5

If random variables X and Y are correlated, but Y and Z are iid, is X and Z always independent? I am able to prove they are independent for specific types of structure on X and Y (for example if $X=Y+U$ where $U$ is iid), but is this true in general. Not sure why I am having such a difficult time proving the general case?

Nanstam
  • 53
  • 3
  • A similar question is asked at http://stats.stackexchange.com/questions/5747, which has several useful answers. – whuber Jun 20 '14 at 15:23
  • Imagine X, Y, Z are a bunch of 3 vectors (arrows) in 3d space going from the same point. "Independent" means 90 degree angle. Any other angle is "correlated". Visualize the possibilities... – ttnphns Jun 20 '14 at 15:24
  • Another thread you could read is: [How can 2 variables each be strongly correlated with a 3rd variable, but uncorrelated with each other?](http://stats.stackexchange.com/q/83922/7290) – gung - Reinstate Monica Jun 20 '14 at 18:49
  • 1. You can't prove a negative. 2. It can be *disproven* and has been by @heropup's counterexample. – Aaron Hall Jun 20 '14 at 19:01

1 Answers1

16

No.

Counterexample: Suppose $Y, Z \sim {\rm Bernoulli}(0.5)$ are IID. Now let $X = YZ$. $X$ clearly is not independent of either $Y$ or $Z$.

Muhd
  • 103
  • 2
heropup
  • 5,006
  • 1
  • 16
  • 25