 |
 |
Hi Jodi, I replied this to the HCP mailing list earlier today. For completeness, here it is the same answer: By default PALM saves p-values directly, not 1-p. If you'd like 1-p, use the option "-save1-p". Better than that, perhaps, is to use instead the option "-logp", that puts the p-values in a log-scale, i.e., -log10(p). This has at least three advantages: - Larger values are the ones more significant. - Contrast becomes more nuanced, without all significant p-values being shrunk to the extreme of the colourscale. - The p-vals in log-scale have a quasi-linear relationship with effect sizes, which makes these images fairly similar to effect size maps, that Tim alluded to. So, if you can, use -logp. For a test level 0.05, the threshold is -log10(0.05) = 1.3. Otherwise, for a similar behavior as randomise, use -save1-p, in which case the equivalent threshold would be 0.95. All the best, Anderson On 16 March 2018 at 10:41, Jodi Gilman <[log in to unmask]> wrote: > Hi, > > In documentation for randomise, when viewing the 1-p results in FSLView > the min/max display range should be set to 0.95/1.0 so that values less > than 0.95 (equivalent to p>0.05) are not shown. If these are corrected > values (i.e. corrp) then the visible areas correspond to the statistically > significant regions. > > Is this the same for PALM results? > > Thanks so much, > Jodi >