I plan to use a poisson-distributed jitter (mean 4 s) for my experiment and wonder whether the jitter can / should be a multiple of the TR (2 sec), i.e., 2, 4, 6, 8 s or whether it should NOT be synced with the TR. Apparently, different people handle this differently but is there a clear argument in favor / against one option or the other?
For many years, the dogma was to TR lock your ER jitters and presentations. It still remains the simplest way to analyze data across all of the analysis platforms. However, over the recent years, a number of off-TR solutions have been implemented. One in particular is the ability to use tent functions in AFNI to span TR for estimates of the hemodynamic response profile. This allows the jitter to be off TR, but you still need to know the exact timing of the stimuli. I refer you to this site: http://www.personal.reading.ac.uk/~sxs07itj/web/AFNI_basis.html or the AFNI homepage to explain the differences in analysis.
Really, the question comes down to your paradigm. If it is important that you have an off TR jitter/presentation, you should go ahead and do that once you have a handle on how you will implement the analysis (which should be tenable). If the paradigm can be done with on-TR presentation without sacrificing the thesis of the project, it may be simpler to TR lock.
As the previous post mentioned, for a very short time everyone thought time-locking to the TR was important. Turns out, it makes you susceptible to slice-timing effects which can affect -- merely by artifact -- your activations if you don't correct for it (see the Huettel et al MRI textbook for an extremely clear explanation of that and the false activation effects it produces). Of course you can correct for this in most software with a simple click here-or-there. But considering that today most software does not need you to necessarily input your timing with one value per TR, and accepts timing onsets specified in seconds or milliseconds, I disagree with the statement that "on-TR presentation is simpler" in any relevant respect (though, this is personal opinion!).
As Keith said, you probably want to think of what is best, psychologically for your design. If jittering and not time-locking affords you a bigger number of trials without affecting the psychology of the task (strategies, boredom, difficulty...) then more trials is always better (though more subjects is even better..!). If for some other reasons it is best to keep the pace of the TRs, then you probably can but you want to make sure that, one way or another (and people discuss about this..), you are correcting for slice-timing effects.
Thank you very much for your helpful comments! Apparently, it truly depends on the task / design and there is no clear answer for this. I found another interesting source that I would like to refer the interested reader to (see p. 197, "timing issues"): http://hfac.gmu.edu/news/fMRIJournalClub/papers_fall_2011/Henson2007.pdf
From a sampling point of view it would be best to not time lock to a TR or even a fixed fraction of a TR. The best is to sample as much of the response as possible. So within limits of your design, and to have it balanced in a statistical sense as well, you should let your jitter run across the continuum of the TR.
- Badges
- Science topic
- Similar topics
- Statistics
- Survey