I am studying hierarchical models, and trying to understand a point in the book where they try to decide on a non-informative hyperprior distribution.

The hyperparameters is $\alpha$ and $\beta$ for a beta distribution. So we are interested in the density $p(\alpha, \beta)$. In the book however they start looking at a reparameterize version instead, $p(log(\alpha/\beta), log(\alpha + \beta))$

It is not clear why this is done in the book. Could anybody explain?

Also, when we have a density such as $p(log(\alpha/\beta), log(\alpha + \beta))$, whats the steps using change of variable formula to get back to $p(\alpha, \beta)$?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.