Dear AFNI experts,
In 3dLME analysis, I included one within-subject factor (three time points), one between-subjects factors (two groups), and age/gender as covariance.
In the 3dLME output, I can see the significant region in the interaction of time points and group.
I was wondering how to do the post-hoc analysis to identify where these differences occur?
- Can I just pull out the significant region and estimate post-hoc analysis in the averaged value in the region? Or I need to check the t-test or ANOVA in voxel-wise?
- For the post-hoc analysis, what kind of post-hoc do I need to do?
Thanks a lot,
If you don’t have missing data, 3dMVM is more preferable than 3dLME.
- Can I just pull out the significant region and estimate post-hoc analysis in the averaged
value in the region? Or I need to check the t-test or ANOVA in voxel-wise?
You can obtain post hoc tests at the voxel level using option -gltCode in 3dMVM or 3dLME. Check out the examples in the 3dMVM/3dLME help for the usage.
For the post-hoc analysis, what kind of post-hoc do I need to do?
It is you as an investigator who decide on the specific research hypothesis. Usually you want to tease apart those main effects and interactions.
Thanks, I have already estimated the multiple comparison using 3dClusSim for the main effect.
(for example, uncorrected p = 0.01, cluster size = 137 voxels, corrected p = 0.05)
In addition, I use -gltCode in the 3dLME for the post-hoc analysis.
I was wondering how to do the multiple comparison in post-hoc analysis inside the significant regions were choose from the main effect.
In the post-hoc analysis, I have three group comparisons. (A-B, A-C, and B-C)
If you take the route of pairwise comparisons based on the main effect, consider conventional approaches such as Tukey’s HSD. Alternatively you can “correct” for multiple testing (e.g., cluster-wise FWE) for each comparison without resorting to the main effect.