Dear RiK / SPMers
I am probably missing something obvious, but if we use an F contrast, how
does this give us any direction of change? Is it because it is acting as a T
contrast because we have two measures. Therefore is a T contrast ok to use
with this model?
Also, if we have nuisance variables, would an ANCOVA be appropriate?
Thanks
Gary
Doron -
SPM does not separate error into between- and within-subject components. So
I would do the following.
Take a subject and use ImCalc to select the scans corresponding to their two
timepoints. First, evaluate the function "i1-i2", and call this, say,
"dif_grp01_sub01" for group 1, subject 1. Then select the same scans again,
but evaluate "i1+i2", and call this, say, "com_grp01_sub01".
Repeat for each group/subject. (You can automatize by calling
spm_imcalc_ui.m from the command line, if you like). Now, if you had 2
groups of 12 subjects, you should have 2 x 12 = 24 images starting "dif_*",
and 24 starting "com_*". Then, for the ANOVA main effect of time, choose a
"two-sample t-test" in the Basic Stats option of SPM2 (i.e, unpaired
t-test), and enter the 12 "dif_*" scans for group 1 as the first sample, and
the 12 scans for group 2 as the second sample.
Then evaluate the F-contrast [1 1]. For the ANOVA time x group interaction,
use the same design matrix as above, but now test the F-contrast [1 -1]. For
the ANOVA main effect of group, create a new "two-sample t-test", but enter
the scans "com_*" for each group, and evaluate the F-contrast [1 -1].
Hope this helps Rik
|