Consider a pair of RVs $X$ and $Y$, with the following conditional distributions:

$$X | Y=y \sim Binom(L, y)$$

$$Y | X=x \sim Beta(\alpha + x, \nu)$$

where $L$, $\alpha$, and $\nu$; are all positive ($L$ is an integer of course). Is there a name for the joint distribution of $(X,Y)$? Or perhaps for the marginal distribution of $Y$? I think that if $x$ is eliminated from the shape "parameter" of the beta distribution, then $X$ is beta-binomial distributed. But in the above bivariate model, $X=x$ affects the shape parameter for the conditional distribution of $Y$, so I do not think $X$ is beta-binomial distributed. Apologies if the above makes no sense; I am not very knowledgeable about probability and statistics.

  • $\begingroup$ Thanks for the formatting edits, mpiktas. Didn't realize that the CV site engine can process LaTeX-style formatting. Nice. $\endgroup$ – Stephen Ramsey May 8 '15 at 15:40

$\begin{align} P(x,y) &= \frac{\binom{n}{x}}{\beta(\alpha+x,\nu)} \times y^{\alpha + 2x-1 }(1-y)^{n-x+\nu-1} \\&= \frac{\binom{n}{x}}{\beta(\alpha+x,\nu)} \times \frac{y^{\alpha + 2x-1 }(1-y)^{n-x+\nu-1}}{\beta(\alpha+2x,n-x+\nu)} \times \beta(\alpha+2x,n-x+\nu) \end{align}$

Integrating w.r.t $y$:

$\begin{align}P(x) &= \frac{\binom{n}{x}}{\beta(\alpha+x,\nu)} \times 1 \times \beta(\alpha+2x,n-x+\nu) \\&= \binom{n}{x} \times \frac{\Gamma(\alpha+2x) \Gamma(n-x+v)}{\Gamma(\alpha +n +x + \nu)} \times \frac{\Gamma(\alpha+x+\nu)}{\Gamma(\alpha+x)\Gamma(\nu)} \end{align} $

This can be further simplified by using properties of the Gamma function, but doesn't lead to a beta binomial $X$

| 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.