For events-related design, do I have to specify actual event duration for basis function? Say I have stimulus A for 4s and B for 0.5s. I am thinking of using BLOCK(4,1) for A, GAM for B, or maybe BLOCK(0.5,1)???. My understanding is that the user-specified durations are (internally) used a boxcar/square wave to convolve with the underlying basis function.
I am thinking of using BLOCK(4,1) for A, GAM for B, or maybe BLOCK(0.5,1)?
For consistency, use BLOCK(4,1) and BLOCK(0.5,1). Read the manual of 3dDeconvolve for the subtle differences between BLOCK and GAM:
3dDeconvolve -help | less
I would like to comment on my own questions and add some notes here for future users’ references.
The introduction of this paper https://www.ncbi.nlm.nih.gov/pubmed/18775784 gives a good theoretical background regarding the specification of event durations. (depending on the different models)
This link by Rick https://afni.nimh.nih.gov/pub/dist/doc/misc/Decon/2007_0504_basis_funcs.html gives pratical choice suggestions
My impression is: if duration is very short (<2 or 1s), go for GAM. If longer, one can then pick BLOCK(d,1).
Thanks for your reply. You have a good point. For my case, I am not going to compare A against B (but within A1,A2,A3; B1, B2, B3). So maybe mixing BLOCK and GAM might be fine.
So maybe mixing BLOCK and GAM might be fine.
Just as p-value thresholding is quite arbitrary, so too the differentiation between event and block trials is just for the convenience of discussion. Practically it might be OK with a mixture, but I would prefer to keep the modeling consistent unless there is a more legitimate reason.
If you are going to compare those conditions, a 4 second
one vs a 0.5 second one, even as part of a contrast, then
the “,1” part is not advised. Consider more simply
BLOCK(4) and BLOCK(0.5).
Equal resulting betas would mean equal BOLD responses,
as if they were measured over the same durations. See the
details of BLOCK(d) in the link you referenced.
Thanks for reminding me of that page.