Is it the case that for random variables $Y$ and $X$, $f(Y | X)$ in the same family as $f(Y)$?

If so how can I prove it? If not are there any situations (some families of distributions) where they can be the same? Or is it related to the link function(s)?



1 Answer 1


There are some specific cases where it is true, such as a bivariate normal, then $f(Y)$ and $f(Y|X)$ are both normal.

But consider the case where $f(Y|X)$ is normal (with mean depending on $X$) and $X$ follows a uniform distribution. Then $f(Y)$ is not normally distributed.

There are also distributions where the marginals ($f(X)$ and $f(Y)$) are both $\text{uniform}(0,1)$, but there is a hole in the square where the probability is $0$, so the conditional $f(Y|X)$ would not be uniform and for some values of $X$ would be disjoint.

  • 1
    $\begingroup$ +1, this is the right answer. For an illustration (eg, pictures) of how $f(Y)$ & $f(Y|X)$ can differ, there are some at my answer here: What if residuals are normally distributed but Y is not? $\endgroup$ Nov 7, 2013 at 17:20
  • $\begingroup$ What about members of exponential family? Or distributions with -inf->inf support? Can we relate $f(Y), f(X), f(Y|X)$ in some way? $\endgroup$ Nov 7, 2013 at 17:43
  • $\begingroup$ @CagdasOzgenc, I have seen a bivariate distribution where both marginals are normal, but there is a hole of 0 probability in the center (where the bivariate normal would have the highest probability), so for that distribution f(Y|X) is certainly not normal (may be approximate for some values of X, but clearly not when passing through the hole). $\endgroup$
    – Greg Snow
    Nov 11, 2013 at 15:20
  • $\begingroup$ @Greg, what if the distributions are nice? Meaning they are smooth, no discontinuities, support from -inf to +inf. $\endgroup$ Nov 11, 2013 at 19:17
  • $\begingroup$ Being 'nice' doesn't solve the problem; You can have both margins and the conditionals all 'nice' ... but still not of the same form, even with normal margins. $\endgroup$
    – Glen_b
    Nov 14, 2013 at 11:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.