[FieldTrip] Comparing/contrasting ft_freqstatistics and eeglab's bootstat
Max.Cantor at colorado.edu
Thu Oct 26 20:12:45 CEST 2017
This is not a fieldtrip question per se, but I'm doing something in eeglab
and I was wondering if anybody could comment on whether what I'm doing is
comparable to fieldtrip's cluster permutation statistic. I'm attempting to
create a statistical mask for an event-related spectral perturbation array
(specifically a morlet wavelet ersp).
The dimensions of the ersp are log-scaled frequency (and where number of
cycles increases as frequency increases), samples, and channels. This
matrix is the grand average across subjects, and the difference between two
conditions. For each subject, each channel, and each condition, the ersps
were baselined. In other words, the data are differences in power from
baseline and between conditions, in units of decibels.
I run the following inputs through bootstat:
* [rsignif rbot] = bootstat(permute(g_ersp, [2,1,3]), 'mean(arg1,3);',
'alpha', 0.01, 'dimaccu', 2, 'naccu', 1000);*
Where the rsignif output is a freq x 2 array which I use as the statistical
mask, g_ersp is the ersp matrix I've been referring to, 'mean(arg1,3)' is
the function, alpha is alpha, dimaccu is the dimension to shuffle, and
naccu is the number times to reshuffle.
This averages across channels (the channels are an ROI so this is what I
want), shuffles across samples 1000 times, and tests for significance at
alpha = 0.01. It is not testing against a baseline as I understand
ft_freqstatistics to do. I use rsignif as an ersp statistical mask, and
when I included the baseline vector in bootstat, it failed to mask
anything. I think this is because I had baselined the ersp prior to the
statistic, so literally any power tested against an empty baseline window
was going to be significant. Running it in this way without testing against
a baseline, I get "sensible-looking" maskings, but it would be nice to get
external confirmation that what I'm doing is methodologically sound, and
that I am correctly interpreting my statistic conceptually.
I have used ft_freqstatistics in the past and would like to frame this
bootstat statistic in a similar manner, which is why I'm asking here. Also,
if I am misunderstanding my statistic, advice either on how to properly
implement ft_freqstatistics-like cluster permutation statistics in this
bootstat function, or alternatively how to convert my ersp matrix in such a
way as to be usable with ft_freqstatistics, would be appreciated.
Cognitive Neuroscience of Language Lab
University of Colorado Boulder
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the fieldtrip