Thursday 22 Mar 2018: Statistical Science seminar- Richard Wilkinson "Subjectivism in inference"
Richard Wilkinson - University of Sheffield
We are used to the idea of subjective choice in prior and model selection. If we move away from the main inferential paradigms (maximum likelihood, Bayes) then there is also potentially need for subjective judgements in the inferential approach taken. Indeed, approaches like approximate Bayesian computation, history matching, and scoring rule based approaches, can be seen in this light. In this talk I'll review some of these areas, and discuss some of the benefits of such approaches. In particular, using inferential approaches that only utilise part of the information in the data, using scores that are robust to misspecification, and thresholding not optimising scores, all look to be promising ways of dealing with model mis-specification.