Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Let $X$ be a random variable with finite expected value $μ$ and non-zero variance $σ^2$. Then for any real number $k > 0$, two-sided Chebyshev inequality states $$ \Pr(|X-\mu|\geq k\sigma) \leq \frac{1}{k^2}. $$

  1. I saw a paper applies two-sided Chebyshev inequality to $\Pr(|X|\geq b)$ and got an upper bound $Var(X)/b^2$, when $X$ does not necessarily have mean zero. I think it is not correct. Is there any way to apply two-sided Chebyshev inequality to this case?
  2. Instead, for $\Pr(|X-a|\geq b)$, I think it is only possible to apply one-sided Chebyshev inequality to $\Pr(X-a\geq b)$ and $\Pr(X-a \leq -b)$ respectively. Am I right?

Thanks and regards!

share|improve this question
add comment

2 Answers 2

up vote 1 down vote accepted

Edited in response to OP's comment and request

It depends on what you mean by applying the two-sided Chebyshev Inequality. The same method that is used to prove the Chebyshev Inequality (viz. bound $E[\mathbf 1_{(-\infty, \mu-b]} + \mathbf 1_{[\mu+b,\infty)}]$ from above by $E[(X-\mu)^2/b^2] = \sigma^2/b^2$ can be used to show that $$P\{|X-a| \geq b\} \leq \frac{E[(X-a)^2]}{b^2} = \frac{\sigma^2 + (\mu-a)^2}{b^2}.$$ In more detail, $\displaystyle \mathbf 1_{(-\infty, a-b]} + \mathbf 1_{[a+b,\infty)} \leq \left(\frac{x-a}{b}\right)^2$ for all $x \in \mathbb R$ since the parabola $\displaystyle \left(\frac{x-a}{b}\right)^2$ passes through $(a-b,1), (a,0)$, and $(a+b,1)$. Consequently, $$\begin{align*} P\{|X-a| \geq b\} &= P\{X \leq a-b\} + P\{X \geq a + b\}\\ &= \int_{-\infty}^{a-b} f_X(x)\,\mathrm dx + \int_{a+b}^{\infty} f_X(x)\,\mathrm dx\\ &= \int_{-\infty}^{\infty}(\mathbf 1_{(-\infty, a-b]} + \mathbf 1_{[a+b,\infty)})f_X(x)\,\mathrm dx\\ &\leq \int_{-\infty}^{\infty}\left(\frac{x-a}{b}\right)^2f_X(x)\,\mathrm dx\\ &= \frac{1}{b^2}\int_{-\infty}^{\infty}(x-a)^2f_X(x)\,\mathrm dx\\ &= \frac{E[(X-a)^2]}{b^2} \end{align*}$$ If $a$ equals the mean $\mu$, that expectation on the right is the variance $\sigma^2$ and we get Chebyshev's Inequality. More generally, it is a standard result in probability theory that $$\begin{align*} E[(X-a)^2] &= E[((X-\mu) + (\mu - a))^2\\ &= E[(X-\mu)^2] + (\mu-a)^2 + 2(\mu-a)E[X-\mu]\\ &= \sigma^2 + (\mu-a)^2 \end{align*}$$ and we get the inequality I stated initially.

With regard to the second question in the OP's comment,

I saw a paper applies two-sided Chebyshev inequality to $\Pr(|X|\geq b)$ and got an upper bound $Var(X)/b^2$, when $X$ does not necessarily have mean zero. I think it is not correct.

Yes, that is not right when the mean is not zero; the upper bound from the variation on the Chebyshev Inequality described above is $(\sigma^2+\mu^2)/b^2$

share|improve this answer
    
Thanks! (1) I wonder how to prove the inequality in your reply using the same method for proving Chebyshev inequality? (2) Are you saying the paper is right, and I am wrong about it? –  Tim May 13 '12 at 19:02
add comment

There can exist no such non-trivial upper bound since, for every fixed $b$, $$\sup\limits_{a\in\mathbb R}\mathrm P(|X-a|\geqslant b)=1.$$

share|improve this answer
    
Thanks! Is the example easy to construct? –  Tim May 13 '12 at 18:42
    
Every random variable $X$ and real number $b$ do the job. –  Did May 13 '12 at 18:45
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.