Dear Helmut,
Thank you for the detailed explanations of this SPM change.
I can look into this issue again to make sure, but in our specific case this specific change seems responsible for most if not all differences between results from SPM8 versions 3042 and 6313.
We restored SPM8 3042 from our backups and run the calculations without and with the 32->33 fix.
* With vanilla SPM8 version 3042, we find differences of the order of 1e-2 in design matrices and contrasts compared to the ones obtained with the latest SPM8.
* With SPM8 version 3042 with the 32->33 fix, we find differences of the order of 1e-15 in design matrices and contrasts compared to the ones obtained with the latest SPM8.
Best,
Dimitri
|