ROBUST BAYESIAN ANALYSIS WITH EPSILON- CONTAMINATED CLASSES (UNIMODAL)
Abstract
In Bayesian analysis, elicitation of prior information can realistically be done only in terms of a class (GAMMA) of all plausible prior distributions. Only the procedures which are robust as prior distribution (pi) varies over (GAMMA) are therefore desirable. Carrying out a robustness analysis requires methods of finding ranges of various posterior criteria. An attractive way of quantifying prior information is in terms of the (epsilon)- contaminated classes. In this thesis, we develop methods of finding the ranges of some posterior criteria for various (epsilon)- contaminated classes.
Degree
Ph.D.
Subject Area
Statistics
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.