Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top asks the following question.

I take the bus to work every day. Every bus has a serial number, but unlike in the German Tank Problem, I don't know if they are numbered uniformly $1...n$.

Suppose the first $k$ buses are all different, but on day $k+1$ I take one I've been on before. What is the best estimate for the total number of buses?

The provided answer gives a maximum likelihood estimator as well as an unbiased estimator of $k(k+1)/2$

If you know the number of buses can't be larger than some given value $N \geq k+1$, how does that change the maximum likelihood estimator and the unbiased estimator ?

share|cite|improve this question
up vote 0 down vote accepted

The unbiased estimator doesn't change. The requirement that the expected value of the estimate is $n$ for all $N$ possible values of $n$ yields $N$ linear contraints on the $N$ estimates of $n$ for the $N$ different possible values of $k$. This $N\times N$ system of linear equations is triangular with non-zero diagonal and thus has a unique solution; since we know that $k(k+1)/2$ is a solution, this is the only unbiased estimator. It yields estimates of $n$ greater than $N$ for most values of $k$, but there is no unbiased estimator without this undesirable property. This is also intuitively clear, since for $n=N$ the estimates below $N$ must be compensated by estimates above $N$ to obtain the expected value $N$.

The maximum likelihood estimator changes in that if the original maximum likelihood estimate would have been greater than $N$, this should be replaced by $N$, since the likelihood is unimodal and thus takes its maximum on $\{1,\dotsc,N\}$ at $N$ if the maximum likelihood estimate is greater than $N$.

share|cite|improve this answer
so the unbiased estimator seems like a terrible idea in this case. Is there a better choice? – Anush Dec 6 '12 at 9:01
@Anush: It's not as terrible as it may seem, since estimates above $N$ are quite unlikely unless $n$ is close to $N$. If $n$ is close to $N$, there's not much you can do; e.g. for $n=N$, either you accept some estimates above $N$ or you get a seriously biased estimator. As always in this area, "a better choice" depends on your criteria for an estimator. Certainly the maximum likelihood estimator is a possible choice that avoids the problem of estimates greater than $N$. – joriki Dec 6 '12 at 9:10
Thank you very much. I am also interested in a Bayesian approach but that is a separate question I know. – Anush Dec 6 '12 at 9:24
Can I ask what the estimators would be if you stop having just seen $k$ different buses? – Anush Dec 7 '12 at 20:01
@Anush: It depends on why you stop. Do you stop at a predetermined $k$? Or randomly, e.g. with a certain probability after each bus? – joriki Dec 7 '12 at 20:18

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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