> as for the extent threshold (k), in the case of SVC on peak level should we leave = 0?
With regard to k in combination with peak level correction, the k is really an additional and arbitrary threshold. See https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=spm;4c8f7781.1402 and responses.
And finally and for sake of completeness, with regard to k in combination with an uncorrected threshold, e.g. something like .001 and k > 10, please see Bennett et al. (2009, https://dx.doi.org/10.1093/scan/nsp053 ):
"Many authors justify this approach by referring to the results of Forman et al. (1995), who examined clustering behavior of voxels in fMRI. The results of Forman et al. suggest that a threshold of P < 0.001 combined with a 10-voxel extent requirement should more than adequately control for the prevalence of false positives. [...] The problem is that this threshold is specific to the parameters of their dataset, and may be inappropriate in other datasets. [...] The cluster size criteria can change quite substantially from dataset to dataset."
You will nonetheless also come across recent papers that refer to a certain k from previous papers without proving that it was a good choice for that study nor that it is a good choice for the current.
Best
Helmut
|