Dear SPM experts, 

I notice that there are two papers recently published about the statistical analysis in neuroimaging. The first is Eklund et al. “Cluster failure: Why fMRI inferences for spatial extent have inflated false-positive rates”. For response, Flandin and Friston wrote a paper of “Analysis of family-wise error rates in statistical parametric mapping using random field theory”.

In the latter study, Flandin et al. exhibited the results (Fig. 1 in the paper, attached Fig. 1) of cluster-level inference for two-sample t-test (two groups of ten random subjects, repeated a thousand times) with the Beijing dataset using a cluster-forming threshold of p < 0.001 and the SPM12 software (r6685). Five levels of spatial smoothing were evaluated (4, 6, 8, 10 and 12 mm isotropic Gaussian kernels) with four different regressors (see (Eklund et al., 2015) for details). We can see from this figure that FWE rates for clusterwise inference are around the 5% level.

However, the Fig. S2 (attached Fig. S2) in the former study, Eklund also had the results of two sample t-test and cluster-wise inference using a cluster defining threshold of p = 0.001, showing estimated familywise error rates for 4 - 10 mm of smoothing and four different activity paradigms (B1, B2, E1, E2), for SPM. But the FWE rates for clusterwise inference are around the 10% level.

I think there are only two differences for these analyses. First, Flandin et al. used voxel size of 3mm, but Eklund et al. employed voxel size of 2mm; second, Flandin et al. used SPM 12 but Eklund et al. utilized the SPM 8.

Does this difference bring about the different result in Fig. 1 and Fig. S2? I am a little confused about this issue. If so, the cluster level inference in SPM 12 is stricter than SPM 8? Or I am wrong in some way to understand these two excellent studies.

Any help would be greatly appreciated!

Best,

Feng