Two hyperpriors
Topic: Hierarchical Bayes · more questions on this topic
‹ Exchangeability is a scientific claim Sample from a grid ›
Transform independent \operatorname{Exp}(\lambda) priors on (\alpha,\beta) to (u,t)=(\alpha/(\alpha+\beta),\alpha+\beta). Then transform p(\alpha,\beta)\propto(\alpha+\beta)^{-5/2} to the same coordinates. Compare properness, the induced prior on u, and the behaviour near t=0 and t=\infty.