4
$\begingroup$

Is it generally true that for random variables $X$ and $Y$, regardless of being dependent or independent, that $E[X] = E[X \mid Y \le a] + E[X \mid Y>a]$ ?

$\endgroup$
  • 1
    $\begingroup$ Assume $X$ and $Y$ are highly correlated standard normal variables. Then $E[X\mid Y\le L] + E[X \mid Y > L] \approx 0 + L = L$ when $L$ is a very large number. If you assume that $L$ is a large negative number you get a contradiction. $\endgroup$ – Hunaphu May 7 '15 at 13:00
  • 3
    $\begingroup$ Consider a homely analogue: is the average height equal to the average height for those below average weight PLUS the average height for those above average weight? Answer: No. $\endgroup$ – Nick Cox May 7 '15 at 13:03
13
$\begingroup$

Not quite, if we use the law of total expectation we would have that

$$ E(X) = E(X| Y \le a)P(Y \le a) + E(X|Y > a) P(Y>a) $$

$\endgroup$
4
$\begingroup$

Exactly Chappers, and I can just add if $X$ and $Y$ are independent continuous random variables, $E[X|Y]=E[X]$.

So Mauro, your formula would be false because in this case, you'll have

$E[X]=E[X|Y\le a]+E[X|Y>a]=E[X]+E[X]=2*E[X]$

Finally, regardless the dependence between $X$ and $Y$, we have always:

$E[X]=E[E[X|Y]]$

$\endgroup$
  • $\begingroup$ Ah yeah sorry I feel dumb now ;_; I was trying to prove it with like integrals. Just to verify though is: f(x, y<=a) +f(x,y>a) = f(x,y) Where f is the joint distribution of X,Y. If so I think I got it from the integrals too. $\endgroup$ – Mauro Augusto May 7 '15 at 13:15

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.