MLE of Erlang Distribution

110 Views Asked by At

For sample $x_1,...,x_n$~Erlang, $f(x_i|k,\lambda)=\frac{\lambda^kx^{k-1}e^{-\lambda x_i}}{(k-1)!}$

Population has expectation $k/\lambda$ and variance $k/\lambda^2$. You may assume second derivative of log likelihood is negative

Suppose parameter k is known and can be treated as constant. Propose MLE for $\lambda$.

If I could get some help, specifically with the log-likelihood part of the problem and the derivative, that would be appreciated

I have got to $L=e^{-\lambda \sum x_i}\lambda^{kn}\Pi x_i^{k-1}(\frac{1}{(k-1)!})^n$

1

There are 1 best solutions below

0
On

Hints:

  • The next stage is $\log_e L = -\lambda \sum x_i + kn \log_e \lambda +\,$ something not involving $\lambda$

  • Then take the derivative with respect to $\lambda$

  • Then see for which values of $\lambda$ where the derivative is zero, positive and negative, and so where the log-likelihood is maximised

  • Which should tell you the maximum likelihood estimator for $\lambda$

It turns out that this will also be the naive estimator for the rate when observing that the sum of $kn$ iid exponential random variables is $\sum x_i$.