I am doing VBM and giving a go at the cluster-based method. I was instructed to look at a conversion table two-tailed, unless I am interested in one contrast and one contrast only to get my t-value. Regardless of being interested in both tails (which I am), isn't each contrast by default a one-tailed test? So my questions are:
1. Since all the contrasts seem to be one-tailed (contrast file below) shouldn't I use the one-tailed p-value? Or do you use the two-tailed since the contrasts are not all together independent from each other.
1 -1 0
-1 1 0
0 1 -1
0 -1 1
1 0 -1
-1 0 1
2. Also to estimate the t-value I need to provide degrees of freedom. If I have a different number of people in each group the df would change for each comparison. For example grp1 has 27 and the other 2 grps have 30. So comparisons between a group with 27 and a group with 30 (N-2=55) would be different than comparing two groups with 30 people (N-2=58) and require different t-values. Do I just run two different randomise commands for the contrasts that have the same dfs?
3. Is it worth just doing an ANOVA since I am interested in looking at differences across all groups in both tails? If so is there an addition to the fslvbm_3_proc command to tell it you have a .fts file instead of a design.con file? I tried to run it as is with my .fts file in the directory and it errored out because of the lack of design.con.
Thanks again for all of your help!
|