Below are the log(PP)-plots and Bland–Altman plots. The shaded regions in the PP indicate the standard deviation of the 100 realisations. The shaded areas in the BA plots indicate the standard deviation of the difference between each method and the respective result with 50k permutations.
| Few permutations |
Negative binomial |
Tail approximation |
No permutation |
Gamma approximation |
Low rank completion |
|
|---|---|---|---|---|---|---|
| Univariate | ||||||
| Uncorrected | P|B | P|B | P|B | P|B | P|B | —|— |
| Corrected | P|B | — | P|B | — | P|B | —|— |
| TFCE unc. | P|B | — | P|B | — | P|B | — |
| TFCE corr. | P|B | — | P|B | — | P|B | — |
| Cluster E | P|B | — | P|B | — | P|B | — |
| Cluster M | P|B | — | P|B | — | P|B | — |
Below are the log(PP) and Bland–Altman plots for the distribution of the maximum statistic (minimum for the Wilks' \(\lambda\) and Tippett) after 100 realisations. The 95% CI are also plotted. Note that, differently than the table above, since there is just one set of realisations, it isn't possible to compute the standard deviations, so only the CI based on the binomial distribution are appropriate.
| Few permutations |
Negative binomial |
Tail approximation |
No permutation |
Gamma approximation |
Low rank completion |
|
|---|---|---|---|---|---|---|
| Univariate | ||||||
| Maximum | P|B | — | P|B | — | P|B | —|— |
| TFCE max. | P|B | — | P|B | — | P|B | — |
| Cluster E max. | P|B | — | P|B | — | P|B | — |
| Cluster M max. | P|B | — | P|B | — | P|B | — |
Below are the histograms (with standard deviation bars), as well as the error rates at \(\alpha=0.05\), with the respective 95% CI (using the Wilson method and the respective number of permutations).
| Few permutations |
Negative binomial |
Tail approximation |
No permutation |
Gamma approximation |
Low rank completion |
|
|---|---|---|---|---|---|---|
| Univariate | ||||||
| Uncorrected | Hist | Hist | Hist | Hist | Hist | — |
| TFCE unc. | Hist | — | Hist | — | Hist | — |