Dear Charlotte,
It's preferable to preprocess the data in a single batch (three sessions during slice timing, realignment)*, and also to set up a single first level model with three sessions, as a multi-session model also allows to construct contrasts across sessions on single-subject level. The beta estimates in the multi-session models won't be indentical to those obtained from three separate models, as some parameters are derived from all the volumes of a model, but dof is not really an issue. When looking at results within the first level model you will indeed have more dofs and thus detect more significant activations. But usually we're not interested in the first level statistics but just want to get some beta estimates and forward these into the second level statistics (with their own dofs based on factors, subjects).
* Even with atrophy rigid-body transformations should usually be sufficient to align one EPI session on another, except stuctural changes are really very strong (e.g. visible enlargement of ventricles over time). This might not hold for developmental studies (e.g. childhood vs. adolescence), but in that case it would also be preferable to align the data with additional non-rigid transformations (possibly indirectly via estimations derived from the structural volumes) within subjects first instead of separate segmentation/normalization onto some templates.
Best
Helmut
|