Dear Alfonso,

I am interested in coducting a seed-to-voxel gPPI analysis, which as I understand is typically implemented with bivariate regression measures. However, since this configuration measures effective (i.e., directional) connectivity, while I am interested in functional connecitivity more generally, I decided to conduct a gPPI analysis with bivariate correlation measures. I know that bivariate correlation measures are typically implemented in the context of a weighted-GLM analysis. However, my experimental task design is event-related and from reading other forum posts I know that in this situation the weighted-GLM analysis is less sensitive (vs. the gPPI analysis) to connectivity effects. Indeed, in my case the clusters identified in the weighted-GLM with bivariate correlation measures (vs. the gPPI analysis with bivariate correlation measures) were fewer and smaller. Thankfully, they overall overlapped with the results of the gPPI analysis with bivariate correlation measures, which I think suggests that the two analysis approaches are accessing the same or highly similar connectivity construct, albeit expectedly the weighted-GLM analysis is less sensitive. Does this sound like a fair conclusion?

Could you explain how exactly CONN computes the bivariate correlation co-efficients in the context of a gPPI analysis implemented with bivariate correlation measures? I do not understand how we arrive at bivariate correlation co-efficients from the equation for the gPPI model presented in the CONN Handbook.

Also, I saw a thread (https://www.nitrc.org/forum/message.php?...) in which a person noticed that the gPPI analysis implemented with bivariate correlation measures produces assymetrical results. In this thread, you mentioned that you would try to force CONN to compute symmetrical results in the context of a gPPI analysis with bivariate correlation measures. For example, if a ROI A (seed) is significantly connected with voxel X (target) in the context of a gPPI analysis with bivariate correlation measures, then this analysis would also always identify voxel X (if selected as seed) as significantly connected with ROI A (target). However, I am not sure whether this has been implemented. Could you confirm?

Finally, how would we interpret significant results (I am specifically only interested in the positive ones) obtained in the context of a seed-to-voxel gPPI analysis with bivariate correlation measures, considering that the 1st-level gPPI model (unlike the 1st-level weighted-GLM model) regresses the seed region's mean BOLD time-series from each voxel's BOLD time-series? Would we interpret significant results simply as functional connectivity of the seed with a given target cluster of voxels DURING a given task condition or is there any more nuance to the interpretation?

Thanks!
Tsvet