3
$\begingroup$

How does one prove probability integral transform? So when $Y = F_X(X)$ where $X$ has a continuous distribution for which the cumulative distribution function is $F_X$, why does $Y$ have a uniform distribution? And what would Y's relationship with $\text{uniform}(0,1)$ distribution?

$\endgroup$
  • $\begingroup$ Why minus vote? $\endgroup$ – rrr Mar 8 '13 at 10:16
  • 3
    $\begingroup$ See the page "How to ask a question" and enumerate the ways in which yours differ. $\endgroup$ – Did Mar 9 '13 at 9:01
5
$\begingroup$

Write for $t \in (0;1)$, $$ P(Y \leq t) = P(F_X(X) \leq t) = P(X \leq F_X^{-1}(t)) = F_X(F_X^{-1}(t)) = t. $$ It is the cdf of the U(0;1).

$\endgroup$
  • 1
    $\begingroup$ Worth mentioning that this is also just an inverse sample transform to a standard uniform distribution. $\endgroup$ – Emily Mar 8 '13 at 21:13
  • $\begingroup$ What if X is not invertible? $\endgroup$ – Peter Jul 4 '13 at 7:15
  • 3
    $\begingroup$ @Peter, we are not inverting $X$ but instead $F_{X}$. Certainly it could be that this function is not invertible when we define the inverse function in the usual way for functions. But here we typically define the inverse of a CDF in a different way to guarantee that an inverse always exists, namely, for a CDF we usually define $F^{-1}_{X}(p) = \inf \{x|P(X\leq x) \geq p)$. $\endgroup$ – Addem Jan 4 '15 at 15:26
  • $\begingroup$ From the answer it is not clear why $F_X$ must be continuous. $\endgroup$ – Sergey Zykov Jul 27 '15 at 15:26
  • $\begingroup$ @Addem How can we be sure that $F_x^{-1}(p)$, in this particular definition, is always a real number? Can't it be infinite? $\endgroup$ – Amontillado Aug 13 '16 at 17:25

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.