Dear SPM experts
We have a GLM model (Model 1) with 8 regressors (columns). Regressors 1-4
will be included as bilinear effects in the DCM model. Model 1 will be
used only to identify coordinates for individual VOIs.
For the DCM itself we would like to create a model with an 'input'
regressor. To do that we have added in Model 2 regressor 9 which includes
all the events from regressors 1-6. This has resulted in the betas of
regressors 1-6 & 9 to be 'not uniquely specified'.
To avoid this problem we have created Model 3 that does not include
regressors 5-6 (so that regressor 9 is not a linear combination of other
regressors in the matrix).
When comparing the maps of the contrast of regressors 1 vs. 2 we find that
Models 1 and 2 give similar maps, while model 3 gives a very different
map.
My questions are :
1) Why are the maps of model 1 and 3 (for the same contrast) so different?
2) Which model should we use in the DCM analysis: Model 2 (with the non-
orthogonality) or Model 3 (which gives entirely different results)?
Does it affect the results of the DCM, given that we only include
regressors 1-4 as bilinear effects and regressor 9 as the input in both
cases?
Thanks a lot
Tali Bitan
Haifa University
|