Proof of a generalized Chebyshev's inequality

326 Views Asked by At

In the probability textbook that I'm reading through right now ("Knowing the Odds, an Introduction to Probability" by John B. Walsh), one of the textbook questions asks us to prove a generalized form of Chebyshev's inequality:

Let $\Phi$ be a strictly positive increasing function, $X$ a random variable, and $\lambda>0$. Prove that: $$\mathbb{P}(\Phi(X)\geq\lambda)\leq\frac{1}{\Phi(\lambda)}\mathbb{E}(\Phi(X))$$

My steps are as follows: we know that $\mathbb{P}(\Phi(X)\geq\lambda) = \mathbb{E}(1_{\Phi(X)\geq\lambda})$. Writing out this latter form as an integral, we have: $$\mathbb{E}(1_{\Phi(X)\geq\lambda}) = \int_{\Phi(X)\geq\lambda}1d\mathbb{P}$$ But at the same time, we can recognize that on the set $\{\Phi(X)\geq\lambda\}$, we have that $\frac{\Phi(X)}{\lambda}\geq 1$. So: $$\int_{\Phi(X)\geq\lambda}1d\mathbb{P}\leq \int_{\Phi(X)\geq\lambda}\frac{\Phi(X)}{\lambda}d\mathbb{P}\leq \int_\Omega\frac{\Phi(X)}{\lambda}d\mathbb{P} = \frac{1}{\lambda}\int_\Omega\Phi(X)d\mathbb{P}$$ What I don't know how to do is turn that $\lambda$ into a $\Phi(\lambda)$. What should my next step be? Any hints would be appreciated. Thanks!

EDIT: it turns out that my textbook has a typo in it. The true generalized Chebyshev's inequality is given by: $$\mathbb{P}(X\geq\lambda) \leq \frac{1}{\Phi(\lambda)}\mathbb{E}(\Phi(X))$$ This follows from a similar line of reasoning as above, and by recognizing that $\frac{\Phi(X)}{\Phi(\lambda)}$, since $\Phi(\cdot)$ is an increasing, positive function.

1

There are 1 best solutions below

3
On

As an alternative proof, the generalized Chebyshev inequality follows from the basic Chebyshev inequality $$P(|X|\ge\lambda)\le\frac{E|X|^p}{\lambda^p},$$ as follows. Argue that the event $\{X\ge\lambda\}$ is the same as the event $\{\Phi(X)\ge \Phi(\lambda)\}$. Now apply the basic Chebyshev inequality with $p=1$ (aka Markov's inequality) to this last event.