Hi FSL experts,
I have a query regarding the use of the 'randomise' method of permutation testing for higher-level analyses. Out of interest, I compared results using the standard FEAT higher-level analysis with the output of randomise, and found several discrepancies. For example, one contrast which produced strong effects in FEAT with a threshold of z=2.3 resulted in no significantly activated clusters using randomise with TFCE. I have seen several posts stating that the randomise method should be more sensitive as it doesn't assume underlying Gaussian distributions, however it produced fewer significant results than FEAT. How should these discrepancies be interpreted?
Thanks for any help with this,
Ciara
|