In the initial development of VBM it was observed that cluster-based inference produces an unreasonable number of false positives, due to calculation of the expected number of clusters depending on local variations ("non-stationarity") in the smoothness of the image (see http://www.fil.ion.ucl.ac.uk/spm/doc/papers/john_vbm_methods.pdf)
There is however an SPM toolbox for implementing Keith Worsley's correction for non-stationarity (http://fmri.wfubmc.edu/cms/software#NS), which should restore control over the false-positive rate (see Hayasaka et al. 2004, Neuroimage)
Perhaps other stats gurus can add to this if there is more up to date advice!
Hope this helps
On 24 May 2011, at 21:22, Deryk wrote:
> Is it appropriate to interpret the cluster level statistics provided in the results table for a VBM analysis?