Approximate Bayesian Inference for Quantiles

Michael Lavine
Duke University

Nov 30 1992

Let Z1,...,Zn be a random sample from F, an uncertain one dimensional cdf, and suppose that a prior distribution is available for some of the quantiles of F. Bayesian inference is difficult because the distribution of the data given the quantiles is not fully specified. This paper compares two methods of approximating the posterior, shows that they provide conservative inferences and gives guidance for choosing between them.


PDF icon 1993-12.pdf