I am not sure I am interpreting the results of my ROI-to-ROI analysis at group level. I have four groups of subjects. I am trying to see if there are any between-network differences amongst the three of them. I am using the networks defined in CONN20 using three covariates of sex, age and education. The order of factors in my Subject effects list is: Sex, Age, Education, Grp1, Grp2, Grp3, Grp4. The selection of subject effects shows:
[0 0 0 -1 1 0 0;0 0 0 0 -1 1 0;0 0 0 1 0 -1 0]
When I run this I detect some between group differences showing up. My interpretation is that some or all of this outcome is caused by one or more of the pairwise comparisons between groups like Grp2-Grp1, Grp3-Grp2, and Grp1-Grp3. If so, then I should so part of the above outcome if I run the connectivity with three separate Subject effects as:

0 0 0 -1 1 0 0
0 0 0 0 -1 1 0
0 0 0 1 0 -1 0

But I get no connectivity difference for any of the above. Can you explain to me how is that possible? Thank you.

Hi Reza,

Yes, that's perfectly fine, because of different in sensitivity and specificity and also because of the arbitrary nature of the .05 significance threshold, the original test which evaluates "any differences" between the three groups can some times detect differences that are not expressed in any of the individual univariate tests (and similarly, significant differences in individual tests can some times fail to reach significance in the multivariate test). In general I recommend post-hoc tests to follow as similarly as possible the details of the original analyses (e.g. for ROI-to-ROI analyses, depending on the choice of thresholding technique used that typically involves exporting the ROI order and groups from the original analysis into the post-hoc analyses) and using post-hoc analyses for interpretation rather than statistical inferences (and this typically involves disregarding or relaxing the thresholds used in post-hoc analyses to focus on the relative strength and size of the effects rather than on whether they reach significance or not). 

Hope this helps
Alfonso
Originally posted by Reza Momenan:

Thank you for the clarification. How can I change the thresholds to see if doing so reveals the potential for being a trend? Thanks in advance.