2DkcTXceO

30 A career in statisticsproblem, but often unclear to outsiders and students. Fisher viewed it asan unnecessary mathematization, but the philosophical issue was important.Years later Neyman gave a talk in which he pointed out that the NP Lemmawas highly touted but trivial. On the other hand it took him years of thinkingto understand and state the issue.Just before graduation I received a telegram offering me a position, whichI accepted, as Junior Physicist at the Dahlgren Naval Proving Grounds inVirginia. After a year and a half I left Dahlgren to study applied mathematicsat Brown University in Providence, Rhode Island. Dean Richardson hadset up a program in applied mathematics where he could use many of thedistinguished European mathematician émigrés to teach and get involved inresearch for the Defense Department, while many of the regular faculty wereaway working at Defense establishments. There was a good deal of comingand going of faculty, students and interesting visitors during this program.During the following year I worked very hard as a Research Assistantfor Professor Stefan Bergman and took many courses and audited a couple.I wrote a Master’s thesis under Bergman on the growth of solutions of partialdifferential equations generated by his method, and received an ScM degree.One of the courses I audited was given by Professor Willy Feller, in which hislectures were a preliminary to the first volume of his two volume outstandingbooks on probability.During the following summer, I took a reading course in statistics from ProfessorHenry Mann, a number theorist who had become interested in statisticsbecause some number theory issues were predominant in some of the workgoing on in experimental design. In fact, he had coauthored a paper withAbraham Wald (Mann and Wald, 1943) on how the o and O notation couldbe extended to o p and O p .ThispaperalsoprovedthatifX n has as its limitingdistribution that of Y and g is a continuous function, then g(X n )hasasitslimiting distribution that of g(Y ).Mann gave me a paper by Wald (1939) on a generalization of inferencewhich handled that of estimation and testing simultaneously. Once moreI found this paper revolutionary. This was apparently Wald’s first paper on decisiontheory. Although it did not resemble the later version of a game againstnature, it clearly indicated the importance of cost considerations in statisticalphilosophy. Later discussions with Allen Wallis indicated that Wald had beenaware of von Neumann’s ideas about game theory. My theory is that in thisfirst paper, he had not recognized the relevance, but as his work in this fieldgrew, the formulation gradually changed to make the relationship with gametheory clearer. Certainly the role of mixed strategies in both fields made therelation apparent.At the end of the summer, I received two letters. One offered me a predoctoralNSF fellowship and the other an invitation I could not decline, to jointhe US Army. It was 1945, the war had ended, and the draft boards did notsee much advantage in deferring young men engaged in research on the wareffort. I was ordered to appear at Fort Devens, Massachusetts, where I was