0
$\begingroup$

Let $X_1, X_2, . . . , X_n$ be a random sample from a Bernoulli distribution with parameter $θ$. Compute the maximum likelihood estimator for $θ$.

In my opinion this is the correct way to solve it:
$L(\Theta)=\theta_1*(1-\theta_1)*...\theta_n*(1-\theta_n)=\theta^n*(1-\theta)^n$
$l(\Theta)=n*ln(\theta)+n*ln(1-\theta)$
$l'(\theta)=\frac{n}{\theta}-\frac{n}{1-\theta}=0$, $\theta=\frac{1}{2}$
For the official solution it should be $\frac{\sum_{i=1}^nX_i}{n}$, why MLE is incorrect, where should I put the $x$ that in my solution is not present?

$\endgroup$
  • 1
    $\begingroup$ bernoulli has no pdf, calculating max with derivatives is here not possible $\endgroup$ – user620174 Jan 25 at 14:52
  • 1
    $\begingroup$ Your likelihood formula is wrong - your formula doesn't depend on the data that you see, only $n$. When you use the correct formula you can differentiate the log likelihood to get the MLE. $\endgroup$ – Alex Jan 25 at 15:28
  • $\begingroup$ @Alex what do you mean? $\endgroup$ – Luke Marci Jan 25 at 15:31
  • 1
    $\begingroup$ The likelihood should depend on the data you see. E.g. suppose you just had $n = 1$ and observed $X_1 = x_1$. If that was a success ($x_1 = 1$) then the likelihood would be $\theta$, and if it was a failure ($x_1 = 0$) then the likelihood would be $(1 - \theta)$. In the first case the MLE would be $\hat{\theta} = 1$ and in the second case the MLE would be $\hat{\theta} = 0$. $\endgroup$ – Alex Jan 25 at 15:36
1
$\begingroup$

The likelihood function for Bernoulli can be written as follows: $$ L(\theta)=\begin{cases}\theta, & X_1=1\cr 1-\theta, & X_1=0\end{cases}\times \begin{cases}\theta, & X_2=1\cr 1-\theta, & X_2=0\end{cases}\times \cdots\times\begin{cases}\theta, & X_n=1\cr 1-\theta, & X_n=0\end{cases} $$ Note that $\sum_{i=1}^n X_i=n\overline X$ calculates the number of units in a sample, and $n-n\overline X$ calculates the number of zeros. In the above product, $\theta$ is multiplied as many times as there are units in the sample. And $1-\theta$ is multiplied as many times as there are zeros in the sample. So $$ L(\theta) = \theta^{n\overline X}\cdot (1-\theta)^{n-n\overline X}. $$ Starting from here you can calculate MLE with derivatives.

$\endgroup$

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.