[FieldTrip] low-pass filtering
Matthias.Franken at UGent.be
Wed Jun 27 11:57:18 CEST 2018
I’ve been wondering about a similar question a while ago. As far as I can see, I think it can affect cluster size, but without compromising the permutation test. More smoothing should (I guess) lead to larger clusters, but this should affect both the observed clusters as well as the clusters in your premutation distribution (after having permuted data labels). So I guess the comparison should be unaffected.
Experimental Psychology Department
From: Simanova, I. (Irina)<mailto:i.simanova at donders.ru.nl>
Sent: Tuesday, 26 June 2018 18:44
To: fieldtrip at science.ru.nl<mailto:fieldtrip at science.ru.nl>
Subject: [FieldTrip] low-pass filtering
We recently submitted a paper where we use a cluster permutation analysis of ERFs (testing conditions exchangeability). The EEG was sampled at 500 Hz and low-pass filtered at 40 Hz during preprocessing. One of the reviewers has indicated that using this temporal resolution for the analysis seems unnecessary given the low-pass filter, which makes each time point not independent due to smoothing. He further asks: "Do all key significant components replicate if the analysis is performed using a mean amplitude that is averaged for larger time bins (25 ms bins) instead of individual time points?"
I understand that when doing parametric analysis one might want to reduce the number of comparisons by averaging ERF samples over time bins. But here we solve the MCP with the cluster analysis. But it also seems like the reviewer is concerned about non-independecy in the data resulting from low-pass filtering. I wanted to check with you: can smoothing time-series indeed compromise the cluster-based analysis (e.g. affect the cluster size)?
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the fieldtrip