> 1. Connectivity analysis. I am using 3dGroupInstaCorr in the batch mode to put the seeds.
> Since no residual timeseries is generated, what to do? I could use the extent threshold from
> task-based analysis as a first guess. But how to justify its use? Is there a more appropriate way?

The seed-based correlation analysis is a little murky to start with. Since there is no clear definition of signal vs. noise in this context, you could estimate the FWHM values after you remove the obvious confounding effects such as slow drift and physiological noise.

> 2. Contrast maps. I have two conditions in the experiment and I want to contrast the brain
> activation across them. What should be the appropriate extent threshold to apply to the
> contrast? Should I first subtract the residual of one condition from the other's, then use that
> difference timeseries as input to FWHMx and proceed as usual?

Even though not ideal, the current practice is to estimate the FWHM values based on the residuals from the individual subject regression analysis. And the FWHM values might be automatically provided if you use afni_proc.py

> 3. Contrast of connectivity maps. I have connectivity maps in two conditions and I want to contrast
> them. How to obtain the appropriate extent threshold?

Same as 1 above.

Gang