> I am writing with some specific questions regarding the creation of a
> between subject analysis with informed basis set (canonical HRF +
> temporal derivative + spatial derivative).
> A majority of the process below has been taken from ch. 27 in the SPM
> 5 manual. However, there is no description of how to perform analyses
> between-group. I will outline my process, and would appreciate any
> feedback on its validity!
> Two groups, BPD (B) and CTRL (C)
> First step was the creation of a basic contrast for each participant
> (first level).
> This resulted in 3 con images for each participant (canonical = c,
> temporal = t, spatial = s).
> I then created a 2x3 full factorial design.
> 1^st level = Group, independent, unequal variance
> 2^nd level = Basis functions, NO independent, unequal variance
> 6 cells: Each cell received on con image from each subject as follows
> Cell # (level 1, level2) = Xx
> Cell 1 (1, 1) = BPD-Canonical x 14 subj
> Cell 2 (1, 2) = BPD-Temporal x 14 subj
> Cell 3 (1, 3) = BPD-Spatial x 14 subj
> Cell 4 (2, 1) = CTRL-Canonical x 15 subj
> Cell 5 (2, 2) = CTRL-Temporal x 15 subj
> Cell 6 (2, 3) = CTRL-Spatial x 15 subj
> This then creates a SPM.mat file and several contrasts are
> automatically defined.
> I have examined between group differences two ways:
> SPM created main effect of group: F [1 -1 1 -1 1 -1]
> - This data looks excellent, activations in predicted regions, however
> I have concerns about correctness.
Unfortunately this contrast (created by SPM by default, but without
knowledge of your particular design) is meaningless. It is contrasting
different basis functions, which is like comparing apples and pears. In
general, be careful of the default contrasts created by SPM5 - they may
not be appropriate (and indeed, if the three basis functions were in
fact three different, comparable conditions, I would still have thought
the main effect of group would be [1 1 1 -1 -1 -1], at least if the
design matrix is ordered the same way that you ordered your cells above).
Anyway, when one factor is basis functions, the contrast you want to
compare groups (in terms of any difference in the shape of the HRF, as
captured by these three dimensions) is the F-contrast:
[1 0 0 -1 0 0
0 1 0 0 -1 0
0 0 1 0 0 -1]
> T-Contrast of Canonical Fx only: T [1 -1 0 0 0 0]
> - This data looks OK. Power is lower than I’d expect. Activation not
> in predicted regions.
This contrast (and the default SPM one above) confuses me, because if
the design matrix is ordered as you ordered the conditions above,
shouldn't the group contrast on canonical HRF only be [1 0 0 -1 0 0] -
or are the factors in the SPM design matrix actually rotated with
respect to what you describe?
> What I am trying to do is analyze areas of functional different
> between the two groups for the same contrast. Directionality is not
> paramount at this point, because I am looking for ROIs to seed further
> Am I correct to use main effect of group in this case? How does SPM
> combine these three functions in this case?
> Why do my F-tests of main effect look so different from my T-contrast?
> Is a large portion of the variance being accounted for by the spatial
> and temporal derivative, and is this in turn not being picking up by
> the T-contrast created post factorial design?
In general, you might find this webpage helpful to answer your questions
(though it refers to SPM2, the same logic applies - indeed, the results
should be very similar if not identical in SPM5):
Dr Richard Henson
MRC Cognition & Brain Sciences Unit
15 Chaucer Road
CB2 7EF, UK
Tel: +44 (0)1223 355 294 x522
Fax: +44 (0)1223 359 062