Dear Joerg, >Dear SPMers, > >I'am planning a random effect analysis in SPM99b (10 subjects x 7 >conditions). The rough outline is as follows (I hope I understood that >right): > >1st level >- define and estimate a model for each subject >- calculate contrasts for each subject independently > >2nd level >- for a given contrast, enter the relevant contrast image of each subject >into a one sample t-test > >The question now is, when or what should be normalized? Should we >normalize all functional images before the first level of analysis or is >it sufficient to normalize only the contrast images before entering the >2nd level of analysis? What is the difference between both procedures? > In principle you are right, both procedures should be identical. However, from a practical perspective you will have problems if normalising the contrast images due to the masking. When calculating the statistics SPM will perform a masking, meaning that calculations are performed only for voxels likely to be within the brain. Hence, the contrast images will contain zeros immediately outside the brain. When the normalisation tries to interpolate the values close to that edge it will interpret zero as missing data, and set the value to zero. I.e. the volume of "data present" will erode by one voxel (for trilinear interplotaion, more for sinc), meaning that you will not include the entire brain in the analysis. For this reason I suggest normalising the individual functional images prior to the fixed effects (first level) step. >And a similar question concerning smoothing: I carried out the 1st level >analysis with smoothed (non-normalized) functional images. Should the >contrast images be smoothed again? > Same concern applies here. Smoothing is a local neighbourhood operation and masking will cause similar problems as for interpolation. >Thank you for any information, >Joerg > Good luck Jesper Jesper Andersson Wellcome Dept. of Cognitive Neurology 12 Queen Square London WC1N 3BG phone: 44 171 833 7484 fax: 44 171 813 1420 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%