search

SnPM-toolbox / SnPM-devel

Statistical NonParametric Mapping, development version
GNU General Public License v3.0
25 stars 10 forks source link

Allow P-value cluster-forming thresholds with pseudo-t #44

Open nicholst opened 7 years ago

nicholst commented 7 years ago

Because it is such a pain, we might as well offer to convert P-values into corresponding T/F statistics based on nominal degrees of freedom when the Pseudo-t statistic is used. Variance smoothing messes up the DF, and hence these statistics won't have the exact same interpretation as usual... however, there is no known mapping between uncorrected P-values and pseudo-t values, and hence it leaves users in a muddle.

In FSL's randomise, we don't worry about this.

We should change our behavior, and provide the P-to-T/F service for users, simply issuing a warning that the conversion is approximate due to unknown effective degrees of freedom.

Relevant code starts here: https://github.com/nicholst/SnPM-devel/blob/master/snpm_pp.m#L518

cmaumet commented 7 years ago

@nicholst: is this implemented in https://github.com/nicholst/SnPM-devel/blob/master/snpm_pp.m#L534-L540?

nicholst commented 7 years ago

Oh... oops... yes.

nicholst commented 7 years ago

No.. but wait... that code assumes that the statistic image is T... not F... current code won't do something sensible with F statistic.

ACTION: Need to revise code to allow sensible clustering forming threshold (from P-value) when STAT == 'F'.