I calculated that $$ Var(\theta_1)=\frac{2\sigma^2}{(x_n-x_1)^2},~~~~~Var(\theta_2)=\frac{\sigma^2}{\sum_{i=1}^{n}(x_i-\overline{x})^2}. $$ Now I have to show that for $\sigma^2 >0$ and $\overline{x}\neq\frac{(x_1+x_n)}{2}$ it is $$ Var(\theta_1) > Var(\theta_2). $$

I have to show that it is $$ \frac{2\sigma^2}{(x_n-x_1)^2} > \frac{\sigma^2}{\sum_{i=1}^{n}(x_i-\overline{x})^2}. $$

Do you have an idea how to show that?

Unfortunately, I have not.


I've already tried to show that $$ Var(\theta_2)/Var(\theta_1)<1, $$ i.e. that $$ \frac{(x_n-x_1)^2}{\sum_{i=1}^{n}(x_i-\overline{x})^2}=\frac{x_1^2+x_n^2-2x_1x_n}{\sum_{i=1}^{n}x_i^2-n\overline{x}^2}<1 $$ but with no success.

  • $\begingroup$ You could look at the ratio and show that it is $ >1$. $\endgroup$ – Hans Engler May 14 '14 at 22:17

We wish to show that

$$ \frac{2\sigma^2}{(x_1 - x_n)^2} \geq \frac{\sigma^2}{\sum_{i=1}^n (x_i - \bar{x})^2} $$

Cancelling the sigmas and inverting the fractions shows that the desired result is

$$ \sum_{i=1}^n (x_i - \bar{x})^2 \geq \frac{1}{2}(x_1 - x_n)^2.$$ Now, in order to avoid typing $\bar{x}$ all over the place, let's define $y_i = x_i - \bar{x}$. We then have $y_1 - y_n = x_1 - x_n$, and the desired result is \begin{align*} \sum_{i=1}^n y_i^2 &\geq \frac{1}{2}(y_1 - y_n)^2\\ &= \frac{1}{2}(y_1^2 + y_n^2 - 2y_1y_n). \end{align*} Subtracting $y_1^2$ and $y_n^2$ from both sides yields that we want to show \begin{align*} \sum_{i=2}^{n-1} y_i^2 &\geq -\frac{1}{2}(y_1^2 + y_n^2 + 2y_1y_n)\\ &= -\frac{(y_1+y_n)^2}{2}, \end{align*} which is trivially true as the left hand side is a sum of non-negative terms, and must therefore be greater than zero.

For $\bar{x} \neq \frac{1}{2}(x_1 + x_n)$ we see that the right hand side is negative, and the inequality is strict.

  • 1
    $\begingroup$ Nice, +1. $ $ $ $ $\endgroup$ – Did May 19 '14 at 15:42

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.