Dear Ryan,
In response to your questions:
1. Yes, the best way is to define parametric modulators in the design matrix. Unless you have adjusted the extracted data for the occurrence of events per se, you will need both the non-modulated and the modulated regressors as inputs somewhere in your model.
2. This could be an appropriate way, but it fully depends on what your SPM results tell you about which regional effects are present. For example, if the SPM tells you that one region is correlated with one parametric modulator and a second region with another, then you might be testing which of the afferent connections to these two regions are modulated by the respective inputs.
3. For historical reasons, SPM adds 32 time bins at the beginning of each regressor. I think this was done because a long time ago some people wanted to model the onset of events before the actual scan started. I don’t think it has any use these days but it is still in there. DCM removes these 32 time bins.
Best wishes,
Klaas
________________________________
Von: Ryan Haynes <[log in to unmask]>
An: [log in to unmask]
Gesendet: Mittwoch, den 12. Mai 2010, 22:20:59 Uhr
Betreff: [SPM] DCM Input specification
Hi,
I have three questions pertaining to DCM in SPM8.
(1) When specifying the inputs for a DCM analysis, what is the
appropriate way to specify inputs that vary over different levels?
Should the first level design matrix contain parametric modulators for
certain factors? If so, are both the non-modulated stick functions
from (Sess.U.u(33:end,1)) and the parametric modulators be included
(Sess.U.u(33:end,2)) or would it be just the parametric modulators?
I'm leaning toward the latter but I'm not sure if the non-modulated
stick functions are needed for part of the DCM analysis.
(2) Consider a situation where an event varies along more than 1
dimension. Let's call the event A and the dimensions 1 and 2. I could
then create a design matrix with all events A and two parametric
modulators for A1 and A2. (This would make three columns A A1 and A2)
Would it then be correct to do something like specify the inputs A1
into one region and then the inputs A2 into another region even though
the events occur at the same time. The hypothesis to be tested would
be that dimension 1 is processed by one region and dimension 2 is
processed by another. Is this the appropriate way to set something
like this up?
(3) Why are the first 32 time bins dropped from the time series as can
be seen in the code of spm_dcm_specify:
U.u = [U.u Sess.U(i).u(33:end,j)];
Thanks in advance for your help.
Ryan Haynes
