I’m wondering if you could provide some advice. I have two conditions (before and after a treatment) for an individual participant. There are between 60-100 trials I can use for each timepoint.
To test these statistically, I intended to do a paired contrast like spatio_temporal_cluster_1samp_test, using trials as the variance (i.e. setup X as trials x samples x vertices). However, there appears to be a time shift in the N1-P2 timing (around 25 ms or so), which means this isn’t really valid as a comparison. Do you have advice on the best way to statistically compare source reconstructions within an individual across two conditions please?
Thank you very much in advance and apologies for any missing details!
That depends on what kind of question you want to get a statistical answer to. Normally, the question is something like: is there a difference in the data between the two conditions? To which in your case the answer would be “yes”, given that the N1-P2 timing is different. You say “this isn’t really valid as a comparison” hinting that you wish to ask another question.
For example, perhaps you want to ask something specific about differences in N1 and P2 amplitude? In that case, I would not do a cluster permutation test. Instead I would define suitable time windows that nicely capture the N1 and P2. Since there is a difference in timing across the conditions, I would design different windows for the different conditions, keeping them the same length but shifted in time. Using these time windows, I would compute the mean signal in them as a measurement of N1 and P2 amplitude. Then finally comes a paired t-test (or permutation test if you want) to compare these values across the two conditions.
Thank you for the very quick reply, this is really helpful. Thinking aloud, as you rightly highlight, a difference given the timing difference would be yes, but wouldn’t get at necessarily what we are interested in - one concern of mine is there could be more mundane reasons for that aspect.
The main thing we want to examine is whether something has changed in the “network” properties of the brain (which may reflect amplitude changes) before vs after our intervention. These might not be at the timing of the N1-P2, but that seemed like a good place to start for sanity checking. Therefore, shifting the later session to have the timing of a known waveform (i.e. the N1-P2) match the former is a very good point. I assume at that point the spatiotemporal clustering function is then valid, as it will be implementing a permutation test between the conditions to examine differences in amplitude across the brain in an appropriate time-aligned fashion.