Title: A predictive approach to statistical problems with multiplicity
Authors: Fumiyasu Komaki - The University of Tokyo (Japan) [presenting]
Abstract: Appropriate adjustments for multiplicity become essential in various statistical problems. Bayesian prediction based on parametric models with multiplicity is investigated. Priors for unknown parameters are constructed by using conditional mutual information between future observables and unknown parameters given observations. The priors depend not only on the parametric model of observed data but also on the choice of target variables to be predicted. The priors are often quite different from the Jeffreys priors or other widely used objective priors. In principle, our approach can be applied to various statistical problems with multiplicity by formulating them as prediction problems.