error in calculating the posterior in basic urn sampling problem

25 Views Asked by At

As I'm reading E.T. Jaynes' Probability Theory and while trying to do some calculations by myself (in this particular case, re-derive equation 6.15) I'm running into a problem where I just can't find my mistake:

We have an urn containing N balls in total of which R are red and the remaining N-R are white. N is known and we draw n balls at random of which r are red. Given the outcome of our draws we now try to estimate R using Bayes theorem:

$$ \underbrace{p(R | r, n, N)}_{\text{posterior}} = \underbrace{p(R | N)}_{\text{prior}} \frac{\overbrace{p(r | n, N, R)}^{\text{sampling distribution}}}{\underbrace{p(r | n, N)}_{\text{marginalization}}} $$

Using a uniform prior and the hypergeometric distribution as the sampling distribution:

\begin{align*} \text{prior} &= \frac{1}{N+1} \\ \text{sampling distribution} &= \frac{{{R}\choose{r}} {{N-R}\choose{n-r}}}{{{N}\choose{n}}} \\ \text{marginalization} &= \frac{\sum_{R=0}^N {{R}\choose{r}} {{N-R}\choose{n-r}}}{{{N}\choose{n}}} \stackrel{\text{vandermonde's idenity}}{=} \frac{{{N+1}\choose{n+1}}}{{{N}\choose{n}}} \end{align*}

the ${N \choose n}$ cancel out and I get: $$ p(R | r, n, N) = \frac{1}{N+1} \frac{{{R}\choose{r}} {{N-R}\choose{n-r}}}{{{N+1}\choose{n+1}}}$$

But this is not properly normalized. The prior $\frac{1}{N+1}$ should cancel out too somehow to arrive at the correct result of just $\frac{{{R}\choose{r}} {{N-R}\choose{n-r}}}{{{N+1}\choose{n+1}}}$.

I just don't see how. What am I missing/what is wrong?

1

There are 1 best solutions below

0
On

And just after posting the question, a friend pointed out the error to me: the marginalization should be: $$\sum_{R=0}^N p(r,R|n,N) = \sum_{R=0}^N p(r|R,n,N) \underbrace{p(R|n,N)}_{=p(R|N)}$$