How can I prove that a random variable taking values in $[0,1]$ has variance no larger than $\frac{1}{4}$? If it matters, discrete and continuous proofs are both welcome.


Let $X$ a random variable with $0\leq X\leq 1$. We have $$\operatorname{Var}(X)=E[X^2]-E[X]^2\le E[X]-E[X]^2=E[X](1-E[X]).$$ Now, we note that for $0\leq t\leq 1$, we have $$t(1-t)=-(t^2-t)=-\left(\left(t-\frac 12\right)^2-\frac 14\right)=\frac 14-\left(t-\frac 12\right)^2\leq \frac 14,$$ with equality if $t=\frac 12$.

With a Bernoulli law ($P(X=1)=1/2$ and $P(X=0)=1/2$) you get an equality.

  • $\begingroup$ Why is $E[X^2] \leq E[X]$? (does that always apply or is it only in the range 0 to 1?) Oh I see, its because $0 \leq X \leq 1$. $\endgroup$ Jan 12 '14 at 8:25
  • 1
    $\begingroup$ We have indeed have $X^2\leqslant X$ because $0\leqslant X\leqslant 1$. $\endgroup$ Jan 12 '14 at 10:01

Simple calculus shows that the mean $\mu_X$ is the constant with smallest mean squared distance from $X$. Combined with the condition $0 \leq X\leq 1$, this gives $$\mbox{Var}(X)=\mathbb{E}[(X-\mu_X)^2]\leq \mathbb{E}\left[\left(X-{1\over 2}\right)^2\right]\leq {1\over 4}.$$


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.