Dear Henrik, > we have run an experiment with 15 Schizophrenics and 15 healthy > controls. > Our real voxel size is 3.59x3.59x3.6 (0.6mm gap). > Our data are normalized to a voxel size of 3x3x3 mm. > > Now we want to compare the groups using different covariates in a 2nd > level analysis. > > Here our question: What kind of spatial smooting is appropriate? > Standard is 2-3 times voxel size. > For group analysis the literature says to use 12-20mm. > As we have found effects when smoothing with 12 instead of 6 mm we are > considering using 15 or even 20 mm. (It is difficult to compare the > effects varying the smoothing systematicallly because every smoothing > needs 1.5 Gigabyte!). > > Are there any principal arguments to stay with 12 or to use 15 or 20 mm? I can't think of any theoretical arguments why you can't use smoothing kernels between 12 - 20 mm. Concerning group studies, the 'optimal' smoothing filter is certainly a function of the quality of normalisation with respect to the underlying functional signal in each subject. If part of the underlying signal is rather focal in each subject, you will attenuate parts of the signal just because of the filter width of say 12 - 20 mm. So, smoothing in group studies is a trade-off story, which is definitely data-dependent: You must smooth given the inability to find transformations which will put your functional signal in homologous voxels for each subject, but you don't want to smooth too much because of the (possibly smaller) signal width. Having to choose a globally applied smoothing kernel for a group study I would go higher than twice the voxel size, i.e. as a first guess I had tried something between 9 - 12 mm as you already did. Viel Glueck, Stefan -- Stefan Kiebel Functional Imaging Laboratory Wellcome Dept. of Cognitive Neurology 12 Queen Square WC1N 3BG London, UK Tel.: +44-171-833-7478 FAX : -813-1420 email: [log in to unmask] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%