Hi Stef,
Sorry for being so slow to reply to this.
Yes, you can separate the final errts dataset into two pieces using 3dTcat. It should not be necessary to run afni_proc.py once for each run (depending on your answer to the following question).
Did the subjects leave the scanner between the two runs? If not, this seems okay. If they did, it would be better to allow for more flexibility in aligning the 2 runs.
To be sure, do you plan to compute correlations on each run separately? Or do you plan to correlate run 1 with run 2, say? The latter might be problematic due to (different) censoring between the runs. Correlating one run at a time should be reasonable.
- rick