I have a rapid event-related design and the duration of my trials are variable. My plan is to run 3ddeconvolve on each individual trial so I can run MVPA. As I was reading over the 3ddeconvolve options, it seemed to me that the recommended options would be to use ‘dmUBLOCK(1)’ with -stim_times_AM2. My confusion is simple (I hope). I don’t really understand what I am meant to be using as my amplitude modulation. I can’t think of what I would be wanting to use other than duration, which feels redundant since I am already providing the event duration in the 1D file after a colon. I know I shouldn’t use AM1, and I know to provide no modulation simply returns the script to using AM1 defaults, so I am stuck. Below is a sample of a command I might run in this example:
3dDeconvolve -force_TR 2.8 -input pb03.sub114.r02.empty+orig.HEAD -polort A -stim_times_millisec -num_stimts 1 -basis_normall 1 -float -local_times -stim_times_AM2 1 sub114_run2_trial1.1D ‘dmUBLOCK(1)’ -stim_label 1 trial1 -tout -fout -bucket 1_DeconResult
Oh, I realized I never mentioned that the trial duration is the reaction time for the trial, which is why it feels redundant. Sorry about that!
My plan is to run 3ddeconvolve on each individual trial so I can run MVPA.
Do you mean that you want to estimate the BOLD response separately for each trial? If so, whether this is feasible or not is a different issue, but you don’t want to modulate each trial in the model.
Yes, that is correct. We do this before conducting mvpa analyses for rapid event-related designs.
Ok, if I do not want to modulate each trial in the model, then would you recommend that I use dmBLOCK with a simple stim_times option (no AM1 or AM2)? The documentation online stated, “If different stimuli in the same class ‘k’ have different durations, you’ll have to use the dmBLOCK response model and ‘-stim_times_AM1’ or ‘-stim_times_AM2’, described below.” So, I wasn’t sure if it would cause problems to use an option other than AM2.
Alternatively, because within a model for an individual trial there is really only one onset time and duration, could I instead just use BLOCK and stim_times? While the run does contain trials of different durations, the individual model only has one trial, so there isn’t variation within the model itself.
Heather, since you want to capture the BOLD response for each trial separately, I think that you should not modulate the trial event, not should you use dmUBLOCK. The reason is that you want to obtain the response magnitude that is associated with (not modulated by) the trial duration. So, just specify the trial duration with BLOCK.
Ok, thank you!
One last question! In a separate set of analyses, we will want to keep data for the whole run in the model (meaning we will not be estimating each trial separately). In this instance, I believe it is the case that we should use dmBLOCK. Could you clarify which stim_times variant would be appropriate in this case? This is with the same dataset, so trial durations are equal to participant reaction times.
In a separate set of analyses, we will want to keep data for the whole run in the model (meaning we will not be estimating each trial separately).
There are two approaches to handling reaction time: one is to treat it as trial duration, and the other is to consider it as a modulation variable. There is no consensus in the field as to which is more preferable.
If you go with the former approach, I suggest dmUBLOCK(x) instead of dmBLOCK so that you’re forced to decide exactly what you want to get out of the analysis by controlling the interpretation of the output: which trial duration do you want to associate with as for the BOLD response magnitude?