Bayesian Inference with Specified Prior Marginals

Michael Lavine, Larry Wasserman, R.L. Wolpert
Duke University, Carnegie Mellon University

Jul 10 1990

We show how to find bounds on posterior expectations of arbitrary functions of the parameters when the prior marginals are specified but when the complete joint prior is unspecified. We also give a theorem that is useful for finding posterior bounds in a wide range of Bayesian robustness problems. We apply these techniques to two examples.
The first example involves a recent clinical trial for ECMO (Ware 1989). Our analysis may be regarded as a followup to a detailed Bayesian analysis given by Kass and Greenhouse (1989) who conclude that the posteriror probability that the treatment is superior to the control is about 0.95. However, their anaylsis assumes a priori independence of the parameters. We consider other prior distributions with the same marginals as KAss and Greenhouse, but in which the parameters are not independent, and conclude that as longa s a priori independence is at least approximately tenable then ECMO seems uperior to the control.
The second example is the product of means problem which has been studied in the Bayesian context by Berger and Bernardo (1989). Here the goal is to find the posterior expectation of αβ where α and β are the means of conditionally independent random variables X and Y. Berger and Bernardo recommend a joint prior π0 propoertional to (α2 + β2)1/2. We find that among all priors with the same marginals as π0, the posterior expectation of αβ can be made arbitrarily large or arbitrarily close to 0. Furthermore, the parameterization is imporant: with a different parameterization the upper bound is strictly finite.


PDF icon 1990-03.pdf