1
$\begingroup$

I'm reading A First Course in Bayesian Statistical Methods by P. Hoff where he is using Jeffrey's prior (J) and Unit information prior (U) for Normal model. For example we can derive Jeffrey's prior for Normal model and get $p_{J}(\theta, \sigma^2) = (\sigma^2)^{-3/2}$ and then obtain a posterior $p(\theta, \sigma^2 | y_{1}, ..., y_{n})$ as the product of prior and likelihood. I don't understand what he means by probability density with subscript J, for example: $p_{J}(\theta, \sigma^{2} | y_{1}, ..., y_{n})$ or $p_{J}(\theta | \sigma^2, y_{1}, ..., y_{n})$. Can someone please explain?

$\endgroup$
  • 1
    $\begingroup$ In the context of the problem, it means the posterior pdf associated with Jeffrey's prior. i.e. it is $p_J(\theta,\sigma^2|y_1,...,y_n)\propto p_J(\theta,\sigma^2)p(y_1,...,y_n|\theta,\sigma^2)$ $\endgroup$ – stats134711 Apr 24 '17 at 13:30
  • $\begingroup$ Thank you for your answer, so it is only the posterior we obtain as the product of Jeffrey's prior and likelihood? And what is $p_{J}(\theta | \sigma^2, y_{1}, ..., y_{n})$ and $p_{J}(\sigma^2 | y_{1}, ..., y_{n})$ supposed to be and how to derive it? $\endgroup$ – Marko Lalović Apr 24 '17 at 14:00
  • $\begingroup$ In general the posterior is the product of the prior and the likelihood. The subscript J is only to indicate that in the particular example the Jeffrey's prior is used. Consequently the $p_{J}(\theta|\ldots)$ and $p_{J}(\sigma^2|\ldots)$ are the full conditional for the parameters $\theta,\sigma^2$ which are obtained from the full posterior under the Jeffreys prior. $\endgroup$ – Christos Apr 24 '17 at 14:16
1
$\begingroup$

In general the posterior is the product of the prior and the likelihood. The subscript J is only to indicate that in the particular example the Jeffrey's prior is used. Consequently the

$p_{J}(\theta|\ldots)$ and $p_{J}(\sigma^2|\ldots)$

are the full conditional for the parameters $\theta,\sigma^2$ which are obtained from the full posterior under the Jeffreys prior. You derive them from the full posterior as the distribution depending only on the parameter of interest up to a constant of proportionality.

For example suppose that we have a posterior of the form

$ p(\mu,\lambda|\ldots) \propto \lambda^{\frac{1}{2}}e^{-\frac{\lambda}{2}(x-\mu)^2}.$

Then the full conditional for $\lambda$ is

$p(\lambda|\ldots) \propto \lambda^{\frac{1}{2}}e^{-\frac{\lambda}{2}(x-\mu)^2},$ which is a $Ga\left(\frac{3}{2},\frac{1}{2}(x-\mu)^2\right),$

and for $\mu$

$p(\mu|\ldots) \propto e^{-\frac{\lambda}{2}(x-\mu)^2},$ which after some algebra is a normal distribution.

$\endgroup$
  • $\begingroup$ Thank you for your answer. I understand now. For my example, Normal model with Jeffrey's prior, I derived full conditionals $p_{J}(\theta | \sigma^{2}, y_{1}, ..., y_{n}) \propto \texttt{Normal}(\bar{y}, \sigma^2/n)$ and $p_{J}(\sigma^{2} | y_{1}, ..., y_{n}) \propto \texttt{InverseGamma}(n/2, \sum_{i=1}^{n}(y_{i} - \bar{y})^{2}/2)$. $\endgroup$ – Marko Lalović Apr 24 '17 at 16:16
  • $\begingroup$ No problem at all. I am glad i helped! $\endgroup$ – Christos Apr 25 '17 at 0:10

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.