[FieldTrip] cluster-based test with Wilcoxon signrank test

[Andreas Wutz]

Dear fieldtrippers,

I am trying to combine the Wilcoxon signrank test statistic with the cluster-based permutation approach. To this end, I wrote my own statfun, which calculates the Wilcoxon test statistic (using signrank.m in the stats toolbox) for each random partition. The output of my statfun is s.stat, which contains a z-value (which is output by the function signrank.m for high enough N).

I understand that the cluster-based test does not depend on the specific test statistic used. My question applies to the initial threshold, which is used for clustering. As I see it, I have three options here.

1) parametric
In this case, I would calculate the critical value(s) for a normal distribution (+/- 1.96) and set this as s.critval.

2 - 3) nonparametric_individual / nonparametric_common
I assume the threshold is here based on the permutation distribution with those samples chosen for clustering that exceed the 1-clusteralpha quantile. is this assumption correct? What is the difference between those two methods?

In general, I am unsure which of the three methods to choose. Which are the advantages / disadvantages of each option? I would be very happy about your feedback!

Best,
Andreas


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20170817/c869e1f6/attachment-0001.html>