hello, 
I hope the following questions are not really basic stuff. 
I'm building a connectivity matrix based on gPPI beta-weights and i want to use this matrix to calculate various graph-theory measures. 
my first question is whether i can refer to the beta-weights as edges, representing connection strength between brain regions ? 
my second question is whether or not i need to use fisher's Z transformation on this matrix? 
are there any other important considerations that i need to think about before calculating graph theory measures on the connectivity matrix?

I think there are three options for doing this, but neither has really been done before:

(1) Use the beta-weights as edges 
(2) Use the z-transform of the T-statistics
(3) Normalize the the beta weights in some way

I think the two important things to consider are:
(a) if the slope of region X and Y is 3 for task A and 2 for task B, then A-B is 1. Ignoring, for a moment the effects of deconvolution on the regressors, if you flip X and Y, you'd get 1/3 for task A and 1/2 for task B, then A-B is -1/6. I believe that something needs to be done to accounted for in the graph theory (e.g. maybe combine these in some way).
(b) I would massively downsample the data first to the number of nodes in the graph. Then you can treat each voxel as an ROI and loop the PPPI code over all ROIs.

If you want to discuss this more, I'm happy to help think about the best way to construct the graph.

-Donald


Originally posted by noamagal: