Dear SPM experts,
I am now cofused by the different results of the following
two types of analyses:
(a) within-subject analysis performed on the pooled data from 10 subjects,
setting the contrast [-1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
[0 0 -1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0] ...and so on
(b) a single-subject analysis performed on the data from the corresponding
one subject,
setting the contrast [-1 1]
As far as I have experienced, activation revealed by (a) is smaller than
that revealed by (b).
In an extreme case, significant activation observed in analysis (b)
completely disappears
when I switch to analysis (a) even though the same significance level (and
the same subject
of course) is chosen.
Here, my questions:
(1) Does this indicate the inhomogeneity of error variance across those 10
subjects?
(2) If so, is it inappropriate to pool those data together and perform
multi-subject analysis?
Or can I still believe that GLM is robust against such violation of
homogeneity?
The details of my data and analysis are as follows:
blocked design fMRI
a simple Rest-Task paradigm
TR: 4s
# of subjects: 10
# of volumes per subject: 60
# of volumes per epochs: 10
order of epochs: Rest Task Rest Task Rest Task
delayed boxcar
high pass filter cut-off period: 160s
global normalization: scaling
Any advice would be appreciated.
Regards,
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
