Dear Alfonso and others,

since I didn't get any response, I am also re-posting my questions in case somebody has time to give an insight please.

I am doing a task-based ROI-to-ROI analysis with 12 seeds (and targets) a priori defined from an independent dataset. I have a number of questions related to my analysis. I also want to confirm my understanding of previous forum threads so I will be as detailed as possible.

My research question concerns whether a sensorimotor network (defined by my 12 seeds) changes in Condition A compared to Condition B. Since I am interested in the network of the 12 seeds, I select ALL seeds and not one seed at the time. I want to test the connectivity of this network both in the individual contrasts (Condition A > Baseline and Condition B > Baseline, where Baseline is another modelled condition) but also in the direct contrast (Condition A > Condition B).

Q1. Connectivity matrix. To create the 12 x 12 connectivity matrix at the group level, I use the ROI.mat that is exported each time I run the Results Explorer. This matrix includes Z transformed correlation coefficients (if I want the original, I do tanh(h)) and the uncorrected one-sided p-values corresponding to the connections between ROIs. For the contrasts I am interested in, I confirmed that the h given in the ROI.mat is the difference between the h's of each subject in the two conditions. I read here (http://www.alfnie.com/software/conn#TOC-...) that for the analysis of the full connectome I must choose between:
a) Use the threshold ROI-ROI connections (by intensity) and p-FDR (analysis level) connection-level threshold. Here I can report all connections that survive corrections, right? Since the corrections are based on the analysis level (and not at the seed level) ROI1 -> ROI2 has the same corrected p-value with ROI2 -> ROI1. So, the matrix is symmetric. Then what I need to do is to plot the connectivity matrix (the hs or tanh(h)) but after having changed the uncorrected one-sided p-values of ROI.mat with the FDR (analysis level) corrected ones in order to display only the significant ones, right?
b) Use the F-statistic for each seed given in the table. I have seen non-significant seeds (FDR corrected for the F test) with significant connections to other seeds (FDR corrected for the t test). How could I interpret this?
It is written here (http://www.alfnie.com/software/conn#TOC-...) that for multiple seeds, I need to use seed-level threshold. So, if I understand correctly, I first choose the seeds with significant F test (by checking the threshold seed ROIs(F-test)) and I also set threshold ROI-to-ROI connections (by intensity) and p-FDR (seed level). Then I can report the connections that survive the seed level connection correction, correct? But then the connectivity matrix will not be symmetric (I have a connection ROI1-> ROI2 that is significant but in the opposite direction it is not). Shouldn't the connectivity matrix be symmetric?
c) Use the seed-level (NBS) 'size' and 'intensity' measures of the table. Same logic as in (b) but now I choose the seeds with significant size or intensity (based on the permutation tests). Then for those significant seeds, I set the connection threshold to seed level and I report the significant ones. But again, how the connectivity matrix will be symmetric?
d) Use Network NBS.
You write 'Option (a) allows highly spatially-specific inferences -about individual ROI-to-ROI connections-, option (d) offers comparatively lower spatial specificity'. Given that I want to be able to make statements with spatial specificity like 'The SMA increased its connectivity with M1, CB and thalamus in Condition A than Condition B, while the thalamus increased its connectivity with S1 and S2, etc', from the above-mentioned webpage I understand that I must go with (a). Is what I understand correct?

Q2. One seed versus all seeds. Besides analyzing the whole connectome, I am also thinking to test specific seeds separately (once at a time). In this case, I would need to set the connection threshold to FDR corrected (either seed or analysis, it is the same) and report the connections that survive. Is this correct?

Q3. One tailed versus two tailed. I went through a script that previous threads have directed us to use for visualizing connectivity matrices. The script contains an if statement with respect to whether we want to test for anticorrelations (two-tailed test for more than 10 ROIs) or not. I was wondering how standard this is (I read paper saying that in purpose they test only for positive correlations). Does it depend exclusively on my research question?

Thank you very much in advance!