1
$\begingroup$

Can the log likelihood ratio for a simple vs simple hypothesis take a negative value? Since it approximately follows a chi-square distribution for large sample sizes and since a chi-square distribution takes only positive values do we reject or accept the null hypothesis if a negative value is obtained for the log likelihood ratio test under a particular simple vs simple hypothesis?

The log likelihood ratio test statistic in this case is $$-2Log\Lambda$$

$\endgroup$
  • 1
    $\begingroup$ Please fix your question; in this state it makes no sense. $\endgroup$ – user88 Oct 16 '12 at 15:41
  • $\begingroup$ well I've tried to fix it .. hope it makes sense in this form :) $\endgroup$ – Nuzhi Oct 16 '12 at 16:16
4
$\begingroup$

Yes, the log likelihood ratio must take on both positive and negative values. If $\log \frac{f_1(x)}{f_0(x)}$ were positive for all possible values of the observed data $x$, then it would be true that $f_1(x) > f_0(x)$ for all $x$, and this is impossible. Remember that $f_1(x)$ and $f_0(x)$ are the probability density functions (or probability mass functions) of the observations under the two hypotheses, and if $f_1(x) > f_0(x)$ were to hold for all $x$, then it would be true that $$\int_{-\infty}^{\infty} f_1(x)\,\mathrm dx > \int_{-\infty}^{\infty} f_0(x)\,\mathrm dx$$ in contradiction of the fact that the integrals displayed above must both equal $1$. A similar argument applies (with sums instead of integrals) for probability mass functions.

$\endgroup$

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.