I was looking at my stats textbook and they claim that the sample variance of a weighted distribution involving i.i.d. $x_i$s will be smallest when each of the weights is equal. I follow this argument up to the point where I reach

$\sum_{i=1}^n$$w_i^2\geq\frac{1}{n}$ given $\sum_{i=1}^n w_i=1$

(this result obtained due to the fact that $Var(\bar{x_w})= \sigma^2\sum_{i=1}^nw_i^2$ and $Var(\bar{x})=\frac{\sigma^2}{n}$, so setting them equal and cancelling the $\sigma^2$ on each side yields that inequality, where $\bar{x_w}$ is the weighted average). Trying out a few examples, it seems pretty obvious that the inequality holds, but the book offers no mathematical justification and I was hoping someone here could help put it more concretely - I'm not sure how to approach it myself.

Any thoughts?


Use Cauchy-Schwarz: $$(\sum_{i=1}^n w_i \cdot 1)^2 \le (\sum_{i=1}^n w_i^2)(\sum_{i=1}^n 1^2).$$

| cite | improve this answer | |
  • $\begingroup$ Sorry, I'm sure I'm missing the obvious here, but I don't see how this helps? $\endgroup$ – user1257768 Jan 31 '13 at 22:38
  • $\begingroup$ The left sum is $\sum w_i = 1$ and the sum of $1$s is $n$. Divide both sides by $n$ to get the statement. $\endgroup$ – Cocopuffs Jan 31 '13 at 22:40
  • $\begingroup$ Ooh, of course! Elegant and precise - thank you very much. $\endgroup$ – user1257768 Jan 31 '13 at 22:42

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.