If $X_1, \ldots, X_n$ are independent but not necessarily identically distributed random variables, is it always true $\mathbb{P}\{X_n > \max\{X_1, \ldots, X_{n-1}\}\} = \mathbb{P}\{X_n > X_1, \ldots, X_n > X_{n-1}\} = \mathbb{P}\{X_n > X_1\}\cdots \mathbb{P}\{X_n > X_{n-1}\}$?


No you need an integral. The probability is equivalent to:

$$P\left(\bigcap_{i\neq n} X_n>X_i\right)=\int P(\cap_{i\neq n} X_n>X_i|X_n=x_n)dP_{X_n}(x_n)=\int\prod_{i\neq n}P(X_n>X_i|X_n=x_n)dP_{X_n}(x_n).$$

  • $\begingroup$ Is there an intuition behind this? In that we need to consider the joint distribution between the events and $X_n$? $\endgroup$ – user321627 Feb 12 '18 at 23:27
  • $\begingroup$ Sorry, I dropped an $|X_n$. Makes sense now? $\endgroup$ – Alex R. Feb 12 '18 at 23:29
  • $\begingroup$ Thanks, I see it now! $\endgroup$ – user321627 Feb 12 '18 at 23:30
  • 1
    $\begingroup$ Please take care to avoid the confusions between the random variable $X_n$ and the real argument $x_n$ of the integrals on $\mathbb R$. $\endgroup$ – Did Feb 12 '18 at 23:38


Let $X_1,X_2,X_3$ be independent with $X_3$ uniform on $\{0,1\}$ and $X_1,X_2$ constant, equal to $0$ w.p. 1. Then, $\mathbb{P}\{X_3>X_1\}\cap \{X_3>X_2\} = \mathbb{P}\{X_3=1\} = \frac{1}{2}.$

But $\mathbb{P}\{X_3>X_1\}\cdot \mathbb{P} \{X_3>X_2\} = \mathbb{P}\{X_3=1\}^2 = \frac{1}{4}$.

The idea is that while the $X_i$'s are independent, the events $(\{X_n>X_i\})_{i\geq 2}$ may not be.

NB: This can easily be adapted to have $X_1,X_2$ non-constant: take $X_3$ uniform on $\{0,2\}$ and $X_1,X_2$ uniform on $\{0,1\}$.


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.