Suppose I have an urn with an infinite number of balls which can be either red or white. I do not know what the proportion of each colour is, but I do know it's a fixed proportion. After drawing $N$ balls, I have observed $r$ red ones and $w$ white ones.

I believe the probability that I will observe a red ball on the next draw is given by Laplace's Law of Succession, $\frac{r+1}{N+2}$. However, how sure should I be of that? That is, before I drew any balls, I believed any proportion other than $0$ or $1$ was the true one. After I drew those $N$ balls, what should be my estimate pdf over the possible values for the proportion of red balls in the urn?

  • 2
    $\begingroup$ The probability you will observe a red ball on the next draw is always the same: it equals the proportion of red balls. Are you perhaps asking how to update an estimate of that probability? If so, you are either performing a Bayesian analysis (whose answer depends on your prior probability distribution for the proportion in the urn) or a non-Bayesian analysis (which, depending on its form, can yield a "fuzzy" probability or a "p-box" or a host of other non-probabilistic representations of your uncertainty). In any case, this question is unanswerable without more information. $\endgroup$
    – whuber
    Commented Nov 8, 2013 at 17:19
  • $\begingroup$ Yes, sorry, that's what I meant, it's an estimate of the probability. $\endgroup$
    – Red
    Commented Nov 8, 2013 at 17:28

1 Answer 1


Okay, so, my first idea was using Jaynes' $A_p$ distribution, in which case we can define $A$ = next draw will be red and $N_r$ = out of $N$ draws, $r$ were red. With an ignorant prior distribution $(A_p|X) = 1$, I get that

$$(A_p|N_rX) = (A_p|X)\frac{P(N_r|A_p)}{P(N_r|X)}$$

We know that

$$P(N_r|A_p) = \binom{N}{r}p^r(1-p)^{N-r}$$

And we can find

$$P(N_r|X) = \int^1_0(N_rA_p|X)dp = \int^1_0P(N_r|A_p)(A_p|X)dp = \int^1_0\binom{N}{r}p^r(1-p)^{N-r}dp$$

from which

$$P(N_r|X) = \frac 1 {N+1}, 0 \leq r \leq N$$

And then

$$(A_p|N_r) = (N+1)\binom{N}{r}p^r(1-p)^{N-r}$$

And that looks like my distribution. Is that correct?


Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.