The two tests below seem to be the same to me. But when I tried an example and constructed the test statistics I couldn't see how they are the same.

$H_0: \lambda=1~~~~H_a: \lambda\not=1$

$H_0: \ln(\lambda)=0~~~~H_a: \ln(\lambda)\not=0$

Use exponential distribution with form $\lambda$=mean

Then the MLE of the mean is sample average $\bar X$, with asymptotic variance $\lambda^2$

Consider the transformation $\ln(\lambda)$, then we get the MLE is $\ln\bar X$ with asymptotic variance 1.

So the test statistics should be:

$\sqrt{n}\frac{\bar X-1}{\bar X}\to Z$

$\sqrt{n}\ln(\bar X)\to Z$

So did I make an error (I asked somebody else and they said it looked fine) or do these tests just converge to the same thing while looking different the whole time?

Edit: I should be using Wald test here instead I think.

  • $\begingroup$ "... variance $\bar{X}^2/n$" ?? $\endgroup$ – Dilip Sarwate Jul 28 '16 at 20:07
  • $\begingroup$ @Dilip Ya that was a type sorry. $\endgroup$ – VCG Jul 28 '16 at 20:12

From Wooldridge's cross section book, there is a problem where a non-linear transformation changes the hypothesis outcome, so the two are not equivalent.

| cite | improve this answer | |

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.