Suppose we start off with the traditional standard bivariate normal distribution:

$$\phi_2(x,y|\rho,\mu_x=0,\mu_y=0,\sigma_x=1,\sigma_y=1)=\frac{1}{2\pi\sqrt{1-\rho^2}}\exp \left(-\frac{x^2-2\rho x y + y^2}{2(1-\rho^2)}\right)$$

where the $\mu$s are the mean parameters, the $\sigma$s are the standard deviation parameters and $\rho$ is the correlation term between $x$ and $y$.

My question is: what happens to $\phi_2$ when we fix $y=y_F$? In other words, what happens when the only "free" variable in $\phi_2$ is $x$?

In this case, can we express $\phi_2$ in terms of a univariate normal distribution $\phi_1$?


After banging my head against my table for quite a bit, I noticed that I was confusing two different concepts:

  • a bivariate normal distribution with $y$ fixed at some point $y_F$
  • the conditional bivariate distribution with a given $y$ (i.e. $prob(X=x|Y=y_F)$)

These are two very different beasts, but I was treating them as the same.

  • $\begingroup$ Are you asking for the pdf of $X\mid Y$ where $(X,Y)$ is bivariate normal? $\endgroup$ Apr 18 '19 at 7:00
  • $\begingroup$ Yes, exactly. I know we can express the pdf of $X|Y$ by the traditional conditional distribution rule $\phi_2(X|Y)=\frac{\phi_2(X,Y)}{\phi_1(Y)}$. But this solution still requires the calculation of a bivariate pdf $\phi_2$. I'm trying to simplify things such that, in the end, I only need to deal with univariate distributions $\phi_1$. $\endgroup$
    – Felipe D.
    Apr 18 '19 at 7:20
  • $\begingroup$ The ratio is obvious since it is equal to the inverse of the value of PDF of Y at Y=1 and as long as you keep Y=1, no matter what X you take, the ratio will remain the same. Moreover, in your comment you write $\phi_2(X|Y)$, which is wrong. Following your notations, since X|Y is a univariate random variable, it's distribution should be represented by $\phi_1(X|Y)$ instead. $\endgroup$ Apr 18 '19 at 19:03
  • $\begingroup$ A bivariate random variable (X,Y) and a random variable (X|Y) are two different random variables. I think you are confusing between the two. $\endgroup$ Apr 18 '19 at 19:07
  • $\begingroup$ You're right, @SanketAgrawal, I was getting my notation mixed up. I finally understand what the heck was going on. Thanks for the help! $\endgroup$
    – Felipe D.
    Apr 18 '19 at 23:03

Here's how you can calculate $\phi_2(x,y,\rho)$ when $y$ is fixed at $y_F$.

First, from basic conditioning rules, we know that:

$$ \text{prob}(X=x|Y=y_F) = \frac{\text{prob}(X=x,Y=y_F)}{\text{prob}(Y=y_F)} \Rightarrow$$

$$ \text{prob}(X=x,Y=y_F) = \text{prob}(X=x|Y=y_F) \cdot \text{prob}(Y=y_F) \Rightarrow$$

$$ \phi_2(x,y_F,\rho) = \text{prob}(X=x|Y=y_F) \cdot \phi_1(y_F|\text{mean}=0,\text{var}=1) $$

Now,using the information from the conditional probability of bivariate normal distributions, we can swap out that $\text{prob}$ term in the middle by an actual function:

$$ \phi_2(x,y_F,\rho)=\phi_1(x|\text{mean}= \rho y_F, \text{var}=1-\rho^2) \cdot \phi_1(y_F|\text{mean}= 0, \text{var}=1)$$


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.