Aggregation as an Epistemic Method: Weak Signals, Strong Conclusions

2026-08-20 · AI-generated insight

There's a structural rhyme between these two notes that goes deeper than their shared subject matter. Both are arguments about how aggregation converts weak evidence into strong evidence — and reading them together reveals that this is the load-bearing move in the entire field.

Winegard and Carl's point about Lewontin is explicitly statistical: any single genetic locus (or nose size) carries almost no classificatory information, but correlated differences across many loci make population structure — like sex differences in faces — nearly unmistakable. The signal lives not in any variable but in the correlation structure across variables.

Warne's abstract makes the same move one level up, epistemologically rather than statistically. He concedes that no single line of evidence is conclusive — not Spearman's hypothesis, not measurement invariance, not admixture studies. His argument is that five converging lines jointly support a conclusion none supports alone. This is Lewontin's fallacy inverted into a method: don't evaluate each piece of evidence in isolation, ask whether they point the same direction.

The interesting tension is that aggregation is only as reliable as the independence of what's being aggregated. Correlated loci genuinely carry independent information about ancestry because recombination shuffles them separately. But are Warne's five lines of evidence similarly independent? If several of them share a common vulnerability — say, a subtle environmental confound that mimics genetic signal in polygenic scores, admixture studies, and heritability decompositions alike — then convergence is less impressive than it appears. Correlated evidence can mean triangulation on truth, or it can mean a shared blind spot replicated five times. Lewontin's critics won the statistical argument precisely because they could show the loci were informative conditional on each other; the analogous demonstration for lines of scientific evidence is much harder to make.

So the pair of notes suggests a useful question to carry elsewhere: when someone offers "converging lines of evidence," the right response is neither dismissal-by-isolation (Lewontin's error) nor automatic deference to convergence, but an audit of whether the convergence is genuine — whether the lines would fail differently if the conclusion were false.

Woven from these notes

The past 30 years of research in intelligence has produced a wealth of knowledge about the causes and consequences of differences in intelligence between individuals, and today mainstream opinion is that individual differences in intelligence are caused by both genetic and environmental influences. Much more contentious is the discussion over the cause of mean intelligence differences between racial or ethnic groups. In contrast to the general consensus that interindividual differences are both genetic and environmental in origin, some claim that mean intelligence differences between racial groups are completely environmental in origin, whereas others postulate a mix of genetic and environmental causes. In this article I discuss 5 lines of research that provide evidence that mean differences in intelligence between racial and ethnic groups are partially genetic. These lines of evidence are findings in support of Spearman’s hypothesis, consistent results from tests of measurement invariance across American racial groups, the mathematical relationship that exists for between-group and within-group sources of heritability, genomic data derived from genome-wide association studies of intelligence and polygenic scores applied to diverse samples, and admixture studies. I also discuss future potential lines of evidence regarding the causes of average group differences across racial groups. However, the data are not fully conclusive, and the exact degree to which genes influence intergroup mean differences in intelligence is not known. This discussion applies only to native English speakers born in the United States and not necessarily to any other human populations.

— Russell Warne Between-Group Mean Differences in Intelligence in the United States Are >0% Genetically Caused: Five Converging Lines of Evidence 2026-07-20

In brief, Lewontin ignored the fact that differences among human populations are correlated. Although population variation at any one genetic locus tends to be small, global population structure becomes clear if one examines correlated differences across loci. For a simple analogy, consider men and women’s faces. If one takes any particular characteristic, say nose size, there is likely to be more variation within a sex than between the sexes (and it would be nearly impossible to classify faces by sex with any accuracy using only nose size). However, if one considers all facial characteristics together (which is, after all, how we actually experience human faces), then sex differences become sufficiently clear that an observer can guess the correct sex more than 95 percent of the time.

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