0
$\begingroup$

Usually the following formulation of the Law of Total Probability is used:

If events $\{H_i\}$ with positive probabilities form the partition of a sample space then for any event $A$ on the same probability space we have $P(A) = \sum_{i} P(A|H_i)P(H_i)$

Do we rigorously need the partition here? Can we use the following alternative formulation instead?

If pairwise disjoint events $\{H_i\}$ with positive probabilities form the cover of the event $A$ (i.e. $A \subseteq \bigcup_i H_i$) then we have $P(A) = \sum_{i} P(A|H_i)P(H_i)$.

$\endgroup$
  • 1
    $\begingroup$ Your alternative formulation is correct! $\endgroup$ – Mike Earnest Apr 3 '18 at 23:09
1
$\begingroup$

Hint: What is $\Pr ( A \vert H_i )$ when $A \cap H_i = \emptyset$?

This should tell you the connection between the two statements.

$\endgroup$
  • $\begingroup$ Yes, when $A \cap H_i = \varnothing$ then $P(A|H_i) = 0$. $\endgroup$ – Rodvi Apr 4 '18 at 9:35
  • $\begingroup$ @Rodvi Good. Now you should hopefully be able to see why the sum in your first statement is the same as the sum in the second. $\endgroup$ – Theoretical Economist Apr 4 '18 at 9:37
  • $\begingroup$ I see. Just want to say that the second statement may be more convenient in situation when we don't know the partition of $\Omega$. In that case it will be enough to find the cover of $A$ to apply LOTP using the second statement. $\endgroup$ – Rodvi Apr 4 '18 at 9:53
  • $\begingroup$ @Rodvi To be honest I can’t imagine a situation where you might not know the partition, though that might be due to a lack of imagination on my part. However, the two statements are equivalent, so you are free to use whichever you find more convenient. $\endgroup$ – Theoretical Economist Apr 4 '18 at 9:55
-1
$\begingroup$

Once you have pairwise disjoint non-empty sets $\{ H_i \}$ that cover $A$, that gives you a partition --i.e., the class $\{ H_i \cap A \}$ partitions $A$. So your second statement is not really an alternative result at all - it is just a way of stating the same result.

$\endgroup$
  • 1
    $\begingroup$ I disagree. Pairwise disjoint non-empty sets $\{H_i \}$ that cover $A$ give us the partition of $A$. While in the first formulation above $\{H_i\}$ form the partition of sample space $\Omega$. And $A \ne \Omega$. $\endgroup$ – Rodvi Apr 4 '18 at 9:32

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.