Posterior sensitivity to the sampling distribution and the prior: More than one observation (Q1962602): Difference between revisions

Bayesian analysis requires specification of the sampling distribution \(f\) and the prior distribution \(P\). When considering the effect of one's modeling assumptions, perturbations of both \(f\) and \(P\) should be taken into account. In the present paper sensitivity of a posterior quantity \(\rho(f, P)\) to the choice of \(f\) and \(P\) is considered. Sensitivity is measured by the range of \(\rho(f, P)\) when \(f\) and \(P\) vary in nonparametric classes \(\Gamma^f\) and \(\Gamma^P\), respectively. Direct and iterative methods are described which obtain the range of \(\rho(f, P)\) over \(f\in \Gamma^f\) when \(P\) is fixed, and also the overall range over \(f\in \Gamma^f\) and \(P\in \Gamma^P\). The method of steepest descent is proposed to obtain the range of \(\rho(f, P)\). Illustrative examples are provided.