2
$\begingroup$

This is an order statistics question. I'm using the notation found on https://en.wikipedia.org/wiki/Order_statistic.

We have $n$ IID random variables $X_1, \cdots, X_n$ that are uniformly distributed on $[0, 1]$. $X_{(1)} = \min(X_1, \ldots, X_n)$ and $X_{(n)} = \max(X_1, \ldots, X_n)$.

I am asked to reason that $\operatorname{Var}(X_{(1)}) = \operatorname{Var}(X_{(n)})$ without resorting to calculations. So it seems some kind of intuitive answer is wanted.

I'm given the hint that $\operatorname{Var}(x) =\operatorname{Var}(1-x)$ for any random variable $x$, but it's not obvious to me how this is helpful.

I know that $$ \operatorname{Var}(X_{(1)}) = E[X_{(1)}^2] - E[X_{(1)}]^2 \\ \operatorname{Var}(X_{(n)}) = E[X_{(n)}^2] - E[X_{(n)}]^2 $$

We known that $E[X_1] = \cdots = E[X_n] = 0.5$.

Intuition tells me that $E[X_{(1)}]$ and $E[X_{(n)}]$ are symmetrically situated about 0.5, i.e, $$ 0.5 - E[X_{(1)}] = E[X_{(n)}] - 0.5 \\ 1 = E[X_{(1)}] + E[X_{(n)}] $$ Not sure if this helps. I still can't figure out how $\operatorname{Var}(x) = \operatorname{Var}(1-x)$ is helpful. Am I going in the right correction with this expected value reasoning?

Honestly, it's kind of intuitive to me that $\operatorname{Var}(X_{(1)})=\operatorname{Var}(X_{(n)})$ by symmetry, but I'm having a hard time putting it into words while incorporating the hint.

$\endgroup$
1
  • $\begingroup$ This also follows from the fact that $X_i-\frac12$ has the same distribution as $\frac12-X_i$ (since the distribution is symmetric about $1/2$), so that $X_{(n)}-\frac12$ has the same distribution as $\frac12-X_{(1)}$. $\endgroup$ – StubbornAtom May 6 '20 at 21:29
1
$\begingroup$

Let $Y_k=1-X_k$. $Y_k$ has the same distribution as $X_k$. (define $Y_{(1)}$ and $Y_{(n)}$ using the same min and max notation).

Therefore $var(Y_{(1)})=var(X_{(1)})$. Meanwhile $Y_{(1)}=X_{(n)}$ and $Y_{(n)}=X_{(1)}$

As a result $var(X_{(n)})=var(X_{(1)})=var(Y_{(n)})=var(Y_{(1)})$

$\endgroup$
14
  • $\begingroup$ I like this proof. This relies on us knowing that $Y_{(1)} = X_{(n)}$ and $Y_{(n)} = X_{(1)}$, which further relies on us knowing that $E[X_{(1)}] - 0 = 1- E[X_{(n)}]$, the symmetry argument I made earlier. Is this obvious by inspection, or does this need to be proven some how? $\endgroup$ – David May 6 '20 at 12:29
  • $\begingroup$ $Y_{(1)}$ is the minimum of {$Y_k=1-X_k$} which corresponds to the maximum of {$X_k$} which is $X_{(n)}$. Essentially changing the sign of all $X_k$ switches max and min. $\endgroup$ – herb steinberg May 6 '20 at 17:31
  • $\begingroup$ Isn't $Y_{(1)}=1-X_{(n)}$ and $Y_{(n)}=1-X_{(1)}$? $\endgroup$ – StubbornAtom May 6 '20 at 21:28
  • 1
    $\begingroup$ @StubbornAtom You are correct. The variance conclusion still holds, since var(x)=var(1-x) for any random variable x. $\endgroup$ – herb steinberg May 6 '20 at 21:45
  • $\begingroup$ @herbsteinberg Isn't $Y_{(1)} = X_{(1)}$? $\endgroup$ – David May 6 '20 at 23:52

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.