Good afternoon!
I'm trying to analyze my data but I'm not sure which approach is the best:
I computed PSD for 4 bands (Delta, Theta , Alpha , Beta) per file
Then I'm trying to perform a permutation test but I have this doubt:
Once I've ran the test and before applying alpha threshold I export the result of the test to a Matlab var,
I checked the Matlab var named "pmap", I guess it contains the p-value for every measure (electrode in this case)
Once I've applied aIpha threshold I usually export the file to excel,
It seems like it automatically performed the test over all the 4 EEG bands, however, I would like to perform the permutation test on just one band (Delta) so when I was running the test I specified in the frequency the same Hz I set for Delta band (1.56 - 3.52), but the result I got is different:
when I check the "pmap" var the p-values are very different from p-values in the first file in Delta band (the file tested with 4 bands)
I don't get why?
I would like to understand the best approach for this testing, I followed tutorial but there is no example about psd comparing.
I'm not sure what correction should I apply: I guess it would be a good idea to select "Signals" since I have 19 electrodes being tested at the same time
The permutation tests are computed for each value in the input files independently: each signal, each frequency band and each time point gets a t-value computed independently from all the others. At this stage, it doesn't matter if you process one or four frequency band.
The number of frequency bands you are processing at once has an impact on the correction multiple comparisons. You will not get the same corrected p-values if you are considering one or four frequency bands at the same time. In this configuration, all the values are heavily dependent on each other, and you are interested in knowing what is significant in your entire file, it is probably a wise choice to keep all three checkboxes selected for the correction for multiple comparisons (time, frequency, signals).
If you knew from the beginning that you are only interested in one frequency band, the easiest would be to run the test only on this frequency band, either by extracting it first, or by using the option to select specific frequencies in the process options.
Otherwise, if you want to exclude the other bands but do not want to recompute the files, uncheck simply the checkbox "frequency" from the correction for multiple comparison, and for the selected frequency band it should apply the same correction as if it were tested alone.
Same goes for the electrodes: if you knew from the beginning (as part of your hypotheses for this experiment) that you are only interested in the power for one electrode: you could decide to run your statistics only only on this one
However, never run a first pass of statistics, identify the strong effects and re-run your stats only on this subset to get lower p-values.
