Very surprised to see no attempt at discussing regression to the mean other than the quickest possible throw away on pg2, when it is so pervasive and must be operating to at least some degree in any result of the form 'after filtering on extreme event Y, variable X decreased/increased!' Most of the data seems pretty silent on this. (Yes, yes, the two groups look pretty similar temporally in raw # of papers before Fields, but you don't get a Fields for simply publishing a lot - by definition the Fields medalists' papers were different from and better than the non-medalists, and that's why they won.)