By the Bayesian Law of Large Numbers for iid data, the post-data measure concentrates on the distribution(s) that minimize the L-divergence (aka the reverse I-divergence). Consequently, methods based on minimization of other divergences are not consistent under misspecification, and in this sense ruled out. This point, which has already been established analytically, is illustrated be means of a Monte Carlo study of the inconsistency under misspecification of the Exponential Empirical Likelihood method that is based on the I-divergence minimization.
Not all empirical divergence minimizing statistical methods are created equal?
9TH INTERNATIONAL CONFERENCE ON MATHEMATICAL PROBLEMS IN ENGINEERING, AEROSPACE AND SCIENCES: ICNPAA 2012 ; 2012 ; Vienna, Austria
AIP Conference Proceedings ; 1493 , 1 ; 432-435
2012-11-06
4 pages
Conference paper
Electronic Resource
English
Not all empirical divergence minimizing statistical methods are created equal?
British Library Conference Proceedings | 2012
|Not All Programmers Are Created Equal
British Library Conference Proceedings | 1994
|Not all solar simulators are created equal
British Library Online Contents | 2011
|Are ATC Subject Matter Experts Created Equal?
British Library Conference Proceedings | 2003
|Not all programmers are created equal — Redux©
IEEE | 2012
|