3dNetCorr - Question on accuracy

Dear AFNI experts,

I am using 3dNetCorr to extract 360-by-360 matrices of functional connectivity among parcels.
A colleague advised me that the standard procedure used by 3dNetCorr (i.e., averaging time series and than correlating) may not be the most accurate choice. In fact, it can be argued that correlating each voxel-pair (among each couple of RoI) and then averaging the correlations is more accurate.

  1. Do you think that the two procedures are somewhat equal, or not?
  2. Is there a way to apply the alternative procedure (correlating voxels among parcels, and then averaging) in 3dNetCorr?
  3. -related to 2)- Is there a way to output also the standard deviation of such correlations (i.e., obtaining not only a matrix of correlations, but also a matrix of dispersion)?

Thanks in advance,