I would like to do some statistics on source localization for different frequency band, and I feel I don't find the proper way to do it. My aim is to realize a statistical test on a group of participants (n=18) in a given condition to be able to anatomically localize in the source space the activity I've found significant in an other test in the electrode space, and for different frequeny bands (alpha, beta, gamma). Actually, I did an average of participants activity that gave me something very coherent with the firts assuption. But I'm looking for a way to "prove" it. Is there any solution for that? I guess I should do a with subject test but I cannot succed doing it (for instance, I cannot erd/ers transform a hilbert map with alpha, beta and gama).
If there are some specific steps you cannot execute but do not understand why, please be more specific (add screen captures of your database, the process options and the error messages).
Thanks françois for your very quick and clear reply,
A problem I encountered so far is I ran the hilbert transform for alpha, beta and gamma, for each trail (power), so I get for each subject ant trial a node called "power, timebands". To go further, I want to standardize it (erd/ers) but I have the following error message : "cannot process values averaged by time band". Should I normalize a step before? I thought normalization should be aplied after time frequency décomposition.
I' cannot go any step further in my analysis, if someone has any helps for this.
I've done source and hilbert transformation (10-11 hz frequency band). I wanted to test the time -frequency decomposition against baseline but I get this error that I don't understand (see screen capture for 3 subjects).
More globally, I did not find in the tutorials the statistics I would like to apply. I have 18 subjects for one condition. I did sources and then frequency décomposition. I would like now to find where in the source space is the main source activation (ex: parietal post, calcarine sissure,..). But I need to find it statistically. Actually I had a comment from a reviewer for a submitted paper that specified that this localisation could not be done with a subject average only but need statistics, and I don't find the way to answer that.
Here the reviewer's comment, it might help to understand;
"""What are the statistics supporting the multi-subject anatomical analysis? Just averaging is simply not acceptable. Please provide pertinent parametric/non parametric statistics of the population for the source reconstruction."""
I have the following error message : "cannot process values averaged by time band"
If you average your time-frequency maps by time bands, you usually do not have enough values to estimate the mean/std over your baseline. So do not use this option "time bands" in the configuration of the Hilbert process if you need to normalize your time-frequency maps.
I wanted to test the time -frequency decomposition against baseline but I get this error that I don't understand.
There is something wrong with your baseline... if you have computed this with the option "time bands", try without. Same issue here: if you have only two values during your baseline, you cannot really estimate the baseline mean and variance you need for the statistic you want to compute here.
I have 18 subjects for one condition
You need to come up with a contrast: stimulation vs. baseline, stimulation vs. rest, etc.
I would like now to find where in the source space is the main source activation (ex: parietal post, calcarine sissure,..). But I need to find it statistically. Actually I had a comment from a reviewer for a submitted paper that specified that this localisation could not be done with a subject average only but need statistics, and I don't find the way to answer that.
It is also a reasonable approach to do the statistics at the sensor level, find when you have a significant difference from your baseline, then just show where the difference of averages localize in source space, as a complement of information. This way, you run your statistical test only on your real data (the recordings) and do not have to question the validity of the full processing chain (forward model+inverse model+post processing+test on source power).
