CURRENT RESEARCH TOPICS

Three B upper limits. Berger and Boos (1994) have pioneered a method of constructing a valid test by maximizing P-values over a confidence set for the nuisance parameters rather than the whole parameter space. This construction can be adapted to the Buehler upper confidence limit where maximizing over the nuisance parameter is often numerically difficult. We have adapted the Berger-Boos construction to this context and have also extended the result to allow the restricted set to be a confidence region for all parameters, not just the nuisance parameters. We call these “3B” upper limits since they combine ideas of Berger, Boos and Buehler. The study will report the computation reductions that these intervals enjoy as well as the possible loss of statistical efficiency. (This is joint work with my Doctoral Student, Max Moldovan.)