2
$\begingroup$

I have generated $n=1000$ samples according to a chi-squared distribution $\chi^2(k)$ using a fixed number of degrees of freedom $k=2$ with MATLAB (chi2rnd function). I just would like to estimate the number of degrees of freedom using maximum likelihood. When calculating the log-likelihood, one should ends up with the following expression:

$$ \begin{align} log(L(k)) & = \log{\left(\prod_{i=1}^n f(x_i;k)\right)}\\ & = \sum_{i=1}^n \log{\left(\frac{x_i^{k/2-1}e^{-x_i/2}}{\Gamma(k/2)2^{k/2}}\right)}\\ & = \left(\frac{k}{2}-1\right)\sum_{i=1}^n \log{x_i} - \frac{1}{2}\sum_{i=1}^n x_i - \log{(\Gamma(k/2))} - \frac{k}{2}\log{(2)} \end{align} $$

However, when I try to plot the resulting expression as a function of the number of degrees of freedom between $k=0.01$ and $k=10$, I have a linear function that does not reach any maximum:

enter image description here

I simply do not understand why I obtain such a result. I wonder if the error comes from the MATLAB program of if the maximum likelihood approach is not applicable for such a PDF. However, when considering that samples follow a Gamma distribution (with appropriate parameters by fixing one of them), it works as expected. Does anyone know what's wrong?

$\endgroup$
3
$\begingroup$

You probably forgot to multiply last two parts by $n$, so it should be

$$\big(\frac{k}{2}-1\big)\sum_{i=1}^n\log{x_i} - \frac{1}{2}\sum_{i=1}^nx_i - n\log(\Gamma(k/2)) - \frac{nk}{2}\log(2)$$

enter image description here

$\endgroup$

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.