Hi all,
I have a fairly complex data learning set with multiple sessions and two parametric modulators of the same event [i don't orthogonalise]. In total, my conditions amount to ~140 unique regressors on top of which I build ~190 different contrasts.
In some rare cases, one parametric modulator fails to estimate and so a particular contrast might be empty for this subject [it happens very rarely and it would mean that my second level analysis will have one subject less, fair enough].
When I provide a contrast full of 0's or NaN's SPM crashes.
I am solving this by marking the failed contrasts, dummy estimating them with 1 at the first regressor and excluding them later but this is prone to error [it's the kind of hack I really hate to do], so I wonder if perhaps there is a way for SPM to still generate the contrast but make it full of NaNs and exclude it automatically at 2nd level?
Any more elegant solutions to this?
Thanks!
Ondrej
|