I have a couple of questions, two theoretical and one technical. I think the theoretical ones will make more sense if I ask the technical question first.
Technical question: I am going to run an experiment where I present a stimulus (CS) paired with a shock which occurs at different times during the stimulus (but the shock occurs on every trial). The CS itself will always be presented for 12 seconds but the shock will occur at 6, 8, 10, or 12 seconds after the CS onset. In terms of making the stimulus timing files for 3dDeconvolve I’m not sure what is accurate. On the surface since the durations of the CS are always 12 seconds, you could assume one can get at this activation by using a BLOCK(12, 1) for the stimulus’ Rmodel and then to have the shock as a regressor of no interest by including their occurrence with the -stim_file option? But the more I think about it the more I’m not sure this will work since the shock will elicit it’s own response that lasts for 12+ seconds itself. So I guess I’m afraid that using BLOCK(12,1) for the CS will include contamination by the shock in the situations in which the shock occurs at 6, 8, or even 10 seconds. Because ultimately 3dDeconvolve is doing a regression perhaps this isn’t a huge issue but I’m not sure. An alternative would be to just model the duration of the CS BEFORE the shock occurs (e.g. model the 8s before the shock on the 8s trials, etc.) but I’m used to stimulus timing files having ‘duration’ measures that are consistent (i.e. BLOCK(12,1)) and am not sure how to model different durations in the same stimulus or if this is even a valid way to do this. Is there an optimal way of dealing with stimuli that act like this?
Theoretical question 1: I have been told for my entire career that fear conditioning studies like this need to have relatively long ITIs in order to allow the signal to return to baseline after the shock is administered. I have old data from a different study where the stimuli were 8s (and the shock always occurred 8s after the stimulus onset) and the ITIs were ~20s. When I plot the timecourse of the signal in areas that respond to the shock the signal seems to return to baseline after ~12s. As such my ITIs for this experiment are 16s. However I was talking to a colleague and we were wondering if this even mattered? Or, in other words, does the BOLD signal HAVE to return to “baseline” in order for 3dDeconvolve to give an accurate estimate to responding to the CS? On the surface it shouldn’t. I’m not at all an expert on how exactly 3dDeconvolve does what it does but if you’re modeling the CS and the shock, the process is basically just a regression so the signal returning to baseline shouldn’t necessarily matter? It’s an issue for us because these long ITIs really lengthen the runs and the participants hate it and now just conceptually I’m not sure it’s even necessary.
Theoretical question 2: This ties into question 1. Again I’ve been told that the shock should not occur too close in time to the onset of the CS because then the response to the CS will overlap with the response to the shock and you won’t be able to disambiguate the two. So shocking 2s after CS-onset would be dumb for this exact reason. I understand that no problem. My problem is with the seemingly arbitrary “cutoff” of an 8s duration CS. So I was told if I shocked at 6, 8, 10, and 12s I’d essentially have to exclude analyzing the 6s trials because of that contamination but the 8, 10, and 12s trials were all fair game. But again I’m now questioning that because it seems like an arbitrary distinction. From the timecourses it seems like the BOLD response definitely peaks within 8s 90% of the time but it still seems to peak within 6s ~75% of the time. Is this what I should be looking for when making the decision about which trials to analyze? I’m just trying to understand if it’s even necessary to not analyze the 6s shock trials at all since it decreases my overall power.
This is really long so literally any correspondence would be much appreciated. Thanks for your time, everyone.