I Finding the pdf of a transformed univariate random variable

  • I
  • Thread starter Thread starter Hamiltonian
  • Start date Start date
  • Tags Tags
    Cdf Integral Pdf
AI Thread Summary
The discussion focuses on finding the probability density function (pdf) of a transformed univariate random variable using the cumulative distribution function (cdf) approach. It clarifies that the integral should be evaluated with respect to the original variable's pdf, f_X(x), rather than directly using the transformed variable. The relationship between the transformed variable Y and the original variable X is established through the cdf, P(Y<y), which can be derived by manipulating the cdf of X. An example involving the chi-squared distribution is provided, illustrating the process of obtaining the pdf from the cdf. The conversation emphasizes the importance of correctly integrating and differentiating to derive the desired pdf.
Hamiltonian
Messages
296
Reaction score
193
TL;DR Summary
Confused as to how to obtain the cdf of a transformed random variable.
1701139076070.png

The above theorem is trying to find the pdf of a transformed random variable, it attempts to do so by "first principles", starting by using the definition of cdf, I don't understand why they have a ##f_X(x)## in the integral wouldn't ##\int_{\{x:r(x)<y\}}r(X) dx## be the correct integral for the cdf of Y.
 
Physics news on Phys.org
No. The way ##r(x)## gets into the equation is only in the limit of the integral, ##r(x) \lt y##. Suppose ##r(x) \lt y##. What is the probability (density) of that? It is the probability (density) of that associated ##x## value, which is ##f_X(x)##. So those are the probabilities that we want to total for the integral. (Notice that the integral is with respect to ##dx##, not ##dy##)

PS. I may have ignored the situation where multiple ##x## values give the same ##r(x)##. That still works out because the integral limit allows the associated ##x## densities to be summed
 
Last edited:
Suppose ##Y=f(X)## and we know the pdf of X. To get the pdf of Y, we can find the CDF of Y ##P(Y<y)## then differentiate it wrt to y to get the pdf.
$$P(Y<y)=P(f(X)<y)$$
And depending on X you have to do an appropriate manipulation to get the cdf of Y. Here's and example, deriving the pdf of a ##\chi^2## function with a deg of freedom of one. This variable here is just ##Z^2##, Z is a standard normal distribution. The pdf of Z is a bit complex but you can find it here.
https://www.thoughtco.com/normal-distribution-bell-curve-formula-3126278 lets call this function f(x).
And cdf of f is ##\int_{-\infty}^{x}f(x) dx## and called F(x). Let X be a standard normal variable, and ##\chi=X^2##. So
$$P(\chi<y)=P(-\sqrt{y}<X<\sqrt{y})$$
which is ##F(\sqrt{y})-F(-\sqrt{y}## differentiating wrt y
$$P(\chi=y)\frac{1}{2\sqrt{y}} f(\sqrt{y})+\frac{1}{2\sqrt{y}} f(\sqrt{y})=\frac{1}{\sqrt{y}} f(\sqrt{y})$$
 
Apologies the website gave gives the formula for the general normal distribution for the standard normal, take ##\sigma=1,\mu=0## in the equation
 
Back
Top