0
$\begingroup$

I have to calculate/approximate the following multiple conditional probability:

$$P(X|C_1,C_2,C_3)$$

with only $P(X|C_1)$, $P(X|C_2)$ and $P(X|C_3)$ given. Is it possible to express the above in terms of these single conditional probabilities only? What independence assumptions would I need to make?

Edit: Concretely, I have an algorithm that can calculate single conditional probabilities $P(X|C_i)$ directly, and I want to use these to calculate multiple conditional probabilities. I'm fairly certain it is not possible to calculate this exactly, but perhaps I can make some independence assumptions so some terms can be dropped?

$\endgroup$
  • $\begingroup$ The question doesn't seem to be clear $\endgroup$ – freak_warrior Nov 15 '13 at 11:03
  • $\begingroup$ What part is unclear? I will clarify where needed... $\endgroup$ – niefpaarschoenen Nov 15 '13 at 11:05
  • $\begingroup$ I will try to add a bit more detail. $\endgroup$ – niefpaarschoenen Nov 15 '13 at 11:10
  • $\begingroup$ I don't quite understand why you get an upvote for a double comment without mentioning what is wrong with the question. In the meantime I have already edited the question... $\endgroup$ – niefpaarschoenen Nov 18 '13 at 10:17
  • $\begingroup$ I think I had accidentally reviewed your question twice. Anyway below is my answer! $\endgroup$ – freak_warrior Nov 18 '13 at 10:37
-3
$\begingroup$

Assume $C_1,C_2,C_3$ are independent of each other. Then

$$P(X|C_1)P(X|C_2)P(X|C_3)=\frac{P(XC_1)}{P(C_1)}\cdot\frac{P(XC_2)}{P(C_2)}\cdot\frac{P(XC_3)}{P(C_3)}\\ =\frac{P(XC_1C_2C_3)}{P(C_1C_2C_3)}=P(X|C_1C_2C_3)$$

$\endgroup$
  • $\begingroup$ Seems to not match with this answer : math.stackexchange.com/questions/1288770/… $\endgroup$ – DollarAkshay Oct 30 '18 at 9:36
  • 3
    $\begingroup$ @AkshayLAradhya Indeed, this is wrong. Consider $C_1,C_2,C_3$ to be the events of throwing a head on, respectively, the first, second, and third of three fair coins, and $X$ to be the event of throwing exactly two heads among those three coins. So $C_1,C_2,C_3$ are mutually independent events (although clearly not conditionally independent). Now $\mathsf P(X\mid C_1,C_2,C_3)=0$ and $\mathsf P(X\mid C_i)=1/2$ for each $i\in\{1,2,3\}$. As $0\neq 1/8$, we see that we do not have $\mathsf P(X\mid C_1,C_2,C_3) = \mathsf P(X\mid C_1)~\mathsf P(X\mid C_2)~\mathsf P(X\mid C_3)$. $\endgroup$ – Graham Kemp Oct 30 '18 at 12:03

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.