3
$\begingroup$

There is already an answer here on the ML estimator for binomial p: Maximum likelihood estimation of p in a Binomial sample

Let me add a twist to the question: let's assume we don't know all the samples, but only count for a single value. So, for example, we have drawn M samples from a binomial distribution with known parameter N and unknown parameter p. The only information we know is that out of these M samples, m of them are equal to some value k.

So my question is: what is the ML estimator for p in Bin(N,p) if we know that m out of M samples are equal to k?

Improve this question
$\endgroup$
5
  • 1
    $\begingroup$ Just for clarity: the problem is that you (for concreteness) have two coins (A and B) with different probabilities of generating a head, and you don't know how many flips come from each coin. Your objective is to estimate the probability of generating a head from coin A? Does that seem about right? $\endgroup$
    – jbowman
    Commented Feb 8, 2018 at 17:20
  • 2
    $\begingroup$ This question definitely needs more clarity! Could you provide a more specific description of your model? What exactly is the relationship between the "1000" and "500 samples" you stipulate at the end? $\endgroup$
    – whuber ♦
    Commented Feb 8, 2018 at 17:22
  • $\begingroup$ Sorry for confusing you, I edited the question. I hope it is clearer now. $\endgroup$
    – Cindy Almighty
    Commented Feb 8, 2018 at 18:58
  • $\begingroup$ If I understand correctly, you can readily reduce this to a simpler problem. Since the chance that a Binomial$(N,p)$ observation equals $k$ is given by $q=\binom{N}{k}p^k(1-p)^{N-k}$, you have data that are equivalent to flipping a coin $M$ times, observing it came up heads $m$ times, and wish to estimate the chance $q$ of heads. From $q$ you can compute $p$. $\endgroup$
    – whuber ♦
    Commented Feb 8, 2018 at 22:49
  • $\begingroup$ Great idea! Can you offer it as an answer? $\endgroup$
    – Cindy Almighty
    Commented Feb 9, 2018 at 10:52

1 Answer 1

Reset to default
1
$\begingroup$

Partially answered in comments:

Since the chance that a Binomial$(N,p)$ observation equals $k$ is given by $q=\binom{N}{k}p^k(1-p)^{N-k}$, you have data that are equivalent to flipping a coin $M$ times, observing it came up heads $m$ times, and wish to estimate the chance $q$ of heads. From $q$ you can compute $p$.

  • whuber
Improve this answer
$\endgroup$
1
  • 1
    $\begingroup$ I was hoping to credit the author $\endgroup$
    – Cindy Almighty
    Commented Mar 20, 2021 at 12:01

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.