Thank you Gang. That's not exactly what we're doing, but Matlab may hold the answer. A thought that crossed our minds was to use 3dmaskdump and manipulate enormous ASCII files, but Matlab might make things easier.

Just to clarify, we're combining two datasets each with TR=3 seconds, but one is sampling with a 1.5-second offset relative to stimulus presentation

relative to stimulus onset:
1st dataset samples at : 0, 3, 6, 9, 12... (seconds)
2nd dataset samples at: 1.5, 4.5, 7.5, 10.5...

Thus, with two scans of identical stimulus order, each with TR=3s, we effectively sample the response every 1.5 seconds by sampling at different lags.

Upon combining these, the resulting dataset would indeed need to have datapoints at a period of 1.5 seconds.

I will look into the Matlab manipulation. Thanks again.

-George