2
$\begingroup$

I am not much of a statistician and I have racked my brains around this problem without success: There are w white balls and b black balls in an urn. I want to get the distribution of the random variable S, the number of white balls obtained after n draws. The tricky part is, when a black ball is drawn, it is replaced in the urn, but when a white ball is drawn it is not replaced.

I have tried to start from the demonstrations of the binomial and hypergeometric distributions, but I get a huge expression that I can't simplify and that is increasingly complex as you increase the number of draws n.

It seems to be a rather simple problem but I couldn't find a solution on the internet. Is there a solution to this problem?

Thank you very much!

$\endgroup$
  • $\begingroup$ Hmmm interesting. I don't know if there is a nice answer. $\endgroup$ – Lost1 Jan 5 '14 at 18:18
  • $\begingroup$ Did you try applying coupon collector's problem here? $\endgroup$ – Alex Jan 5 '14 at 18:40
  • $\begingroup$ Yes, but as I understand it the coupons are always replaced in the coupon collector problem, whether you already drew them or not, right? $\endgroup$ – jblugagne Jan 5 '14 at 18:44
  • $\begingroup$ You are probably right. Finding the $\textit{mean time}$ until $r$ white balls are sampled is much easier. $\endgroup$ – Alex Jan 5 '14 at 18:51
  • $\begingroup$ Sorry, I am not sure I understand what you mean. Do you suggest using the geometric distribution to get the expected number of black balls before I sample r white balls? Would it give me the distribution? $\endgroup$ – jblugagne Jan 5 '14 at 18:56
0
$\begingroup$

The number of samples until each success follows Geometric distribution with parameter $p_k$, e.g. $p_1 = \frac{w}{w+b}, \ p_2 = \frac{w-1}{w+b-1}$, etc. The mean time until the success is $\frac{1}{p_k}$, i.e. $\mathbf{E}X_k=\frac{1}{p_k}=\frac{w+b-k}{w-k}$. Mean time until $r^{\text{th}}$ success is $$ \mathbf{E}X=\sum_{k=1}^{r}\mathbf{E}X_k $$

which you can find using Harmonic sum approximation with logarithms, i.e. $\sum_{k=1}^{n}\frac{1}{k}=O(\log n)$.

$\endgroup$
  • $\begingroup$ Indeed, working out expectation is not hard, can do variance similarly. This does not really answer the question which distribution it is, but i don't think there is a way working out pmf of the distribution. The only way to characterise it seems to be mgf or pgf. $\endgroup$ – Lost1 Jan 5 '14 at 22:18
  • $\begingroup$ Yes, the way I see it you have to account for the possibility of sampling a white ball in any time between $1$ and $n$ for $P(S_n=r)$ $\endgroup$ – Alex Jan 6 '14 at 1:00

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.