 |
 |
Hi Ciara > When looking at 2nd level between group results is it valid to set the threshold to p=0.001 instead of FWE and FDR which gives an empty brain > and then report clusters from the whole brain results table instead of the voxel? if so, the clusters say they are corrected but how? and at what level? The short answer to your question is that cluster-level correction is a perfectly valid way to correct your data. The long answer: Historically (i.e. through SPM5) FWE and FDR both attempted to control for false-positives at the voxel level (although now there is an implementation of topological FDR, but I'll ignore that for now). The cluster- and set-level p values listed in the SPM results table take into account the spatial extent your clusters (identified using your voxelwise p value) relative to the overall smoothness of the data to determine how likely these clusters are to have occurred by chance. In other words, the cluster-level correction takes into account the spatial distribution of your signal, whereas voxelwise FWE and FDR do not. To come up with this p value, SPM needs some estimate of how likely clusters are to occur by chance. This is done using random field theory; there are lots of resources for this on the SPM bibliography page: http://www.fil.ion.ucl.ac.uk/spm/doc/biblio/Keyword/RFT.html There is also a more general-audience paper from Keith Worsley: Worsley KJ (1996) The geometry of random images. Chance, 9, 27-39. http://www.math.mcgill.ca/keith/chance/chance3.pdf and slides from Will Penny: http://www.fil.ion.ucl.ac.uk/spm/course/slides05/ppt/infer.ppt Finally a recent discussion of FDR might also be relevant: Chumbley JR, Friston DJ (2009) False discovery rate revisited: FDR and topological inference using Gaussian random fields. Neuroimage 44, 62-70. http://dx.doi.org/10.1016/j.neuroimage.2008.05.021 Hope this helps! Jonathan