Hi Anders,
When dealing with multiple tests in this way, the significance level still needs to be corrected to 0.05/2 - using TFCE doesn't change this. An alternate approach would be to run a F-test on the two contrasts and _only_ use the F-test p-values to determine significance at 0.05 ( although data from the two contrasts can be used to establish directionality ).
Hope this helps,
Kind Regards
Matthew
> Dear all,
>
> A very basic question:
> I want to use randomise with TFCE to compare two groups of subjects (A and B).
> My design has two contrasts: A > B and B > A.
> I would like to apply a two-sided significance level of 0.05 for testing the hypothesis of A not equal to B.
> Randomise in this case carries out two one-sided tests.
> Does this mean that I should threshold the randomise output at p<0.05/2, i.e. 1-p>0.975 ?
> I supposed that this was the correct approach, but I was recently informed that this is not nescessary when using TFCE, since each one-sided test in this case only deals with exactly half of the voxels.
> What is the correct way of doing this?
>
> Best,
> Anders
>
|