Fisher exact test statistic

[JorgeCamachoS]

Dear Jamoviers:

Past versions of jamovi included the value of Fisher exact test statistic as well the associate probability. The 2.3.9 version include de probability but not the statistic value. Could you include this value?

Thanks in advance for your support

Jorge Camacho

[reason180]

There's a test statistic for Fisher's exact test?

[jonathon]

Past versions of jamovi included the value of Fisher exact test statistic as well the associate probability. The 2.3.9 version include de probability but not the statistic value. Could you include this value?

i don't think we've changed anything here ... are you saying an option that used to be there, is no longer there?

jonathon

[JorgeCamachoS]

Dear Jonathon:

In an old analysis, using a past version of Jamovi, I got a value for Fisher Exact test as well as the related probability. The 2 x 2 table was:

Output1 Output2
Trat1 9 3
Trat2 4 9

The statistics value was 6.181 and the probability 0.0472.

Thanks for your support

Jorge Camacho

[MAgojam]

Hey Jorge,
the elegant question from @reason180

There's a test statistic for Fisher's exact test?

suggests that there is no statistic for Fisher's exact test.
The reported value, returned from the stats::fisher.test(...)$estimate test used, for a 2x2 table was the odds ratio, but in a larger table it was always NULL.
The exact test does not have a statistic, it directly provides a p-value.
I believe there was a fix for the contingency table in jamovi's v1.6.5 that changed the code to only show the p-value for the exact test (Value always blank) to avoid misunderstanding with the estimation as Value.

Take a look at the screenshot.

Fisher's exact test.png (70.29 KiB) Viewed 9328 times

Cheers,
Maurizio

[reason180]

I used to think I understood Fisher' exact test, but now I don't think I do. I've been under the impression that Fisher's exact test is a special case of a permutation test. See https://journals.physiology.org/doi/pdf ... 00072.2012

However, if that's true, there would need to be a test statistic. For example: When all possible re-pairings of the levels of the X variable with the levels of the Y variable, ____% of the time is the observed difference between two proportions at least as large as it was prior to the re-pairings. So I wonder if either the odds ratio is THE permutation test statistic, if the difference between two proportions is THE permutation statistic, or if the result is the same regardless of which of those statistics is used? (Presumably the odds ratio and the difference between two proportions are monotonic functions of each other.)

[JorgeCamachoS]

Dear reason180:

Thanks for your reply about Fisher exact test

Best regards,

Jorge Camacho S