Figure 6. Group maps made from various size of smoothing kernels applied to data for SVM-searchlight.

When smoothing is applied to the SVM-generated maps, the resulting random effects maps do not become similar to the GNB-generated map. Instead, the resulting maps simply look like slightly smoother versions of the random effects map generated from unsmoothed SVM images. This increased number of clusters makes the SVM analyses bear even less resemblance to the GNB analyses, regardless of the level of smoothing applied to them. Group-level random effects maps are shown at p<0.001 uncorrected without any cluster-level thresholding (k = 0).