Hello Everyone,

I have been asked to work on an fMRI data set including 3 groups of 20 each.  Each subject received 3 runs of a Go-No Go task where subjects were required to press a button (Go trial) or inhibit a button press (Stop trial) every TR (2500 ms).  Hence every single image collected during the run is associated with a trial and there are approximately 3 times as many Stop trials as go trials.  Initially, I modeled only the Stop trials (1 EV), which amounted to a Stop v Go contrast.  However, several anonymous reviewers suggest that we model correct and incorrect responses separately (and made publishing the manuscript contingent upon the new analysis).  Hence, I have re-analyzed these data modeling correct Go trials, incorrect Go trials, correct Stop trials and incorrect Stop trials (4 EVs).  The results are very poor, with fewer than 3 subjects showing any activation above threshold.  Indeed, the zstat files tend to vary between -0.01 to 0.02.  Furthermore, the analysis for the 180 runs lasted a very long time (over 6 days on a dual processor G5).  How should improve this design to improved to better model these data and to improve the subsequent results?

Best,

David