 |
 |
At 22:50 31.05.02 +0100, you wrote: >We have run an experiment where each subject goes through three separate >runs. We were wondering if we should combine the three runs into >one "session," or if we should count each run as a separate "session" in >our analysis. We are under the impression that doing a separate "session" >for each run would be giving the program the impression that we have run >three times as many subjects as we really have. Could somebody please let >me know the best way to do this? Hi Eric, I assume that you are using fMRI and that you interrupt the physical scanning process after each run is completed. If that is the case, then I would recommend to model the three runs as three separate sessions in your design matrix. I am certainly not the world's greatest expert on this topic and happily accept any corrections, but from my point of view, you can encounter various problems if you pretend that different sessions are one. For example, data from different sessions can have different frequency structure or different autocorrelation structure and this is important for temporal filtering (see spm_filter.m which performs filtering in a session-specific fashion). Another thing that comes to my mind is slice timing where SPM implicitly assumes that within a session the acquired time series is sampled from a "continous" process - this assumption is violated if you concatenate three time series from different runs between which the the acquisition has been interrupted. These are just two examples and there may be more important points that I am missing. I do not know whether, in practice, you would get significant errors if you combined the three runs into a single session but from my point of view it does not sound like a good idea. Just in case that you should worry about loosing df's by modelling each run as a separate session: remember that this is no worry if you are striving for a random effects group analysis - in this case, the df's depend on the number of contrast images entering the second-level analysis, not on the structure of the individual design matrices. Hope this helps, Klaas ________________________________________________ Klaas Enno Stephan Institute for Medicine (IME), Cognitive Neurology Group Research Centre Juelich, 52425 Juelich, Germany phone +49-(0)2461-61-4007 / fax +49-(0)2461-61-2820 [log in to unmask] / [log in to unmask] http://www.hirn.uni-duesseldorf.de/~klaas/