I am running for the first time a task based fMRI analysis with afni. It is a very fast event related design (short events and short inter trial intervals).
I was visiting the afni bootcamp a few weeks ago and I was recommended to run my analysis with TENT function (TENT for the regressors I care about and GAM for the regressors I don t care about, as I had some collinearity issues when using only TENT), but also try GAM for all regressors and compare the two analysis.

My question is now, what should I focus on when comparing the results from different basis functions?
Would I only look at which basis function gives me a higher beta and t-stats for the regressor I am interested in? But this might also differ for the region I am looking at.
Should I instead compare the full model F statistics?

what should I focus on when comparing the results from different basis functions?

I assume that you’re referring to the individual level analysis. For the result with TENT per condition (or task), in the output from 3dDeconvolve/3dREMLfit you should have a t-stat for each tent and an F-stat for the overall shape of the hemodynamic response. You can compare the two modeling approaches (TENT vs. GAM) by pitting the statistical inference based on the F-stat from TENT against the t-stat from GAM.

I have a follow up question regarding the F-stats when using TENT. I can not find the answer in the afni slides. Maybe a stupid question.

I have a condition Lure_Object_FA, which I model in 3dDeconvolve by using 5 TENT regressors. I also performed a linear test
-gltsym ‘SYM: +Lure_Obj_FA[1…3]’ -glt_label 1 ‘Lure_Obj_FA’

What is the difference between the partial F-stats from the TENTS and the F-stats from this GLT?
When I understand correctly, the partial F-stats for the TENTS (#0 - #4 for Lure_Object_FA) is comparing the error sum of squares from the full model with the error sum of squares from the reduced model (where all 5 TENT parameters that are related to the condition Lure_Object_FA are removed).

I guess I still don t understand what exactly is analyzed/compared when performing a linear contrast like in the above GLT.
I would really like to understand this.

What is the difference between the partial F-stats from the TENTS and the F-stats from this GLT?

By partial F-stat from the TENTs, do you mean the F-stat 3dDeconvolve/3dREMLfit automatically provides? If so, that F-stat tests whether there is statistical evidence for the signal at any of the 5 tents. As for the test for your GLT

the F-stat tests whether there is statistical evidence for the summed effect of those 3 tents. If you want something similar to the partial F-stat, do the following with double square brackets:

The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.