I'm trying to get a better understanding of conditional independence. Specifically, I am interested in what is sufficient to condition on. I have read this thread and this thread but they do not explain the particular aspect that I am trying to understand better.

Let X be a random variable, $X\in \mathbb{R}$ and


where h is some function. It trivially follows that

$Y \perp\!\!\!\perp X |X $


$Y \perp\!\!\!\perp X |h(X) $.

However, there are clearly other functions of X also fulfilling this, i.e. $Y \perp\!\!\!\perp X |g(X)$.

My question is: What are the (measure theoretic?) conditions on $g(X)$ for this to hold?

My attempt: Intuitively, as long as $g(X)$ contains more or equally as much "information" about $X$ as $h(X)$ it should hold. For example if $h(X)=|X|$ and $g(X)=X^2$, I would suspect it to hold that

$Y \perp\!\!\!\perp X |g(X)$

However, with my limited measure theoretic knowledge it is difficult to confirm/reject this formally. I am temped to, based on the results in van Putten and van Schuppen (1985), write something like

$Y \perp\!\!\!\perp X |g(X)$


$\sigma(g(X)) \supseteq \sigma(h(X))$

But then I do not really know what $\sigma(g(X))$ and $\sigma(h(X))$ is in the example above, although I think I have seen every clip on youtube regarding the definition of a $\sigma$-algebra by now.

Any help is much appreciated!


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.