If you want to use AlphaSim instead of FWE or FDR for multiple comparisons correction, then using it for a two-sample t-test at an uncorrected p < 0.001 seems perfectly reasonable.
When you use AlphaSim to estimate a minimum acceptable cluster size, you are supposed to use the ResMS (residuals) image to estimate smoothness, rather than the data. Since there's only one ResMS image for each model, it's the same across contrasts. You would, of course, want to run AlphaSim separately for each t-test that you have, but not for each contrast in a t-test. The exact steps (and rationale) used to convert your ResMS into an AlphaSim-compatible image, to compute the smoothness of the noise, and to run AlphaSim are discussed in this listserv thread: https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1206&L=spm&D=0&1=spm&9=A&J=on&d=No+Match%3BMatch%3BMatches&z=4&P=15546
As for your last question, FDR and AlphaSim attempt to correct for multiple comparisons in different ways, so they aren't interchangeable. You should go in with one multiple comparisons correction method in mind and use that throughout your analyses. This AFNI forum thread discusses differences and similarities a bit: http://afni.nimh.nih.gov/afni/community/board/read.php?1,77089,77089#msg-77089
Hope that helps.
- Taylor
|