[Eeglablist] bootstrap statistics with averaged data?

[Adrian Guggisberg]

Hi

I have multi-trial data from 15 subjects which each performed a
repetitive task. Now, I would like to use the timef() function to find
out whether significant spectral power changes occurred during the task
for all subjects. Timef() provides bootstrap statistics for multi-trial
data. My question is, how do I have to deal with the subject's average.
I can think of 3 different possibilities:

1. I calculate mean log(powers) for each subject (the first output of
the timef() function), and perform a bootstrap using these mean values
of each subject. Is this allowed and feasible?

2. I calculate log(powers) for each subject, and use the individual
bootstrap to set insignificant power changes to 0. Afterwards, I
calculate the grand average of the "bootstrap-weighted" data. However, I
figure that this way would introduce a considerable bias, and this
possibility does not seem to yield valid information about the overall
significance of power changes.

3. Bootstrap statistics can only be performed on individual multi-trial
data, and other statistics (which?) have to be used in this case.

Does anybody have experience with this problem? Thank you very much for
your support.

Adrian Guggisberg, MD
Neurologische Klinik
Inselspital
CH-3010 Bern
Switzerland
aguggis at swissinfo.org