When the joint probability for events $A_1,A_2,A_3,A_4$ is desired, the chain rule is used as follows:$$ P(A_4, A_3, A_2, A_1) = \mathrm P(A_4 \mid A_3, A_2, A_1)\cdot \mathrm P(A_3 \mid A_2, A_1)\cdot \mathrm P(A_2 \mid A_1)\cdot \mathrm P(A_1)$$ However, I have a problem in which I need the probability of the joint event and calculation of the conditional probabilities is not possible. Is there any approach that I could find an upper or lower bound for this probability? or any sort of approximation, no matter how elementary it is. I only have the probabilities of events alone, i.e. $P(A_4), P(A_3), P(A_2), P(A_1)$.

  • 1
    $\begingroup$ It is $P(A_i|A_j)\geq P(A_i)$. Therefore one (rough) lower bound is $P(A_1)\cdot P(A_2)\cdot P(A_3)\cdot P(A_4)$ $\endgroup$ – callculus Jul 9 '16 at 7:40
  • $\begingroup$ Yes, however this was very far from the actual (I think there is a considerable dependency between them in my problem). I am looking for something more accurate. $\endgroup$ – asterix Jul 9 '16 at 7:46
  • $\begingroup$ @callculus If $A_i\cap A_j=\varnothing$ then $P(A_i\mid A_j)=0$ and $P(A_i)>0$ is not excluded. $\endgroup$ – drhab Jul 9 '16 at 7:50
  • $\begingroup$ Do you have any further information ? $\endgroup$ – callculus Jul 9 '16 at 7:50
  • $\begingroup$ @drhab Yes, I haven´t considered this special case. I thought that all events are dependent. $\endgroup$ – callculus Jul 9 '16 at 7:54

Without more information upperbound is: $$\min(\Pr(A_1),\Pr(A_2),\Pr(A_3),\Pr(A_4))$$

This is based on $A_1\cap A_2\cap A_3\cap A_4\subseteq A_i$ for $i=1,2,3,4$ together with the fact that $=$ instead of $\subseteq$ is not excluded here.

If e.g. $\Pr(A_1)+\Pr(A_2)\leq1$ then it is not excluded that $A_1\cap A_2=\varnothing$ so in such cases $0$ serves as lower bound. Not quite useful of course. For a useful lower bound more information concerning the events is needed.

  • $\begingroup$ Is it possible to derive some approximation for the conditional probabilities as well? $\endgroup$ – asterix Jul 9 '16 at 8:27
  • $\begingroup$ I can only think of things like: $P\left(A_{1}\right)+P\left(A_{2}\right)-1\leq P\left(A_{1}\cap A_{2}\right)\leq\min\left(P\left(A_{1}\right),P\left(A_{2})\right)\right)$ and dividing both sides by $P\left(A_{1}\right)$ gives you something like $\cdots\leq P\left(A_{2}\mid A_{1}\right)\leq\cdots$. $\endgroup$ – drhab Jul 9 '16 at 8:40
  • $\begingroup$ OK, I will work more on it (possibly some simulations) then will get back to this post. $\endgroup$ – asterix Jul 9 '16 at 8:48

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.