After I solve a quadratic equation which is the derivative of log-likelihood function to find MLE, I get one positive root and a negative root. Which one do I choose? Sample space is between $$0<y<1$$

  • 1
    $\begingroup$ Add the full question. $\endgroup$ – StubbornAtom Mar 16 '20 at 15:43

If your derivative is a quadratic, your log-likelihood is a cubic. Since you find two roots, your cubic has a local maximum and a local minimum. There are three ways to proceed:

  • the "zeroeth derivative test": evaluate the log-likelihood at both roots. Since you want the MLE, pick the root that gives a larger value.
  • the first derivative test: On the intervals of the sample space partitioned by the roots, evaluate the derivative to see in which intervals the log-likelihood is increasing/decreasing. A maximum occurs when the function turns (reading left-to-right) from increasing to decreasing.
  • the second derivative test: differentiate again and plug the roots into this second derivative. When the second derivative is positive, you have a local minimum. When the second derivative is negative you have a local maximum, which is the one you want to keep for your MLE.

Also, don't forget to evaluate the MLE at the endpoints of the sample space. The derivative of the log-likelihood cannot tell you about maxima at the boundaries of the sample spaces (since nothing interesting happens to the derivative at the boundary of the sample space).

  • $\begingroup$ Can the test statistic of L.R.T W(y) = -2(l(theta_hat) - l(theta_zero)) be negative? $\endgroup$ – moli Mar 16 '20 at 15:50
  • $\begingroup$ @moll : You seem to be asking another, different Question. You should post it as a new Question and give enough context to make sense of it. (I'm unable to guess what you mean from the far too short comment you posted.) $\endgroup$ – Eric Towers Mar 16 '20 at 15:52
  • $\begingroup$ can you take a look at this question please?math.stackexchange.com/questions/3584282/… $\endgroup$ – moli Mar 17 '20 at 15:20

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.