# Confidence interval for a sample from the norm dist

Let $Y_1<Y_2$ be order statistics from a random sample of size $2$ from a normal distribution, $\mathcal{N}(\mu,\sigma^2)$, where $\sigma^2$ is known. If $\bar{X}$ is the mean of this sample, how would I find the constant $c$ which solves this equation: $P(\bar{X} - c\sigma < \mu < \bar{X} + c\sigma)$. Thanks for the help.

-
The order statistics don't even matter in this case because you are considering their mean. – Emre Nov 14 '11 at 22:02
Any progress on this? – Did Nov 16 '11 at 15:19

The expression $P(\bar{X} - c\sigma < \mu < \bar{X} + c\sigma)$ is not actually an equation until you put "$=\text{something}$" after it. Presumably you want it to be equal to some number between $0$ and $1$. Let's call that number $p$.
Notice that $\bar{X} = (Y_1+Y_2)/2$, and $\bar{X} \sim \mathcal{N}(\mu, \sigma^2/2)$. So $$\frac{\bar{X} - \mu}{\sigma/\sqrt{2}} \sim \mathcal{N}(0,1).$$ Now find $c$ such that $P(-c<Z<c)= p$ if $Z\sim\mathcal{N}(0,1)$. You get this from a table or from software unless you want to go through a moderately elaborate discussion of how to do it numerically.
Then notice that $$-c < \frac{\bar{X}-\mu}{\sigma/\sqrt{2}} < c \text{ if and only if }\bar{X}-c\frac{\sigma}{\sqrt{2}} < \mu < \bar{X}+c\frac{\sigma}{\sqrt{2}}.$$
The value of $c$ for which $\Pr(-c<Z<c)=0.5$ is the same as the value for which $\Pr(Z<c)=0.75$, and if I can trust the software package I'm using, that is 0.7733726. – Michael Hardy Nov 15 '11 at 17:58