2
$\begingroup$

Suppose, I am using Wald test to test following hypothesis. $$H0:\theta = \theta_0$$ $$H1:\theta \neq \theta_0$$

Given the MLE estimate $\hat{\theta}$ , Wald test makes the following normality assumption: $$\frac{\hat{\theta}-\theta_0}{\hat{se}}\rightsquigarrow \mathcal{N}(0,1)$$

Suppose $\theta = \theta^*$ is the true value of parameter, then for MLE estimate $\hat{\theta}$ we know: $$\frac{\hat{\theta}-\theta^*}{\hat{se}}\rightsquigarrow \mathcal{N}(0,1)$$

Why can we make the assumption of normality around $\theta_0$, when we know $\hat{\theta}$ is normal around $\theta^*$ and still trust the results?

$\endgroup$
  • $\begingroup$ You are misinterpreting the arrow symbol. It means that the quantity approaches the N(0,1) distribution as n tends to infinity. It is a central limit result. Now I will answer the question based on the proper interpretation. $\endgroup$ – Michael Chernick Nov 28 '16 at 14:59
1
$\begingroup$

The implied result requires assuming the null hypothesis is true. We need theta star to equal theta 0 and this will happen if the null hypothesis is true. Otherwise it will be a non-zero constant say m. In that case the asymptotic distribution will be N(m,1).

$\endgroup$
  • 1
    $\begingroup$ Both $\theta^*$ and $\theta_0$ are fixed values, which have nothing to do with sample size $n$. $\endgroup$ – user2329744 Nov 28 '16 at 15:57
  • $\begingroup$ Sorry I meant that theta star would have to equal theta 0 i.e. the true theta is the null value of theta. I will edit my post. $\endgroup$ – Michael Chernick Nov 28 '16 at 23:56
  • $\begingroup$ If you down voted my answer please note my correction. $\endgroup$ – Michael Chernick Nov 28 '16 at 23:58
  • $\begingroup$ Thanks for your answer. I didn't down vote it. $\endgroup$ – user2329744 Nov 29 '16 at 6:45

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.