 |
 |
Hi John, In PALM, FWER is always present across voxels (even without "-corrmod" and "-corrcon"). The output always includes files named "fwep" that are FWER-corrected across voxels. If there are multiple contrasts and you'd like to correct across these, use "-corrcon" and files named "cfwep" will be produced. If there are multiple input modalities and you'd like to correct across these, use "-corrmod" and files named "mfwep" will be produced. If both "-corrmod" and "-corrcon" are used, files named "mcfwep" will be produced. None of these is Bonferroni -- PALM does't use Bonferroni in any circumstance, although in some particular cases, e.g., complete independence and exhaustive permutations, the permutation-based correction becomes equivalent to Bonferroni. For a one-sample t-test, 2 contrasts, and sign-flippings, the options you are using are fine. The -corrmod is only needed if there are more than one modality. Otherwise, the mfwep will be the same as fwep, and mcfwep will be the same as cfwep. Hope this helps! All the best, Anderson On 20 April 2017 at 17:33, John Zackary <[log in to unmask]> wrote: > Dear all, > > I’m working with PALM and although I read the great paper “Non-Parametric > Combination and Related Permutation Tests for Neuroimaging” my knowledge > of statistics is not enough. Please, could you help me to solve some > questions about p-value corrections? > > Which is the FWER procedure to correct for multiple comparisons? It's not Bonferroni, > what is performed when the arguments corrcon and corrmod are used? Does > sign-flipping account for multiple voxels? > > I’m not sure if this is the answer “Each test is applied separately, > statistics converted to a z-statistic based on their asymptotic behavior, > and the distribution of the extremum used to obtain adjusted p-values for > all in a closed testing procedure using a union–intersection test”. > > I’m performing one-sample t-test, 2 contrasts, with sign-flippings with > the arguments -corrcon –corrmod -ise -save1-p -T -n 5000 > > Thank you, > > Best > > John >