity assignable to race is about right, or a slight overestimate considering that Hpop is overestimated. It is clear that our perception of relatively large differences between human races and subgroups, as compared to the variation within these groups, is indeed a biased perception and that, based on randomly chosen genetic differences, human races and populations are remarkably similar to each other, with the largest part by far of human variation being accounted for by the differences between individuals. Human racial classification is of no social value and is positively destructive of social and human relations. Since such racial classification is now seen to be of virtually no genetic or taxonomic significance either, no justification can be offered for its continuance.

— Lewontin 2026-07-21