'Restriction of range'. When you look at the general population, from janitor to tenured professor, one of the single best predictors of job performance is IQ; but when you look at, say, 'MIT physicist grad students', well, they're all very smart, so when you look at the minor differences in intelligence between them, the scores barely correlate with their future success even though no one becomes a MIT grad student in physics without being terrifyingly smart. Because they have been selected on that trait already. Hence the apparent paradox.

(No surprise Friedman doesn't understand this; the NYT article on the Google findings did not explain this.)