In "real science" what is shown to be true this year will probably be superseded within a couple of years, but when you prove a mathematical theorem it is "true" forever. Buried in past issues of the Annal of Math Stat, Biometrika, JRSS, JASA, etc., are a great many answers to questions people may or may not be asking. Often, the major work of a great man like Neyman gets codified by the textbook writers, and his (or her) later work is ignored. Neyman, in his final years, was pushing the use of contaminated distributions. His fellow faculty at UC Berkeley were pushing his idea of restricted significance testing. Both of these insights should have extensive usage in modern problems but are ignored. Just before he died, Kolomogorov was working on a modified definition of probability, not as a mathematical measure but as the limit of sequences of numbers. Modern problems of N<<p are just beginning to be attacked with nearest neighbor algorithms, but nearest neighbor techniques have been around since at least 1965.
I used to enjoy browsing through the library volumes of back issues of journals. Alas, nowadays we all locate specific articles via JSTOR and the internet and never have a chance to examine the forgotten gems that come just before or just after that article we thought we wanted. Perhaps we should train some of our graduate students to be archeologists, digging through the layers of theorems that are just as "true" now as they were when they first graced the printed page.
------------------------------
David Salsburg
------------------------------