 |
 |
Dear Marc > I have two questions on study design and contrasts. I am using SPM96. > > 1) Multi-study design main effects. > > I understand that SPM calculates intra-subject variability, and > therefore cannot appropriately assess main effects: except for the case > of a single scan per subject per group. (Perhaps this has changed?) This is correct. By use of a two-level procedure appropriate random-effect analyses are however easily implemented (in SPM99). > We have a PET [15]Ostudy in which an activation scan is repeated twice > in a control group and twice in a disease group. Is there any way I > can a) compare the mean's of the activation scans between the groups, Yes you can. The simplest way would be to use the implicit fixed effect analysis and test directly for a group x condition interaction with the appropriate contrast (assuming you have a basline condition as well), or simply compare the mean of the replications in a separate analysis (if you do not have a baseline). > and b) can this be extended to three groups? Must I calculate the mean > scans outside of SPM? If you do not have baselines than your inference is simply about a main effect of group (under the same condition). The mean enters as a subject-specfic measure and this can be created using SPM (see 'AdjMean' in the lower GUI panel). > 2) Applying a covariate of interest to a delta scan. We have an > activation and control scan for all subjects in a group. A > psychophysical score is calculated during the activation scan. Is it > possible to correlate the psychophysical scores with the > Activation-minus-Control sc ans? Again, should I calculate the > difference scan outside SPM? It would by simpler to test for a condition x psychophysical score interaction (assuming the score = 0 under basline and is mean corrected under activation) using the original data. This reduces to a simple main effect of score under activation. (using delta scans in SPM96 is tricky because of thresholding to define brain voxels - this will be changed in SPM99) With best wishes - karl %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%