What is the posterior distribution for parameter $b$ with $X \sim Gamma(a,b)$, under the Jeffreys prior? We can assume that $a$ is known.

The Jeffreys prior is the square of the Fisher information of $b$:


Then using Bayes' rule we have

$p(b|x) \propto p(x|b) \,p(b) = \dfrac{b^a}{\Gamma (a)}x^{a-1}e^{-xb}\cdot\frac{\sqrt(a)}{b}$

Next we look for the kernel of a Gamma distribution. But this is where I am stuck.

What is the next step for deriving the posterior distribution for $b$?

  • $\begingroup$ It's not obvious. stats.org.uk/priors/noninformative/YangBerger1998.pdf $\endgroup$ – HStamper Feb 18 '16 at 6:55
  • $\begingroup$ @EricMittman: I do not understand the link since the authors write therein (p.13) that $b$ being a scale parameter the natural choice is $\pi(b)=1/b$. $\endgroup$ – Xi'an Feb 18 '16 at 7:34
  • $\begingroup$ @Xi'an: I didn't realize that $a$ was considered to be known. I was thinking of the Jeffrey's prior for $(a,b)$. $\endgroup$ – HStamper Feb 18 '16 at 17:27
  • $\begingroup$ @EricMittman I'll edit the question so that this is clear. $\endgroup$ – ADB Feb 18 '16 at 22:39

It is difficult to perceive where you get stuck: $$\begin{align*}p(b|x)&\propto \dfrac{b^a}{\Gamma (a)}x^{a-1}e^{-xb}\cdot\dfrac{\sqrt{a}}{b} \\ &\propto b^{a-1} e^{-xb} \\ &\propto \dfrac{x^a\,b^{a-1}}{\Gamma(a)}\,e^{-xb}\end{align*}$$ which shows the posterior is a Gamma $\mathcal{G}(a,x)$ distribution.

  • $\begingroup$ I didn't see that $\sqrt{a}$ could be absorbed into the proportionality. Later it seems that $x^a$ appears from nowhere... Is this allowed because in the posterior we are conditioning on $x$, which means that it is fixed (constant)? $\endgroup$ – ADB Feb 18 '16 at 7:45
  • $\begingroup$ Indeed, everything but $b$ is a constant for the proportionality $\propto$ sign. $\endgroup$ – Xi'an Feb 18 '16 at 8:39

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.