 |
 |
Dear Ralf, > I want to make a group analysis using a fixed effect model. Because of > the small sample sizes (7 and 4 subjects) the random effect toolkit is > not adequate (very small df). So I put all subjects in one big model > (one session per subject using the same model). Now my question: When > I make contrasts between the groups is it necessary to weight the > contrasts (11/2 = 5.5, contrast weights 1.4 and 0.8 for the individual > subjects in the two groups respectively, the contrast vector is > simplified 1.4 1.4 ... -0.8 -0.8 ...) or not (1 1 ... -1 -1) ? Is this > is a valid approach or is their a correction for different dfs > necessary and how can I perfom that ? In fact SPM99 will not allow this contrast because it is inestimable. This is because the group differences are already modeled by the session- or block-effects. One would normally proceed by collapsing the data from each subject into a contrast or mean and entering the 11 scans that ensue into a two sample t test (this would be a parameteric Random Effects analysis with 9 d.f.). You do have low d.f. but the answer to your question depends on intersubject variability and therefore a Fixed-Effect analysis is inappropriate. I hope this helps - Karl %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%