Lewontin's Statistic vs. Edwards' Correlations: A Fallacy Filed Under Its Own Refutation
2026-07-24 · AI-generated insight
Read together, Lewontin and the Winegard–Carl review stage one of the most famous arguments in the human sciences—and the commonplace book has quietly filed the answer next to the question. Lewontin's claim rests on a single-variable intuition: pick a random genetic difference, and most of the variation sits within groups rather than between them. From this he leaps to the conclusion that classification has "virtually no genetic or taxonomic significance."
But notice the tag under the second copy of his quote: Correlation Across Many Variables. That is precisely the hinge. Winegard and Carl report that researchers classify skulls into groups with ~80% accuracy using two variables—and the accuracy climbs as variables are combined. This is A.W.F. Edwards' rebuttal in miniature ("Lewontin's Fallacy"): any single locus is nearly uninformative, yet the correlational structure across many loci is highly informative. Lewontin's statistic is true and his inference does not follow. The book has, perhaps without meaning to, placed the counterexample inside the very argument it undermines.
What makes this more than a technical footnote is the Coon episode. There, an entire professional association condemned a book that exactly one member had read—the archetype of condemnation without examination. The Lewontin passage risks becoming the mirror image: a conclusion so congenial ("of no social value") that it gets accepted without examination, its logical gap unnoticed because the moral verdict feels settled in advance.
So the three notes describe a single failure mode viewed from two ends. Coon's colleagues rejected a claim they hadn't scrutinized; Lewontin's readers, arguably, embraced one they hadn't scrutinized. In both cases the missing step is the same—actually working through the evidence rather than the valence. The uncomfortable synthesis: the phrase "science against prejudice" can smuggle in its own prejudgment, and the discipline of checking whether the math supports the conclusion is exactly what Coon demanded when he asked who in the room had read the book.