Let $X_{11},...,X_{1n}\sim \mathcal{N}(\mu_1,\sigma^2)$ and $X_{21},...,X_{2n}\sim \mathcal{n}(\mu_2,\sigma^2)$ be iid distributed and such that $\{X_{i1}\}$ are independent from $\{X_{i2}\}$ and $\sigma^2$ is known. Now we consider the region $$R=\{(\mu_1,\mu_2): \frac{n}{\sigma^2}\left((\bar X_1-\mu_1)^2+(\bar X_2-\mu_2)^2\right)\leq Q\}$$ I want to find $Q$ for which $R$ is a confidence region with level $1-\alpha$. ($\bar X_1=\frac{1}{n}\sum_{k=1}^n X_{1k}$)

I know that I need to show that $\Bbb{P}_{(\mu_1,\mu_2)} ((\mu_1,\mu_2)\in R)=1-\alpha$. therefore I first tried to compute the probability: $$\begin{align}\Bbb{P}_{(\mu_1,\mu_2)} ((\mu_1,\mu_2)\in R)&=\Bbb{P}_{(\mu_1,\mu_2)}\left(\frac{n}{\sigma^2}\left((\bar X_1-\mu_1)^2+(\bar X_2-\mu_2)^2\right)\leq Q\right)\\&=\Bbb{P}_{(\mu_1,\mu_2)}\left(\frac{\bar X_1-\mu_1}{\sigma/\sqrt{n}}+\frac{\bar X_2-\mu_2}{\sigma/\sqrt{n}}\leq \sqrt{Q}\right)\\&=\Bbb{P}_{(\mu_1,\mu_2)}\left(Z\leq \sqrt{Q}\right)\\&=\Bbb{P}_{(\mu_1,\mu_2)}\left(\frac{Z}{\sqrt{2}}\leq \sqrt{\frac{Q}{2}}\right)\\&=\Phi\left(\sqrt{\frac{Q}{2}}\right)\end{align}$$ where I used that $\frac{\bar X_i-\mu_1}{\sigma/\sqrt{n}}\sim\mathcal{N}(0,1)$ and the sum of two $\mathcal{N}(0,1)$ random variables has distribution $\mathcal{N}(0,2)$ so I took $Z\sim \mathcal{N}(0,2)$ and additionally I used in the last step that $\frac{Z}{\sqrt 2}\sim \mathcal{N}(0,1)$. So I only need to solve $1-\alpha=\Phi\left(\sqrt{\frac{Q}{2}}\right)$ which is equivalent to say that $\sqrt{\frac{Q}{2}}=z_{1-\alpha}$ where $z_{1-\alpha}$ is the $1-\alpha$ quantile of a normal distribution. But then I would have said that $Q=2(z_{1-\alpha})^2$.

Now somehow our prof. told us that $Q=\chi^2_{2,1-\alpha}$ where $\chi^2_{2,1-\alpha}$ is the $1-\alpha$ quantile of a $\chi^2_{2}$ distribution. I don't see how to get from my solution to his solution. Am I doing something wrong?

Thanks for your help.


1 Answer 1


Note, that for any standard normal variable $$Z \sim N(0,1) $$ it holds that $$ Z^2 \sim \chi_1^2$$. I suppose this is what you need here.

  • $\begingroup$ Thanks for your answer. Yes I know this, but in the first attempt I would not have seen it. I would have done it as above and I would like to know where my error is or how do I can conclude. $\endgroup$ Commented Jun 15, 2023 at 13:56
  • $\begingroup$ +1. This exactly responds to the question. $\endgroup$
    – whuber
    Commented Jun 15, 2023 at 14:02
  • $\begingroup$ @whuber can I say that since $Z^2\sim \chi^2_1$ then also $z_{1-\alpha}^2=\chi^2_{1,1-\alpha}$ and since I got $2z_{1-\alpha}^2=2\chi^2_{1,1-\alpha}=\chi^2_{1,1-\alpha}+\chi^2_{1,1-\alpha}=\chi^2_{2,1-\alpha}$ since sums of $\chi^2$ distributions is again a $\chi^2$ distribution with degree of freedom to be the sum of the degrees of freedom? $\endgroup$ Commented Jun 15, 2023 at 14:03
  • $\begingroup$ @user1294729 I think that the squared quantile does not match with the chisq quantiles. As far as I see it, you took the root of sum of squares which should not result in a simple sum. $\endgroup$ Commented Jun 15, 2023 at 14:08
  • 1
    $\begingroup$ @user1294729 the comment from whuber technically solves your question. The sum of $\chi_1^2$ variables is again $\chi_n^2$ distributed. Equipped with that you should take a look at your computations again. $\endgroup$ Commented Jun 15, 2023 at 14:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.