Jeffreys Priors and Alternative Noninformative Solutions for Location-Scale Mixtures

• Activity Number: 601 - Prior Specifications for Finite Bayesian Mixture Models
• Type: Invited
• Date/Time: Thursday, August 2, 2018 : 8:30 AM to 10:20 AM
• Sponsor: International Society for Bayesian Analysis (ISBA)
• Author(s): Christian Robert* and Clara Grazian
• Companies: Universite Paris-Dauphine and University of Oxford
• Keywords: mixtures; Bayesian Analysis; Jeffreys priors; non-informative priors

Abstract

While Jeffreys priors usually are defined for the parameters of mixtures of distributions, they are not available in closed form. Furthermore, they often are improper priors. We study in this talk the implementation and the properties of Jeffreys priors in several mixture settings, show that the associated posterior distributions are most often improper, and then propose different non-informative alternatives for the analysis of location-scale mixtures.