1
$\begingroup$

This thread shows that if two events are to be mutually exclusive and independent, one of them should have zero probability. I worked the following example that seems to contradict conditional probability.

We pick a real number in range [0,1]

A = event that the number is rational

B = event that the number is irrational

P(A and B) = $0$ (because they are disjoint by definition)

Also, P(A) = $0$ (from measure theory/discrete maths)

Thus P(A and B) = P(A) * P(B)

This means A and B are independent.

Note that the probability that number is irrational given that it is rational is zero. This does not satisfy conditional probability.

P(B | A) = 0 $\ne$ 1 = P(B)

P(B | A) $\ne$ P(B)

This just means that occurrence of one event (A) is affecting the probability of other (B), and not independent. Where did I go wrong?

$\endgroup$
  • $\begingroup$ $\mathbb{P}(B\mid A)$ is not defined if $\mathbb{P}(A)=0$. $\endgroup$ – carmichael561 Jul 12 '16 at 23:20
  • $\begingroup$ What makes this "not defined"? If I write it out, it will be 1/0, agreed. But we are dealing with probabilities here, P(B|A) must have some value between 0 and 1. $\endgroup$ – User49582934 Jul 12 '16 at 23:41
  • $\begingroup$ I wish to update my comment above. If A = getting $-5$ on a fair dice throw, and B = getting any number from 1 to 6 on the same throw, then P(A)=0, P(B)=1, and P(B|A) is undefined because A cannot occur, and conditional probability on A is undefined. But in rational/irrational example, A is a valid event, and P(B | A) is a valid phrase, and I believe P(B | A) = 0 $\endgroup$ – User49582934 Jul 13 '16 at 0:26
1
$\begingroup$

It's fine.   Contradictions are allowed to happen when the precondition is not satisfied.

If $\mathsf P(A)>0$, then $\mathsf P(B\mid A)\cdot\mathsf P(A) = \mathsf P(B)$ .

Even if the measure $\mathsf P(B\mid A)$ can be evaluated from the model, the consequent doesn't have to hold if the antecedent doesn't; ie: $\mathsf P(A=0)$.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Are A and B really independent if P(B | A) is not equal to P(B), i.e, the probability of event B is getting affected by the occurrence of A? $\endgroup$ – User49582934 Jul 13 '16 at 0:08

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.