Is there an alternative to the Fishers's exact test that has as null hypothesis that the relative proportions of one variable are dependent of the second variable?
4
-
1$\begingroup$ I would be surprised if there are: The usual practice is to have a 'simple' null hypothesis (no association, pop mean 100, 2 pop means equal, and so on). The null hypothesis $H_0$ provides a probability distribution, and marked divergence from that dist'n leads to rejection of $H_0.$ If you're talking about 'dependent', then there are degrees of dependency to be considered. // Why not explain your data and objective. There might be a way to look at your situation so that you can use Fisher's exact test. $\endgroup$– BruceETCommented May 25, 2020 at 22:49
-
$\begingroup$ @BruceET, if you look at the example here: docs.scipy.org/doc/scipy/reference/generated/… I compute the p-value with as matrix [[1278,209],[1244,243]]. The p-value I get is 0.0918 therefore I cannot reject the null hypothesis of whales and sharks not having a preference for a given ocean. What I would like to do instead is to have as null hypothesis that whales and sharks do have a preference for an ocean and test whether I can reject it or not. $\endgroup$– NewbieCommented May 25, 2020 at 23:37
-
1$\begingroup$ Sorry, I don't see what you are trying to accomplish with that. And as i explained in my previous comment, this is not a useful kind of test. $\endgroup$– BruceETCommented May 25, 2020 at 23:42
-
1$\begingroup$ There is much good material about this situation in our answers that mention "TOST". Check it out! $\endgroup$– whuber ♦Commented May 26, 2020 at 13:21
Add a comment
|