Likelihood is central to statistical inference, and classical results on the properties of maximum likelihood estimators, likelihood ratio statistics etc. are widely used. These results are not always accurate, but they can be improved to be highly accurate without much further effort.  After reviewing the main ideas in the scalar case, the talk focuses on improved tests on vector parameters, which have been much less studied.  It turns out that extremely high accuracy is also possible even in high-dimensional settings.  The work is joint with Don Fraser, Nancy Reid and Nicola Sartori.  

Anthony Davison is professor of statistics at EPFL, Lausanne, prior to which he held posts at universities in Texas, London and Oxford.  His main current research interests are extreme-value statistics and likelihood methods, though he has published on a wide range of topics.  He has been editor of Biometrika and joint editor of Journal of the Royal Statistical Society, series B.