3
$\begingroup$

Let $X_1,X_2,...;X_n$ be a random sample with $X_i$~$Binomial(m,p)$ for $i=1,...,n$ and $m=1,2,3,...$ and let $p\in (0,1)$. We assume $m$ is known and we are given the following data $x_1,...,x_n\in\{0,...,m\}$

Write up the log-likelihood function and find the MLE $\hat{P}ML$ for p

I'm not quite sure how to approach this. This is what I've tried:


I believe the likelihood function of a Binomial trial is given by

$P_{X_i}(x;m)=$ ${m}\choose{x} $$p^x(1-p)^{m-x}$

From here I'm kind of stuck. I'm uncertain how I find/calculate the log likelihood function.

I've understood the MLE as being taking the derivative with respect to m, setting the equation equal to zero and isolating m (like with most maximization problems). So finding the log likelihood function seems to be my problem

Edit: I might be misunderstanding it but could the log likelihood function simple be log of the likelihood function? so $log(P_{X_i}(x;m))$

$\endgroup$
3
$\begingroup$

The likelihood is

$$L(p)=\prod_{i=1}^nP_p(X=x_i)=\prod_{i=1}^{n}{m\choose x_i}p^{x_i}(1-p)^{m-x_i}$$

The log-likelihood is thus

$$\log L(p)=\log\left(\prod_{i=1}^{n}{m\choose x_i}\right)+\log(p)\sum_{i=1}^nx_i+\log(1-p)\left(nm-\sum_{i=1}^nx_i\right)$$

Let $M=\log\left(\prod_{i=1}^{n}{m\choose x_i}\right)$ (which does not depend on $p$):

$$\log L(p)=M+\log(p)\sum_{i=1}^nx_i+\log(1-p)\left(nm-\sum_{i=1}^nx_i\right)$$

$$\log L(p)=M+n\log(p)\bar x+n\log(1-p)(m-\bar x)$$

$$\dfrac{\partial\log L}{\partial p}=\frac{n\bar x}{p}-\frac{n(m-\bar x)}{1-p}$$

This last expression is zero if

$$\frac{\bar x}{p}=\frac{m-\bar x}{1-p}$$

$$\bar x-\bar xp=mp-\bar xp$$

Hence the ML estimator is:

$$\hat p=\dfrac{\bar x}{m}$$

$\endgroup$
  • $\begingroup$ Thank you for the thorough explanation. You would'nt happen to know of a well documented page online that could explain a similar general example for other widely used distributions? My textbook does not really provide them. $\endgroup$ – Sirmimer May 23 '18 at 21:22
1
$\begingroup$

I think you are correct, but usually you take the product of P_X over all observations X_1,...,X_n to compute the Likelihood of a sample. This translates to the sum of log probabilities.

I think you are interested in finding the parameter p that maximises the (Log)Likelihood rather than the known quantity m.

$\endgroup$

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.