# Beginner 3dDeconvolve GLM question: Unmodeled TRs included in contrast between two modeled conditions?

Hi all,

I’m an fMRI beginner and I have a very basic mathematical question about what 3dDeconvolve does and does not include in each GLM. I consulted with a good friend who has used AFNI for years, but he wasn’t sure so I thought I would inquire here.

I am using a block design with six conditions. My timing files for each condition include all the TRs that occur during that condition, but there are a few TRs after each block that are not represented in any of the timing files.

My question: When I get my coefficients for each contrast (e.g., conditionA_vs_conditionB), do those GLMs include the unmodeled TRs in any way (e.g., are they modeled as “rest” and/or effectively covaried out)? Or are data from unmodeled TRs entirely excluded from analysis?

Similarly, I know my “conditionA_vs_conditionB” contrast includes regressors for covariates of non-interest (e.g., motion). Are the data from the other conditions (e.g., conditions C, D, E, F in this example, and/or rest) accounted for in the GLM in any way?

I think these questions both boil down to: How do I know what else is being statistically controlled in the GLM?

Thanks so much!
Lauren

Hi all, just following up to see whether the AFNI gurus or any other end users have any insight into this. Thanks so much!

Hi Lauren,

Indeed, it is good to have time when no stimulus is presented.
Yes, those time intervals are used to estimate the baseline,
with which each condition is contrasted, implicitly. The
conditionA beta weight is with respect to that baseline.

Keep in mind that the GLM is solved using all regressors at
once, including baseline, drifts, motion and conditions of
interest, for example. And that regression is solved at each
voxel independently.

You know what the GLM consists of by knowing what is in the
regression matrix, probably X.xmat.1D.

Consider reviewing this tutorial on first level analysis
using afni_proc.py. It is a bit old now, but should still
be useful. In particular, review t15_regression.txt and
t16_X_matrix.txt (though using the data from your own study).

• rick

Hi Rick,