0
$\begingroup$

Suppose that $X$ and $Y$ are two random variables defined on a say, finite sample space $A$ and suppose that $p$ is a probability distribution on $A$. Is there a special name for the following quantity? $$\sum_{a\in A} X(a)Y(a)p(a)-\sum_{a\in A} X(a)p(a)\cdot \sum_{a\in A} Y(a)p(a)$$ It's certainly not covariance. Covariance of $X$ and $Y$ is $$\sum_{a,b\in A} X(a)Y(b)p_{X,Y}(a,b)-\sum_{a\in A} X(a)p_{X}(a)\cdot \sum_{a\in A} Y(a)p_Y(a)$$ where $p_{X,Y},p_X,p_y$ are the joint distribution on $A\times A$ and marginals on $A$ respectively. I don't know whether this can be thought of as a covariance of $X$ and $Y$ in some sense. Or am I missing something here?

$\endgroup$
  • $\begingroup$ You might try to give an example, perhaps where $X$ and $Y$ each take two values and they are neither independent not completely dependent. In particular, you might try to say what $p_X(a)$ means, and look at whether $\sum_{a \in A} p_X(a) =1$. $\endgroup$ – Henry Jan 4 '14 at 12:30
  • $\begingroup$ @Henry: Yes $X(a)$ and $Y(a)$ could be different and independent. $P_X(a)=\sum_{b\in A}P_{X,Y}(a,b)$. $\endgroup$ – Ashok Jan 4 '14 at 12:43
2
$\begingroup$

Your first expression is covariance, but the space $A$ may not be what you expect, and is not the same as the $A$ in your second expression.

For example if $A$ has four distinct parts $e_1, e_2, e_3, e_4$ with

  • $p(e_1)= 0.4, \qquad X(e_1)=6, \qquad Y(e_1)=6$
  • $p(e_2)= 0.1, \qquad X(e_2)=6, \qquad Y(e_2)=7$
  • $p(e_3)= 0.2, \qquad X(e_3)=7, \qquad Y(e_3)=6$
  • $p(e_4)= 0.3, \qquad X(e_4)=7, \qquad Y(e_4)=7$

then your first expression gives $41.7-6.5\times 6.4 =0.1$, which is the covariance of $X$ and $Y$.

$\endgroup$
  • $\begingroup$ Thank you Henry. But, the actual covariance is 0 because $E[XY]=\sum_{x,y}x~y~P(X=x,Y=y)=41.6$. What's wrong here? Am I missing something? $\endgroup$ – Ashok Jan 5 '14 at 7:54
  • $\begingroup$ $0.4 \times 6 \times 6 + 0.1 \times 6 \times 7 + 0.2 \times 7 \times 6 + 0.3 \times 7 \times 7 = 41.7$. In my example the covariance must be positive as low/low and high/high are more likely than high/low or low/high. $\endgroup$ – Henry Jan 5 '14 at 16:20
  • $\begingroup$ Yes you are right. Sorry for the confusion. $\endgroup$ – Ashok Jan 6 '14 at 5:12

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.