Bayes' rule $$ \pi(\theta|y)\propto\pi(y|\theta)\pi(\theta) $$ gives the posterior distribution. However for the first time I have encountered a problem where I have two parameters (because the likelihood function is a gamma distribution), $\alpha$ and $\beta$.

  • I know the likelihood function (a gamma distribution).
  • I know the distribution for $\beta$ given that $\alpha$ is known (a gamma distribution).
  • I know the distribution of $\alpha$ (an exponential distribution).
  • I have data $y$.

The goal is to find the posterior distribution. In general, how do I approach a problem like this? How can I find the the prior when the prior is given as two separate distributions and one depends on the other?


1 Answer 1


You can use the product rule $P(A, B) = P(A) P(B | A)$ to figure this out.

Take your model above. You can start with the joint distribution over everything and expand it recursively until you have everything you need:

\begin{align*} \underbrace{\pi(y, \alpha, \beta)}_{\text{joint distribution}} & = \underbrace{\pi(y \, | \, \alpha, \beta)}_{\text{likelihood}} \underbrace{\pi(\alpha, \beta)}_{\text{prior}} \\ & = \pi(y \, | \, \alpha, \beta) \underbrace{\pi(\beta \, | \, \alpha)}_{\text{gamma}} \underbrace{\pi(\alpha)}_{\text{exponential}} \end{align*}

You can similarly expand the left-hand side into $\pi(\alpha, \beta \, | \, y) \pi(y)$, and then dividing everything by the latter term gives you Bayes' theorem.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.