Can someone define independence of two random variables with this "product rule", or are there any counterexamples?

  • 2
    $\begingroup$ Sure there are, try X standard normal and Y=UX with U symmetric Bernoulli. $\endgroup$ – Did Oct 29 '15 at 19:13
  • $\begingroup$ @Did, what do you mean by symmetric Bernoulli? $P(U=0) = P(U=1)$? $\endgroup$ – Carlos Mendoza Oct 29 '15 at 19:16
  • $\begingroup$ P(U=1)=P(U=-1)=1/2 $\endgroup$ – Did Oct 29 '15 at 19:29
  • $\begingroup$ @Did Is that a Bernoulli variable? I don't think so. $\endgroup$ – Carlos Mendoza Oct 29 '15 at 19:32
  • 1
    $\begingroup$ @CarlosMendoza Many people call Bernoulli what you probably call a two-points distribution. $\endgroup$ – Did Oct 29 '15 at 19:33

Independence implies uncorrelation

We say that two random variables are uncorrelated if $E[XY] = E[X]E[Y]$.

If $X$ and $Y$ are independent and discrete (this can also be extended to continuous random variables), then

$$p_{X,Y}(x,y) = p_X(x)p_Y(y)\qquad \forall (x,y)$$

Using this in the definition of $E[XY]$ and doing some rearrangements of terms, we have

\begin{align} E[XY] &= \sum_x\sum_y xyp_{X,Y}(x,y)\\ &= \sum_xxp_X(x)\sum_yyp_Y(y)\\ &= E[X]E[Y] \end{align}

Therefore, if $X$ and $Y$ are independent, then they are uncorrelated.


Uncorrelation does not imply independence (in general)

In general$^1$, it is false to say that if $X$ and $Y$ are uncorrelated, then they are independent. Let's see a counterexample. Let $X$ and $Y$ be random variables taking values in $\{(0,1), (1,0), (-1,0), (0,-1)\}$ with equal probability $0.25$. Then, $XY = 0$ with probability $1$, and therefore $E[XY] = 0$.

By other hand,

$\displaystyle p_X(x) = \left\{ \begin{array}{ll} \frac{1}{4} & x=-1\\ \frac{1}{2} & x=0\\ \frac{1}{4} & x=1\\ \end{array} \right. $ $\hspace{1cm}$ and $\hspace{1cm}$ $p_Y(y) = \left\{ \begin{array}{ll} \frac{1}{4} & y=-1\\ \frac{1}{2} & y=0\\ \frac{1}{4} & y=1\\ \end{array} \right. $

which implies that $E[X]=E[Y]=0$. Then, $E[XY] = E[X]E[Y]$ and the random variables are uncorrelated.


$$p_X(1)p_Y(1) = \frac{1}{4}\frac{1}{4} \neq p_{X,Y}(1,1) = 0$$

which shows that $X$ and $Y$ are not independent.

We can conclude then that independence is a sufficient but not a necessary condition for uncorrelation.

$^1$ The only two exceptions I am aware of are the jointly Gaussian variables and two Bernoulli random variables...any other?


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.