 |
 |
Dear Mateja de Leonni Stanonik, >How come that SPM can assume independence among voxels in order to >determine those which are activated at a certain level of significance >when performing multiple comparisons? Are these observations truly >independent of each other, I think that they are linked. The correction for multiple comparisons does not assume independence among voxels. Indeed, for the correction for multiple comparisons to be valid, the data should be spatially smoothed, in which case the voxels are definitely not independent. SPM calculates the number of resels, which can be thought of as the number of independent comparisons (across space) that have effectively been made. Then the correction for multiple comparisons is based on this number of resels. See http://www.mrc-cbu.cam.ac.uk/Imaging/randomfields.html for Matthew Brett's introduction to Random Field Theory, the statistical theory behind the correction. >Is this why analysis is also done at the cluster level, i.e. to aggregate >cases to a higher level? Doesn't one sacrifice a lot of statistical >power, making a Type II error more likely? The analysis can be done at the voxel level or at the cluster level, according to taste. This doesn't have a direct relationship with the correction for multiple comparisons, which can be done at the voxel level. There is not necessarily a loss of statistical power moving to the cluster level. Indeed, if you had a very wide cluster, with low differential BOLD responses across this wide area, then a cluster-level analysis might actually be more sensitive. >Thanks, No problem. Best of luck, Richard. Dr Richard Perry, Clinical Lecturer, Wellcome Department of Cognitive Neurology, University College London, Gower Street, LONDON WC1E 6BT Tel: (020) 7679 2187 Fax: (020) 7679 7316 E mail: [log in to unmask]