[Eeglablist] One-sample cluster-based permutation test for identifying spatiotemporal ERP locus

[Uziel Miguel Rayas Hernandez]

 Hi everyone,

I'm working with EEG data in a within-subject design, where all participants completed two experimental conditions (each subject served as their own control). The preprocessing steps include re-referencing to the average, low-pass filtering, and ICA decomposition using ‘runica’.

My goal is to identify the spatiotemporal locus of ERP components. I'm following a two-level permutation approach inspired by Robinson et al. (2022), where the first level identifies the temporal locus and the second identifies the spatial locus.

In standard permutation tests, we typically shuffle condition labels during a t-test to generate a null distribution under the assumption that condition labels have no effect. However, our research interest is in analyzing each condition independently—we want to identify the spatiotemporal ERP patterns within each condition, without influence from the other. The idea is to treat each condition as its own population, rather than comparing them directly.

To do this, we considered applying a one-sample permutation test, where we would:


  *   Flip the sign of each subject’s data randomly across permutations (within a condition),


  *   Compare the observed ERP signal against zero or baseline activity.

This approach seems to make logical sense to us, as the null hypothesis here would be “no significant deviation from baseline,” and we could then test for consistent ERP activity within each condition separately.

However, I found a discussion in the FieldTrip mailing list (https://urldefense.com/v3/__https://mailman.science.ru.nl/pipermail/fieldtrip/2018-August/038192.htmlwhere__;!!Mih3wA!Hi5SLETHfOBcDLjDtl41-DqCMlrXnfER4L5w0eiF5SIG3XO7P0sTy_TuDqPPrTBwJwMBE1eXaPdIdln_poOqjpBk4k4b-9il5aA$ ) Eric Maris advised against this type of test, emphasizing that permutation tests are designed to compare two or more conditions, and that the null hypothesis is about the distribution of those conditions.

This raises my main question:

Is it valid to use a one-sample t-test in permutation test (with sign-flipping) to identify significant ERP activity within a single condition, relative to baseline or zero?
And if so, are there precedents in the EEG literature where this method has been used to locate spatiotemporal ERP patterns within conditions (not between)?

We’d appreciate any insight or references to studies that have used this or a similar approach or any suggestion about other methods more adapted to our needs. Thank you very much in advance!

Best regards,
Uziel Rayas