Hello Jeanette,
Thank you - I really appreciate it.
I'm actually more concerned about how big the difference between F and t results is.
What you describe in my case looks like this:
2x2 F-contrasts: no significant results even on p>0.999 (FWE corrected)
2x2 t-contrasts: significant results on p<0.025 (FWE corrected).
Not a 1:2 difference in p-values as 1-tailed, 2-tailed explanation suggests.
I know in FSL the tests are not parametric, but can it account for such a difference?
Not only Maya, but also Timothy Meeker asked about it in TBSS analysis (11/08/2012). Anderson suggested 2 solutions:
a) more permutations. In my case increasing it to 5000 didn't change it.
b) setting exchangeability blocks to 4 groups. In my case it didn't change anything, because it's not a repeated measures design.
Would you have any other explanation why the difference in p-values is so big between F and t-tests? Or isn't it anything unusual?
All the best and thank you,
Danusia
|