For which distributions x, other than beta, is the x-binomial distribution nice? The beta and binomial distributions are famously conjugate but I am curious if other non-conjugate distributions will give comparably simple compound pmfs. By nice I mean that the pmf is nice to compute without resorting to numerical integration; I'm not referring to the easyness of sampling from the compound distribution.

  • $\begingroup$ Nice question. I was going to suggest a Jeffreys prior, but it turns out that's just a special case of the beta distribution for binomial models. I'll keep thinking and see if I can come up with any others. I have some vague sense that there are transformations of Gaussians that work, but Google isn't helping me pin that down. $\endgroup$ – David J. Harris Aug 22 '12 at 22:14

If you just want to be able to write down the probability mass function, you have a lot of flexibility, basically because you can repeatedly use integration by parts. As long as you can integrate the distribution of $X$ repeatedly to get closed form expressions, you get at worst a double sum of closed form expressions for the pmf of the compound distribution. I think you often get single sums, so maybe there is an even simpler way of expressing these.

For example, let $\text{pdf}_X(x) = -\log x$ on $[0,1]$.

$$ \int_0^1 {N\choose k}x^k (1-x)^{N-k} (-\log x) ~dx \\\ = \frac{1}{N+1}\sum_{i=k}^{N} \frac1{i+1}.$$

| cite | improve this answer | |

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.