> Does it make a difference whether 3dDeconvolve is run on individual runs
> that are then merged using 3dMerge post-deconvlolution vs. running
> 3dDeconvolve after concatenating multiple runs using 3dTCat and the -
> concat option in 3dDeconvolve?
There are actually 3 different approaches to handling multiple runs:
(A) Concatenate all runs and run regression with each condition/task as one regressor across runs.
(B) Concatenate all runs and run regression with each condition/task as a separate regressor across runs. In other words, you would have 4 separate regressors for condition A if you have 4 runs.
(C) Analyze each run separately.
People usually go with (A), which is fine if there is little across-run difference for each regressor effect. If such an assumption does not hold, the approach might not be a good choice.
(B) is most appealing to me if there are enough sample size for each run because it would allow me to test whether there is across-run difference such as habituation for each condition/task. Without even testing the hypothesis of across-run difference, approach (A) could be problematic.
Gang