Dear Dr. Smith,
Thank you very much for your prompt reply.
About the multiple 68 comparisons correction, do you mean that if my significant level is 0.05, using Bonferroni correction, p <0.007 (0.05/68)for each ROI?
I did not do that. I checked my data again, even with that. still couple of ROIs survived. While randomise showed none significant voxel. Does it mean "randomise" is conservative (less sensitive)? Or shall I change the TFCE default setting to less connectivity (default=26)? Sometime, the randomise will give a huge number for the permutation needed for the contrast. For example, it will say "1.8347 e+10 permutation needed. I only did 5000. Will that affect the results since not enough permutations are performed?
From you paper, skeletonized images follow Gaussian distribution. Is it ok for me to use parametric statistical methods?
Sorry for throwing lots of questions to you. I want to make sure that I did not mess up my data analysis before publishing my results.
Thanks a lot again.
Have a nice weekend!
On 4 Nov 2011, at 06:05, Yingying Wang wrote:
Dear Dr. Smith,
When I used mean_FA_skeleton_mask as the mask for randomise procedure, I did not find significant voxels surviving from TFCE correction. After I exported the mean values for 68 regions from JHU DTI-81 and tract templates, I used SAS to do the modeling. I found significance in couple of ROIs according to the mean in the ROI.
Did you correct for multiple (68) comparisons?
Then, I went back to use "randomise" with the ROI as mask instead of mean_FA_skeleton_mask, the randomise result showed some voxels surviving from TFCE correction. Why is that?
Because you now have a greatly reduced multiple comparison correction..this is a rather circular (i.e. not safe) analysis I'm afraid!
Cheers, Steve.
Can I use small mask to do randomise procedure then?
Thank you so much.
Yingying
===========================================================
Yingying Wang, Graduate Student, Biomedical Engineering, University of Cincinnati.
Pediatric Neuroimaging Research Consortium, Cincinnati Children's Hospital Medical Center.
MLC 5033, 3333 Burnet Avenue, Cincinnati, OH 45229-3039, United States.
===========================================================
