Most articles that I have found on the subject of testing for the independence of two events state if $P (A \cap B) = P (A) P(B)$, then $A$ and $B$ are independent. If I only had estimates of $P (A\cap B)$, $P (A)$ and $P(B)$ collected from sample data rather exact probabilities calculated using the entire population, will this test for independence still work? If for example, if $P (A\cap B)$ was almost equal to $P (A) \times P(B)$, can I assume independence? Should I be using the chi-square test for independence instead?

  • $\begingroup$ What does your sample data consist of? $\endgroup$
    – Glen_b
    Jan 27, 2017 at 3:41
  • $\begingroup$ Not to nit-pick, but what would be defined as "almost equal"? $\endgroup$
    – Bensstats
    Jan 27, 2017 at 4:20
  • $\begingroup$ Ploni Almni, by "almost equal" I mean something akin to the statement "I believe with a confidence of 95% that event A is not dependent on B" $\endgroup$
    – eglinker
    Jan 27, 2017 at 4:33
  • $\begingroup$ Glen_b, I don't have a particular data set to provide for you off hand. What I want to be able to do is make observations of events A and B so that I can get estimates of the probabilities, and then use a hypothesis test infer whether or not events A and B ar independent of each other based on whether or not the difference between P(A&B) and P(A)P(B) is statistically significant. $\endgroup$
    – eglinker
    Jan 27, 2017 at 4:51

1 Answer 1


You seem to have observations which each can be classified as either A or not A, and either B or not B. That is exactly the pattern behind a $2\times 2$ contingency table, so just make your contingency table and use a chisquare test.

This is all we can say without more context and details.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.