Hi SPM experts,
I had a (potentially simple) question regarding my design matrices (attached):
My task had six runs (three task blocks - let's say A, B, and C - that were repeated twice). I analysed this in two ways: first, by concatenating all the runs together using the spm_fmri_concatenate function (the left matrix), and second, by keeping the runs separate (i.e., defining multiple sessions at first level; the right matrix). As you can see, in the concatenated design, I modelled the same task conditions together (so that both 'A' blocks were modelled as one), as well as all the errors and movement parameters in the same columns (ending up with 99 regressors in total), whereas everything is kept separate in the other design (resulting in 240 regressors in total).
My problem is: I get very different looking results from these two designs, despite both containing the same data. Generally, I get very robust activations when using the concatenated design, and in some cases, there is nothing to see for the 'separate-runs' design at all. So, I was wondering if anyone could suggest a reason for such differences? Is this a simple power issue? I am aware of the potential problems with concatenating data, but it seems that the function I used takes care of the most obvious problems (e.g., high-pass filter, temporal non-sphericity). Any suggestions would be much appreciated!
Best regards,
Haeme
|