Table 3.
Number/Proportion of voxels deemed “Significant” by voxel-wise and cluster-wise thresholding, by analysis method
| Method | Voxel-wise stringency requirement |
Cluster-size stringency requirement |
Cluster numberΨ actual (Maximum) |
Mean # voxels “Significant” |
Percent of total voxels |
|---|---|---|---|---|---|
| Cluster-Threshold | P<0.05 | 3971 (each analysis) | 1 (1) | 25,964 | 1.4% |
| P<0.01 | 771 (each analysis) | 4–5 (5) | 9826 | 0.5 % | |
| P< 0.001 | 233 (each analysis) | 3 (50) | 1714 | 0.09 % | |
| RMRD | P<0.01 | 771 (full analysis) | 5 (5) | 8668 | 0.45 % |
| P< 0.001 | 233 (full analysis) | 3 (50) | 1300 | 0.068 % | |
| Bootstrap | >90 % of T-Tests | None | 5 (N/A) | 6784 | 0.35 % |
| >95 % of T-Tests | None | 5 (N/A) | 4265 | 0.22 % |
Ψ
In both cluster-threshold analysis and the RMRD method clusters are selected on both a voxel-wise and cluster size stringency requirement. There is at present no defined cluster-size stringency requirement for a bootstrapped process; five regions were selected empirically to test prediction accuracy