In the middle of page 64 of the third edition of Bayesian Data Analysis, Gelman writes...

We saw in Chapter 2 that a sensibly vague prior for $\mu$ and $\sigma^2$, assuming prior indipendance of scale and location, is uniform on $(\mu, \sigma^2)$, or eqivalently $$ p(\mu, \sigma^2) \propto (\sigma^2)^{-1} $$

I don't think I saw where this was referenced. Gelman writes that a prior or a normal with known mean but unknown variance is an Inverse Chisquared distribution. This distribution has two hyperparameters, $\alpha, \beta$, which if were both set to 0 would yeild $(\sigma^2)^{-1}$.

Is that the logic behind the prior for a normal with two unknown params?


The flat prior for $\mu$ with support on $(-\infty,\infty)$ is $p(\mu) \propto 1$, i.e. an improper uniform prior over the real line.

The independence assumption gives you $p(\mu,\sigma^2) = p(\mu)p(\sigma^2) \propto (\sigma^2)^{-1}$.


Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.