Hey Andrew,

I've been working hard over the past week on both the NBS and the BCT. I have a few final questions that I hope you would be able to help me with. You've been extremely gratious and  forhtcoming with your expertise, and I honestly could not appericiate it more. I tried getting most of these answers from the available pages of your recent book, but as an unpaid intern living abroad I can't afford it at the moment.

1) So if I understand you correctly, only the functions that include "signed networks" as an option can deal with negative values? (i.e. local_assortativity_wu_sign.m or clustering_coef_wu_sign.m). For the sake of simplicitiy, I feel it is probably wise to stick to binary undirected values. I've given up on all but the original BCT and am just writing scripts instead of a UI. In this case, should I apply some form of normalization on the thresholded matrices before computing/plotting the metrics? 

2) I've seen different approaches to this in different articles, but should I pick a single or small range of thresholds (i.e. 30% or 20-30%, as these values are more often significant than the 30-50% range), or plot changes from density 1 to 50 (relative)? 

3) Follow-up question 2. Should I enter the individual values (1 per subjects so 12 per theshold) into a repeated measures anova with 2x2(x50 in case of 50 density levels), or should I run paired t-tests on each of the pairs (resulting in 50 t-values, one for each density level) and create contrasts by subtracting the measures for music from the no-music scan for each drug, and then a paired t-test between the 2 contrasts? Then conduct FDR on the pvalues?

4) Do I need to control for multiple comparisons for the NBS? Since the effects of music on functional connectivity seem to be diametrically opposed for  (Drug reduces FCD in certain ROIs, which is increased by music, while placebo has higher FCD but is decreased by music) I've looked at both main effects seperately, as well as in the full model. This leaves me with 6 outcomes for the full model (Drug connectivity increase (+) and decrease (-), Music + and -, Interaction + and -), and 8 additional outcomes (Drug effect no-music + and -, Drug effect music + and -, Music effect placebo + and -, and Music effect drug + and -). For each of these outcomes (8 out of 14 have significant edges) I also ran several thesholds from ~3 to ~4.5 to funnel down the number of edges to the most influential. 

5) How should I interpret negative interaction effects? Based on my design matrix(PCB NM, PCB M, Drug NM, Drug M), is the contrast DrugvPCB or PCBvDrug and similar for Music V No music? When calculating t-tests based on mean functional connectivity (mean for each colmun in the 90x90 matrix), I subtracted Music from No-Music and Drug from Placebo. I'm unsure if the NBS uses the same logic in the contrasts. 

6) Most articles seem to only mention the benefit of partial correlation (models direct connection by reducing suprious/unrelated connectivity between the remaining pairs), but what exactly is the downside of this? I've decided to follow your advice and stick to full correlation, but not entirely sure how to defend that decision.

7) Are the p-value for the network and t the t-values for the significant edges saved anywhere in the NBS.out output file? I've saved the NBS file and binary matrix for each analysis but I can only find the alpha level I selected in the UI, but not the lowest p-value that was found for the network). I'm afraid I'll have to run the analyses again and write down each p-value as well as the t-values between the signfiicant edges.

Hopefully these will be my final questions. If you're interested in the results, I can probably send you my internship report once it is finished. Thanks again for your help.

Kind regards,

David