Bayesians, Frequentists, and Physicists

PHYSTAT2003 brought statisticians together with particle physicists, astrophysicists, and cosmologists. This paper, which is taken from the text of the keynote address, concerns the uneasy relationship between Bayesian and frequentist statistics, with particular attention to the “neutrino problem”: how to set confidence limits for a parameter known to be non-negative. Model selection, an objective Bayes technique, gives a different answer than the classic Neyman confidence construction.