I have been curious about this for a long time. Is there a sensible notion of a conditional random variable? I have been told "no" before, but I don't feel that I ever got a clear answer. And, in fact, I see them get used in statistics all the time. In particular, $x \sim (X|Y)$ is shorthand for

$$y\sim Y\\ x\sim (X|Y=y).$$

This "sample and bind" shorthand seems completely straight-forward to me, since we can "always" recover the distribution of $Y$ from the joint distribution of $(X,Y)$ by marginalizing $X$ out. Are marginalizing and conditioning adjoints? It seems that the notation witnesses triple-ability, if they are adjoint at all.

  • 2
    $\begingroup$ In terms of the probability behind this, you should look up disintegrations and regular conditional probabilities. I don't know about the categorial side of the question. $\endgroup$ Apr 29 '14 at 16:50

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.