This is very basic, but I have been stuck on this problem for a while. Suppose $Y_1, \dots, Y_n|\alpha,\beta\sim Gamma(\alpha, \beta)$ is iid with $\alpha$ known. I want conjugate prior for $\beta$ and the posterior.

My work:

$p(\beta|y_1, \dots, y_n)=p(y_1, \dots, y_n|\beta)p(\beta)=(\prod_{i=1}^{n} y_i)^{\alpha-1} \exp(-\beta\sum_{i} y_i) p(\beta)$

$\propto \exp(-\beta\sum_{i} y_i) p(\beta)$.

Therefore, the conjugate prior for $\beta$ would be gamma$(\alpha_0, \beta_0)$.

In this case, we can derive the posterior as:

$p(\beta|y_1,\dots, y_n)\propto \beta^{\alpha_0 -1} \exp(-\beta(\sum_{i} y_i+\beta_0))$.

So, the posterior is gamma$(\alpha_0 , \sum_{i} y_i+\beta_0)$.

However, Wikipedia says the posterior should be gamma$(\alpha_0 +n\alpha, \sum_{i} y_i+\beta_0)$. I don't understand where does $n\alpha$ come from. I also don't understand why my derivation is not correct. Thanks in advance!


I think you may have missed an important term from the likelihood

  • The density of a Gamma distribution with parameters $\alpha$ and $\beta$ is $\frac{1}{\Gamma(\alpha)}\beta^{\alpha }y^{\alpha-1}\exp(-\beta y)$ when $y \gt 0$

  • so the likelihood for the observations here is proportional (with terms involving $\beta$) to $$\beta^{\alpha n}\exp\left(-\beta \sum_{i=1}^{n} y_i\right)$$

  • while your prior for $\beta$ is proportional (with terms involving $\beta$) to $$\beta^{\alpha_0-1}\exp(-\beta_0 \beta)$$

  • so the posterior for $\beta$ is proportional to $$\beta^{\alpha_0+\alpha n-1}\exp\left(-\beta\left(\beta_0+\sum_{i=1}^{n} y_i\right)\right)$$

which is proportional to the density of a Gamma distribution with parameters $\alpha_0+\alpha n$ and $\beta_0+\sum_{i=1}^{n} y_i$


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.