[FieldTrip] FW: Combining CBPT & LMM
Bekke, M.E. ter (Marlijn)
marlijn.terbekke at donders.ru.nl
Tue Feb 22 11:10:25 CET 2022
Dear FieldTrip community,
My name is Marlijn ter Bekke and I'm a PhD student at the Donders Institute and Max Planck Institute in Nijmegen. I work on multimodal language processing (speech + gesture) during conversation. Currently, I'm planning the analysis for my first EEG study, and I have a question about the combined use of cluster-based permutation tests and linear mixed effects models.
I've read about a strategy (e.g. here<https://www.frontiersin.org/articles/10.3389/fnins.2018.00048/full>) where people first use cluster-based permutation tests to e.g. find an effect of Condition, and then average over this cluster (in time/space/frequency) and use this average value as the dependent variable in linear mixed models. The benefit of this approach is that parametric predictors that vary by subject (e.g. performance) or by item (e.g. word frequency) can easily be modeled.
I wonder whether this strategy is good practice, given that the cluster-based permutation test cannot be used to conclude that the effect occurred in a particular time window, for particular electrodes or in particular frequencies (see Fieldtrip FAQ<https://www.fieldtriptoolbox.org/faq/how_not_to_interpret_results_from_a_cluster-based_permutation_test/> or Sassenhagen & Draschkow, 2019<https://doi.org/10.1111/psyp.13335>). Does it then make sense to use the found cluster as a basis for averaging? For example, if the test cannot say there was an effect specifically from e.g. 200-400ms, then why average only over this time window?
I'd be very interested to hear your opinions on this matter!
Marlijn ter Bekke
Marlijn ter Bekke<https://www.mpi.nl/people/bekke-marlijn-ter>
PhD student @ Donders Centre for Cognition
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the fieldtrip