Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I want to characterize the probabilistic ordering of some (random) variables without going into a parametric from of the variables themselves. I couldn't easily find any theory for this and I am not sure what it is called, but I can try to explain through an example what I mean. Consider two random variables, $X$ and $Y$. There are only two possible orderings and I can characterize it as $P(X\le Y) = p_0$ and $P(X>Y) = 1-p_0$. What if I now have a third variable $Z$. Obviously, there now $3!=6$ orderings and we can give 6 parameters summing to 1, but I want to give more "separable" parameters, utilizing $p_0$ that we already have. Is it sufficient to also give $P(Y\le Z)$, $P(X\le Z)$? Once I give these parameters I should be able to compute the probability of observing any ordering. It seems to me that there must be a theory of such probabilistic orderings but I am unable to find it anywhere, or come up with a solution on first principles.

share|improve this question
1  
@Michael: If you don't have enough letters to add in an edit, you can always add ${}{}{}{}{}{}$ at the end of a paragraph. –  t.b. Aug 2 '11 at 15:12

1 Answer 1

Once I give these parameters I should be able to compute the probability of observing any ordering.

No, one cannot reconstruct quantities like $P(X\le Y\le Z)$ from $P(X\le Y)$, $P(Y\le Z)$ and $P(Z\le X)$ only. Note that there are 5 free parameters on the side of the 3-dimensional ordering and only 3 on the side of the 2-dimensional orderings hence this would be surprising if one could. Here is an explicit counterexample.

First consider $X$, $Y$ and $Z$ i.i.d. uniform on $(0,1)$. Then, by symmetry, $P(X\le Y\le Z)=\frac16$ and $P(X\le Y)=P(Y\le Z)=P(Z\le X)=\frac12$.

Consider on the other hand $X$ and $Y$ i.i.d. uniform on $(0,1)$ and $Z=X+Y$ mod $1$. Then $Z$ is uniform on $(0,1)$ and each pair $(X,Y)$, $(Y,Z)$ and $(Z,X)$ is uniform on $(0,1)\times(0,1)$ hence $P(X\le Y)=P(Y\le Z)=P(Z\le X)=\frac12$. But $[X\le Y\le Z]=[X\le Y\le 1-X]$ hence $P(X\le Y\le Z)=\frac14$.

share|improve this answer
    
I understand that the 3 parameters are insufficient, I just want to have parameters to characterize these orderings, even if they are 5. But is there a way to do this other than just write the different probabilities for all the cases separately? –  highBandWidth Aug 2 '11 at 13:09
    
You could use the 5 parameters $P(X\le Y)$, $P(X\le Y\le Z)$, $P(X\le Z\le Y)$, $P(Y\le X\le Z)$ and $P(Y\le Z\le X)$, but there are many other, noncanonical, possibilities. –  Did Aug 2 '11 at 13:49

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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