how to use depsamplesF as an omnibus anova?

Manish Saggar manish.saggar at GMAIL.COM
Tue Nov 30 23:25:52 CET 2010


Dear all,

I have 81ch EEG data (no trials only continuous data) from two groups:
A and B. Each group was tested at three test points t1, t2,and t3. One
of the group received treatment and other one didn't (control).

Now in order to test the effects of treatment across time in each
group, we want to use nonparametric analysis. I am doing depsamplesF
test for three test-points in each group separately (followed by
Bonferroni Correction for doing 2 such tests) and if I find a positive
cluster in depsamplesF test then I will take that as a permission to
go in and test t1 vs. t2, t2 vs. t3, and t3 vs. t1 for that group.
Ideally I should not get any such cluster in control group. So my
first question is that does this approach sounds reasonable? For these
two depsamplesF test I am keeping the foi = [0.5 - 100], with
avgoverfreq='no'. Reason for this broad range frequency is that I am
keeping this level as more of an exploratory type analysis and want to
keep bonferroni corrections to minimum (not sure if I can assume
different freq bands as independent or dependent and thereby need
posthoc corrections or not?).

Now in addition to saying that three test-points differ (if positive
cluster), depsamplesF test also gives us <channel, freq>
pairs/clusters that differs in each group. In my case, it usually
gives only one such cluster and that usually covers 2/3 of the
channels and a freq range of 10-40Hz. Now since the sensitivity of
these depsamplesF might be low (due to huge foi range). Is it
reasonable to just take these depsamplesF as an indication that
something differs but ignore the cluster where things differ. Then do
t1 vs t2... and so on tests with defined low-freq bands (delta, theta,
alpha) and high-freq bands (beta and gamma) and using
avgoverfreq='yes' for those, so that the sensitivity can be increased.

Thus, using depsamplesF test as an omni bus anova and just find out if
something differs. If it does, then *not* take the cluster that
differs rather re do cluster analysis for three lower level tests (t1
vs t2... so on) with more sensitive freq bands and avgoverfreq='yes'
over all channels. Is it a reasonable approach?

I apologize in advance if something is not clear or if I am using
wrong terminology.

Thanks for all your great work and help.

Regards,
Manish

---------------------------------------------------------------------------
You are receiving this message because you are subscribed to
the  FieldTrip list. The aim of this list is to facilitate the discussion
between  users of the FieldTrip  toolbox, to share experiences
and to discuss  new ideas for MEG and EEG analysis.
See also http://listserv.surfnet.nl/archives/fieldtrip.html
and http://www.ru.nl/neuroimaging/fieldtrip.
---------------------------------------------------------------------------



More information about the fieldtrip mailing list