Wolfgang Weber - Fahr wrote:
> Dear SPM Experts,
> I still have some basic questions about the Contrasts ins SPM99. I'm
> looking at a quite simple paradigm with the conditions [A B R] and would
> like to test for Voxels where the activation for A is significantly
> higher than for B.
> If I look at the results for the Contras 1 -1 and plot the
> event/epoch-related response for these voxels, all plots show me
> activation for A and deactivation (negative response function) for B. I
> just wonder what happened to the voxels where there is activation for
> both contrasts but more for A....
> Is this a problem with my data or do I understand something wrong about
I don't think there is necessarily a problem with your data. It might simply be a question of power.
Obviously the largest differences in your contrast [1 -1] will be found in the cases weher there is an
activation in A and a deactivation in B. If your power isn't very high (i.e. if you have relatively small
no. of events) it may be that you are unable to detect the more subtle differences resulting from an
activation in A and a smaller activation in B. Try to lower the threshold (you may have to lower it in
the SPM-defaults to ensure data are saved for plotting) and have a look at some voxels with smaller
z-scores and see what you find.
You should also be aware that an "activation" or a "deactivation" is always relative to some baseline
which may be more or less well defined. If you are using rapid stimulus presentation (short SOA) without
null events it will be less well defined, and it will be very difficult to determine between activations
and deactivations. In that case the interpretation of a positive finding in the contrast [1 -1] can be
larger activation in A than in B, or less deactivation in A than in B, or anything in betweeen.
If R in your design denotes null events you are in a better position and the question of activations or
deactivations should not be determined from plots of event-related responses of event types A and B, but
rather be based on the [1 0 -1] and [0 1 -1] contrasts.
> Is there any other Contrast that would test for A>B? For Example would
> something like [+2 +1] make any sense at all?
No, not really.
> What would be the apropriate Contrast for three conditions A B C that
> tests for A > B & C ?
The contrast [2 -1 -1] tests if the activity in A is larger than the mean activity in B and C.
The conjunction between contasts [1 -1 0] and [1 0 -1] tests if the activity in A is larger than the
activity in B, AND is larger than the activity in C.
I suspect the latter makes a little more sense.
> Finally I've got a completely different problem with the SPM
> Using the following parameters the routine constantly crashes before
> doing anything:
> 1 Session, 300 Images, Ascending ,TR=3.2, TA=2.4, Ref.Slice=21 (Top)
> The Error Message is:
> ??? In an assignment A(matrix,matrix) = B, the number of rows in B
> and the number of elements in the A row index matrix must be the same.
> Error in ==> /usr/local/spm99/spm_slice_timing.m
> On line 255 ==> stack(nimgo+1:end,g) =
> ??? Error while evaluating uicontrol Callback.
> Could anybody give me a clue about what's going wrong here?
I'll pass on this one.
> Thanks a lot in advance!
> Wolfgang Weber - Fahr
Good luck Jesper