 |
 |
Dear Sarah, You should correct for both, but as in fMRI the correction for the number of voxels can be done in SPM based on Random Field Theory (FWE correction). In practice I suspect that a threshold FWE-corrected for the whole brain and also divided by 10 will be extremely conservative and you won't get anything. So you should think how you can use prior knowledge to limit your test number e.g by limiting your test in advance to some VOIs and maybe also only looking at the most interesting time windows. Alternatively, you can do your ANOVA at the sensor level corrected for both space and time and then only reconstruct the significant peaks and use an uncorrected threshold at the source level or only report the peaks. See also http://www.ncbi.nlm.nih.gov/pubmed/23046981 http://www.fil.ion.ucl.ac.uk/spm/course/video/#MEEG_Stats Best, Vladimir On Mon, Dec 10, 2012 at 9:19 PM, Sarah Vanhoutte <[log in to unmask]> wrote: > Dear SPM experts, > > I would like to ask your advice on the statistical analysis of EEG source > reconstructed images, more specifically on the p-value to be used to decide > whether a difference is significant or not. > > We used the normal source imaging approach in 10 different time windows. The > reconstructed activity was statistically analyzed per window with ANOVA to > find possible differences between different groups of patients. > > Do you have guidelines to choose an appropriate p-value threshold to correct > for multiple comparisons? Should I only correct for the number of ANOVA?s > that we did? Or should I correct for the number of EEG dipoles, like what is > done in fMRI research for the number of voxels? > > Thanks a lot for your answer. > > Kind regards, > Sarah Vanhoutte > > > -- > Doctoral researcher > Speech-language therapist > Ghent University (Belgium)