Would you recommend starting from trials and then average all coherence values, or starting from concatenated data?
Different research groups do different things, there are pros and cons for both approaches.
In favor of concatenating (the default option if you select all the trials in Process1): it provides longer time series, you get more stable estimators for cross-spectra. Necessary when the epochs are very short
Against the concatenation: it introduces discontinuities in the signal, which creates large artifacts in spectral domain.
With 4s epochs, maybe this is long enough not to have to concatenate the trials, and average the coherence values computed for each trial separately. But I'm not sure...
@Sylvain @hossein27en @pantazis What do you think?
In both cases, you can't make any direct conclusion with one or the other graph. In general, you should not try to compare the coherence value obtained for two different pairs of ROIs.
You need to contrast these results between two experimental conditions, or two groups of participants: compute this measure for all the subjects and then run a non-parametric statistical test to explore the significant differences between the two conditions or groups.
Ideally, you would obtain the same maps of statistical differences no matter which solution you selected for the computation of the coherence.
Which kind of connectivity estimation (coherence, PLV, Envelope, etc...) fits better with study design (interictal EEG connectivity) in your experience