Print

Print
David,

Your model does not fit an interaction between subject and time, so you can't test it.  And even if you did, you won't be able to assess the significance of the subject-by-time interaction, since it would require (nSubj-1)*(nTime-1) terms and you'd have no degrees-of-freedom left to estimate the residual error.

If you constrain yourself to look for subject-specific linear time effects, you might have a chance of seeing something (you'll have low-ish DF, but it won't be hopeless).

You'll need to add nSubj-1=5 columns, having the time effect for each subject's scans (skip the one for the last subject, to avoid a rank-deficient design).  E.g. For the first subject the column would be
[ 1 2 3 4 0 0 0 ...... ]'
and for the second subject it would be
[ 0 0 0 0 1 2 3 4 0 0...]'
An F-test looking like
... 1 0 0 0 0 ...
... 0 1 0 0 0 ...
... 0 0 1 0 0 ...
... 0 0 0 1 0 ...
... 0 0 0 0 1 ...
on the corresponding columns will give you your interaction test.

I'm afraid I'm not sure how you could get the interface to create this model for you.   You could possibly drop the time effect, add a time covariate (that was simply [1 2 3 4 1 2 3 4 ...]) and then request a Factor-by-Covariate interaction, but then you won't have the full time factor fit. 

Hope this helps.

-Tom


On Wed, Feb 27, 2008 at 12:16 AM, Keator, David <[log in to unmask]> wrote:

 

Hi,

 

I have the following design matrix in SPM 5, six subjects with 1 FDG-PET per time point and 4 time points.  One subject only had 3 time points (ie., subject 6).  I'm wondering how to define the contrast to look at the interaction between subject x time.

 

Thanks,

 

Dave Keator

UC Irvine

 


--
____________________________________________
Thomas Nichols, PhD
Director, Modelling & Genetics
GlaxoSmithKline Clinical Imaging Centre

Senior Research Fellow
Oxford University FMRIB Centre