[FieldTrip] Statistics on freq data

Алексей Бенин albenin at mail.ru
Wed Feb 5 15:57:47 CET 2014

Dear Fieldtripers,

First of all big thanks for a wonderful tool and support!

I have some questions about cluster based permutation test on TF data.
I perform a between-trials experiment and want to search differences say 
on frequencies 10-30Hz and latencies 200-1200 ms from onset.
So I need to explore these ranges for possible differencies in powerspectrum.
When I use for example

cfg.latency          = [0.2 1.2];
cfg.frequency        = [10 30];

I find no significant clusters, and lowest p_value is very high, say 0,8.
But when I try to look at smaller time/frequency areas, like

cfg.latency          = [0.2 0.3];
cfg.frequency        = [15 16];

There can be one or two significant clusters with p_values<0.01.
So my question is: does the algorithm look for the whole specified TF window like
0.2-1.2s   10-30Hz and tries to find significant difference based on the whole length
of interval? This could explain why there are no significant clusters on bigger windows.
Such clusters appear when we shorten the toi and foi - it's much more probable to find differences
at smaller scales. 
So the next question is: what should I do to explore the whole number of frequencies and latencies?
Shall I run a loop with parameters like this:

Iteration 1
cfg.latency          = [0.2 0.3];
cfg.frequency        = [15 16];

Iteration 2
cfg.latency          = [0.3 0.4];
cfg.frequency        = [15 16];

... and so on?

Or maybe I do not understand something in basis of method? 
BTW, I've read all tutorials and appropriate papers like
Nonparametric statistical testing of EEG, but could not find the answer.

Thanks in advance,


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20140205/543c714c/attachment-0001.html>

More information about the fieldtrip mailing list