Dear all
We have collected ASL data on drug and placebo administration at several time-points (different sessions), within-subject design (drug and time, both as within subject factors). For each time point i have run paired t tests between drug and placebo, using globals as a covariate (Ancova option) to identify drug effects on cerebral blood flow. I have also run some analysis where i calculated difference images first (drug-placebo) for each time point and then have run one-sample t test (controlling for globals of the difference images in the same way), where i was expecting to get similar results to the ones achieved in the paired t tests.
Surprisingly, i could not achieve similar results between paired t tests and one-sample t test (on the difference images), unless i enter the globals as a separate covariate without mean centering. Results between paired and one sample t test on the difference images are exactly the same when not controlling for globals. Adding manually the globals as a covariate, with or without mean centering, do not change the results of the paired t test. However, i got very strange results on the one sample t tests when i ask spm to calculate and add globals as a nuisance variable (using an explicit grey matter mask), when i add globals manually in the user option for globals (both with spm centering the covariate to the mean automatically) or when i add a covariate manually with mean centering. If i enter the globals manually as a covariate and do not ask for overall mean centering, then i achieve results similar to the ones i got in the paired t tests.
Did someone come across with the same issue or may advice on the use of centering to the mean in difference images?
Thanks in advance
BW
|