[FieldTrip] Second-level one-sample t test with clustering
Eelke Spaak
eelke.spaak at donders.ru.nl
Thu May 16 10:07:47 CEST 2013
Dear fellow FieldTrippers,
At the moment, I want to do second-level source statistics on my data
(but the question is the same for sensor statistics). I have computed
two-sample t-scores for each subject, between the trials in conditions
A and B.
Now, the appropriate per-voxel test to do across subjects would be a
one-sample t-test. This is rather easy to do, but I would like to use
cluster permutations to correct for multiple comparisons. My idea is
to generate fake data of all zeroes, the same size as my t-score
descriptive data per subject, and then do a paired-samples t test
between the real data and these zeroes (using ft_statfun_depsamplesT).
The voxel-level t statistic of data vs 0 should be the same as a
one-sample test of course. However, I am not sure the permutations of
the data would give me a valid distribution of the cluster statistic.
The test for the real data would be (comparing column 1 to column 2):
dataA 0
dataB 0
dataC 0
dataD 0
whereas a permutation might look like so:
0 dataA
dataB 0
dataC 0
0 dataD
Intuitively, I think this makes sense, as the t score of 0 vs data
will be the negative of t(data vs 0), I should get an appropriate
randomization distribution. But I'm not completely sure of this.
So, my question: does anyone see a problem with my approach? Or is
this a valid way of cluster-correcting a one-sample T test? If not,
what would be the right way to proceed?
Many thanks,
best,
Eelke
More information about the fieldtrip
mailing list