URA 1378 Publication 9401
Auteur : GOUTIS Constantinos and ROBERT Christian P.

Titre : What's robust, anyway? (A discussion of James Berger's ``An overview of Robust Bayesian Analysis'')

Année : 1994

Référence : publié dans

Mots-clefs :

Classification AMS :

Résumé :

Abstract :
We first want to congratulate Professor Berger on such an extensive coverage of the multiple facets of robust Bayesian analysis and for giving us further insights on the various issues at stake in this fascinating field. We want to take advantage of this tribune to address some directions only briefly mentioned by Berger. We discuss the ambivalent aspects of robust Bayesian analysis and, as a corollary, the problem of defining a measure of robustness. We then move to a related issue, namely the goal of re-placing robust Bayesian analysis within a decision-theoretic framework with loss functions, merely exhibiting some open problems in this direction. We conclude this discussion with remarks on a robustification of conjugate priors by mixture distributions and their wide diversity, along with indications on the corresponding inference in such setups.

Pour obtenir le fichier pub9401.ps.gz.