[FieldTrip] Statistics of coherence (DICS)

[David Pedrosa]

Dear list,

I would be very grateful if someone might answer some questions on the statistics of coherence. Maybe first of all an outline to our experiment. We have measured a 128 channel EEG with additional peripheral signals from the most affected arm on 20 subjects suffering from tremor and the same number of healthy control subjects with a simple motor paradigm (activation condition). In another condition, rest (baseline condition) was also measured (we have about 20-35 trials per condition).

For the data analysis we have preprocessed the EEG in a rather standard way. For localizing the cortical sources we created a headmodel from individual MRI and the source reconstruction was obtained using the DICS beamformer (with the peripheral signal as 'refchan' and after computing a common spatial filter). 

Everything works out well and regarding the activation condition by itself the sources appear to be where we you would suspect it (motor areas) in all individuals. Of course coherence is much bigger during activation (about 1-3 dimensions). But otherwise coherence is never 0, not even in the rest condition (as to be expected). This makes us suspect a permutation test much more valid for testing the 'null-hypothesis' of interchangeable coherence between the conditions. Now we are struggling with the stats of the individual but also the group results. 

My first question is am I right that averageing over trials per condition is not necessary? And do I need to normalise the coherence after computing it with ft_freqanalysis? I was thinking of computing (activation  - baseline)/baseline. And, finally, regarding the stats. The idea is to look individually at differences between the conditions and in a second step perform a group analysis. But what to test if we are not assuming a null-hypothesis, or am I completely on the wrong track?

Thanks in advance.

Best,
David