Let $X_1, \dots , X_n \sim \mathrm{N}(\mu,\sigma^2)$. $\sigma^2$ is known. We want to test $\mathrm{H}_0: \mu = 0$ versus $\mathrm{H}_1: \mu > 0$.

For the likelihood ratio I got: $\Lambda_1 = \exp(\frac{n(\mu_0^2 - \mu_1^2)}{2 \sigma^2}) \cdot \exp(\frac{\mu_1 - \mu_0}{\sigma^2} \cdot \sum x_i)$. Where the first term is a constant. Hope this is correct.

Now we know that for expected value of normal randowm sample $T(X_{1:n} = \sum X_i)$ is a sufficient statistic. I have to rewrite $\Lambda_1$ as a function of $T$, which gives me $\Lambda_2$. Can I now just exchange $\sum x_i$ in $\Lambda_1$ with $T$?

Another question is what I can say about the rejection region of $\Lambda_1$ and $\Lambda_2$, keeping in mind that $\mu$ is zero or bigger. I do not know what is meant here...


For your first question: Yes, you can substitute sufficient statistic as it is just a re-writing of $\sum x_i$

For your second part, since your data are normal, you can use the Wilks Likelihood ratio test to get a rejection region.

  • $\begingroup$ Thanks for the hints! Is it true, that $\Lambda_1$ and $\Lambda_2$ are basically the same and have the same rejection regions? I would say so, because the substitution above did not change anything, right? $\endgroup$
    – Michael
    Nov 5 '13 at 13:59
  • 1
    $\begingroup$ Not exactly...$\Lambda_1$ is in $\mathbb{R}^N$ while $\Lambda_2$ is in $\mathbb{R}$...which is one of the nice features of a sufficient statistic...dimension reduction. The two are related in that if we let $R_T$ be the rejection region for T, then the rejection region for $\Lambda_1 = \{x_i: \sum x_i \in R_T\}$ I hope this clarifies it...the two problems are in different dimensions (literally), but are equivalent in the sense I just pointed out. $\endgroup$
    – user31668
    Nov 5 '13 at 14:29

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.