[FieldTrip] Help understanding cluster-based permutation implementation

[Anton Malko]

Dear FieldTrip community,

I am a psycholinguist at the Australian National University, working with eye-tracking data. I have been writing a small custom cluster-based
permutation analysis script, and I wanted to ask for your help in understanding some details of the algorithm (as implemented in FieldTrip).

1. Computation of cluster p-values:

In "clusterstat.m" file, strict inequalities are used when comparing the observed test statistic to the permutation distribution, e.g., at line 443 for positive clusters:
 prob(j) = (sum(posdistribution>stat(j)) + 1)/(Nrand + 1);

My understanding has been that "greater than or equal to" needs to be used for computing cluster p-values (e.g., p. 187 in Maris & Oostenveld 2007). And since
test statistic can be discrete (e.g., cluster length), there can be cases where it will be exactly equal to one or several values in the permutation distribution.

Why are strict inequalities used in the code?

2. Permutation distribution for two-tailed tests:

In FieldTrip, two test statistics are computed on the observed data (for pos and neg clusters), and each is compared to the respective permutation distribution
(followed by a Bonferroni corection).

It makes sense, but I can't figure out how it compares to an alternative, namely, recording the maximum *absolute* value of the test statistic, and comparing it
against a single permutation distribution of maximum absolute test statistics (probably, without a Bonferroni correction?). My understanding is that
Maris & Oostenveld (2007) paper describes this option.

Is it just an implementation choice which would give the same results? Or is using two distributions for pos/neg clusters better than using one distribution
of max absolute values -- if so, why?

I would be grateful for any advice!

Best wishes,
Anton

--------------------------

Dr Anton Malko
Research Fellow
School of Literature, Languages and Linguistics | College of Arts & Social Sciences
The Australian National University, Canberra
anton.malko at anu.edu<mailto:anton.malko at anu.edu>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20231207/71c91b41/attachment.htm>