3
$\begingroup$

From $P(A\mid B)$ and $P(B\mid C)$ can you compute $P(A\mid C)$?

Is this statement true? I couldn't find any formulae to that would relate these events. Bayes theorem certainly wasn't helpful here...

$\endgroup$
1
  • 3
    $\begingroup$ To see this is impossible, consider the case where $P(B) = 1$. Then $P(A|B) = P(A)$ and $P(B|C) = 1$, but $P(A|C)$ could be anything. $\endgroup$
    – fblundun
    Jan 15 '21 at 12:36
2
$\begingroup$

You can only if you assume $A$ is conditionally independent of $C$ given $B$, i.e. $P(A | B, C) = P(A | B)$. In other words, once you know the value of $B$, the outcome of $C$ tells you nothing further about $A$.

This conditional independence assumption allows: \begin{align} P(A | C ) & = \sum_{b\in\Omega_{B}} P( A | B = b, C) P(B = b | C) & \text{Law of total probability} \\ & = \sum_{b\in\Omega_{B}} P( A | B = b) P(B = b | C) & (A \perp\!\!\!\perp C)~ |~ B \end{align}

where $\Omega_B$ is the set of possible outcomes of $B$.

$\endgroup$
2
  • $\begingroup$ But this assumes that $P(A|B^c)$ is known which it isn't - we only know $P(A^c|B)=1-P(A|B)$. $\endgroup$ Jan 15 '21 at 11:54
  • $\begingroup$ I'm interpreting $A$, $B$, and $C$ from the original post to refer to random variables, not specific outcomes, in which case known $P(A | B)$ and $P(B | C)$ would be known conditional probability tables. $\endgroup$
    – jkpate
    Jan 15 '21 at 12:04

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.