Hello,
I would like to run the non-parametric equivalent of a one-sample t-test against zero (e.g., Wilcoxon signed-rank). At the moment with one group of files I can only select parametric test against zero or baseline. It seems like non-parametric alternatives are implemented for paired samples but not for one group against zero. Is there an easy way to compute it?
Thanks,
Svetlana
Hi Svetlana,
Permutation tests are based on the idea that if two sample sets have the same properties (null hypothesis), then you can exchange samples between the two sets freely:
http://neuroimage.usc.edu/brainstorm/Tutorials/Statistics#Nonparametric_permutation_tests
In a one-sample test, there is nothing to permute the samples with…
Alternative suggestions are welcome.
Francois
Hi Francois,
Indeed, a permutation test isn’t possible with only one group of observations. However it seems like a Wilcoxon signed rank test should be possible. See below:
http://biostat.mc.vanderbilt.edu/wiki/pub/Main/AnesShortCourse/NonParametrics.pdf (slide 10)
or https://www.xlstat.com/en/solutions/features/one-sample-wilcoxon-signed-rank-test
The procedure seems similar to the paired version (where it’s the difference between samples that is compared to zero), but in the case of a one-sample test, it would be the observations directly compared to zero. What do you think?
Svetlana
@pantazis: What do you think?
@spinet: Want to code this?
I probably won’t do it myself I have very little time left to develop new things in Brainstorm at the moment. You can look at the process_test_* functions to see how this is organized.
If it seems statistically sound to you, I could yes. I would need it for a project anyway!