0
$\begingroup$

I am trying to figure out what the CDF and PDF are of a log likelihood ratio test. My research online has led me to believe that they follow a $\chi^2$ distribution according to Wilks:

Given an observed sequence $Y_n$ which is IID according to either hypothesis $H_0$ & $H_1$, where:

$\Lambda = log\frac{f_{Y|H_1}(Y)}{f_{Y|H_0}(Y)} $

is the likelihood ratio. Is the distribution (PDF and CDF) of $\Lambda$ $\chi^2$? How do I find out the degrees of freedom?

$\endgroup$
0
$\begingroup$
  1. The distribution is $\chi^2$ only asymptotically and only if $H_0$ is true, finite sample distribution generally depends on distribution of $\{Y_n\}_{n=1}^{N}$.
  2. The number of degrees of freedom depends on hypothesis. For simple $H_0: \theta = \theta_0$, it is the number of dimensions in $\theta$, for general $g(\theta) = 0$ it should be like the rank of Jacobian of $g(\theta)$.
$\endgroup$

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.