Ranges of posterior measures for priors with unimodal contaminations (Q758036)

Consider that we observe X with density function f(x\(| \theta)\) and desire to perform a Bayesian analysis concerning the parameter \(\theta\). It may not be possible to quantify exactly the prior distribution of \(\theta\). Consider that after an elicitation process, we are led to a prior \(\pi_ 0\) and that any prior close to \(\pi_ 0\) could also be a reasonable representation of the prior belief. Thus we consider that the uncertainty in the elicited prior is modelled by an \(\epsilon\)- contaminated class \(\Gamma\), \[ \Gamma =\{\pi =(1-\epsilon)\pi_ 0+\epsilon q:\;q\in Q\}, \] where Q is the class of allowed contaminations. This paper considers the range of the posterior quantity \(\rho\) (x,\(\pi\)) as \(\pi\) ranges over \(\Gamma\). The quantities considered are the posterior mean, posterior variance and posterior probability of a set. It is assumed that the base prior \(\pi_ 0\) is unimodal. It is considered that Q the class of contaminations is (i) the class of all unimodal distributions and (ii) the class of all symmetric unimodal distributions. Examples with normal density functions are given.