The average duration of trials is 14 seconds (11.5 - 16.5 s);

There are two events: Stimulus and Response. Stimulus starts at trial onset; Response starts 4.5, 5 or 5.5 seconds after trial onset.

Individual level analysis: we are considering both a TENT (11 basis functions) and a GAM (Stimulus and Response regressors) approach.

While 3dMVM seems really appropriate to analyze TENT results at the group level, I was wondering if it would be appropriate to analyze also results from the GAM GLM with 3dMVM. My question is: can the two betas associated with the two regressors (Stimulus and Response) be considered multiple response functions (and be better analyzed in a single model with 3dMVM)?

Useful note. There is no collinearity in the GAM approach; the two activation maps are different and coherent with what was expected.
May-be-useful note. 2x2 factorial paradigm (2 voice types x 2 trial types).

However, I still have this doubt: can the two betas associated with the two regressors (Stimulus and Response, modeled at the individual level using GAM functions) be considered multiple response functions (better analyzed in a single model with 3dMVM)?

EDIT: ok, I think you mean using 3dANOVA3 and treating the two regressors (Stimulus, Response) as a random factor. Is that so?

I still have this doubt: can the two betas associated with the two regressors (Stimulus and Response, modeled at the individual level
using GAM functions) be considered multiple response functions (better analyzed in a single model with 3dMVM)?

I’m confused with your description as well. My suggestion was simply based on the fact you mentioned a 2x2 factorial paradigm. However, it remains unclear to me what exactly those two factors are. Could you describe the data structure in more details? How many effect estimates do you for each subject that go into the group analysis?

In the GAM approach, for each subject I estimated:

the activity associated with the stimulus presentation;

the activity associated with the motor response.

Important infos:

The two are both estimated using the function GAM in 3DDeconvolve;

There is no collinearity between regressors. And the two activation maps (at the single subject level) are different, and coherent with what was expected.

It is a 2x2 factorial paradigm, so there are 4 conditions.

Question:
It is legitimate to treat the activity associated with the two events as multiple response functions (stimulus = T0; response = T1), analyzing them in 3dMVM?

the activity associated with the stimulus presentation;

the activity associated with the motor response.

Is the reason you have a 2 x 2 structure because each of the two stimulus presentations is associated with one motor response?

It is a 2x2 factorial paradigm, so there are 4 conditions.

If you consider those 4 conditions as a 2 x 2 structure, then it’s a two-way within-subject (or repeated-measures) ANOVA, which can be directly handled by 3dANOVA3 -type 4. Of course, you can also use 3dMVM, but it would take much longer runtime.

Unless it’s not really a 2 x 2 factorial structure.

Is the reason you have a 2 x 2 structure because each of the two stimulus presentations is associated with one motor response?

I’m sorry I wasn’t clear. No, there are 4 runs, and each run represents a condition:

First run (first condition): factor I level A, factor II level A;

Second run (second condition): factor I level A, factor II level B;

Third run (third condition): factor I level B, factor II level A;

Fourth run (fourth condition): factor I level B, factor II level B.

Within each run, there are 32 trials. Each trial has two events: STIMULUS presentation and RESPONSE.

I understand the rationale of using 3dANOVA3 -type 4, and i agree it is an optimal model. But in 3dANOVA3 I will use as a dependent variable the estimated activity during STIMULUS or the estimated activity during RESPONSE. Not both (as far as I understand).
I am still wondering if 3dMVM would instead allow me to use as inputs the betas associated with both STIMULI and RESPONSE.

In the attempt of being exhaustive, I attach a single-subject 3dDeconvolve generated with afni.proc.py and putative group analysis with 3dANOVA3 and 3dMVM (using only 2 subjects to spare space)

GROUP ANALYSIS USING 3dMVM - note that there are both STIMULUS (t1) and RESPONSE (t2) as dependent variables:

3dMVM \
-prefix MVM_MultFun_results \
-wsVars "run*condition*Time" \
-bsVars 1 \
-SS_type 3 \
-mask GroupMask+tlrc \
-dataTable \
Subj run condition Time InputFile \
s1 S A t1 'S1/SoA_S1_GAM.results/stats.SoA_S1_GAM+tlrc[1]' \
s1 S NA t1 'S1/SoA_S1_GAM.results/stats.SoA_S1_GAM+tlrc[7]' \
s1 NS A t1 'S1/SoA_S1_GAM.results/stats.SoA_S1_GAM+tlrc[13]' \
s1 NS NA t1 'S1/SoA_S1_GAM.results/stats.SoA_S1_GAM+tlrc[19]' \
s2 S A t1 'S2/SoA_S2_GAM.results/stats.SoA_S2_GAM+tlrc[1]' \
s2 S NA t1 'S2/SoA_S2_GAM.results/stats.SoA_S2_GAM+tlrc[7]' \
s2 NS A t1 'S2/SoA_S2_GAM.results/stats.SoA_S2_GAM+tlrc[13]' \
s2 NS NA t1 'S2/SoA_S2_GAM.results/stats.SoA_S2_GAM+tlrc[19]' \
s1 S A t2 'S1/SoA_S1_GAM.results/stats.SoA_S1_GAM+tlrc[4]' \
s1 S NA t2 'S1/SoA_S1_GAM.results/stats.SoA_S1_GAM+tlrc[10]' \
s1 NS A t2 'S1/SoA_S1_GAM.results/stats.SoA_S1_GAM+tlrc[16]' \
s1 NS NA t2 'S1/SoA_S1_GAM.results/stats.SoA_S1_GAM+tlrc[22]' \
s2 S A t2 'S2/SoA_S2_GAM.results/stats.SoA_S2_GAM+tlrc[4]' \
s2 S NA t2 'S2/SoA_S2_GAM.results/stats.SoA_S2_GAM+tlrc[10]' \
s2 NS A t2 'S2/SoA_S2_GAM.results/stats.SoA_S2_GAM+tlrc[16]' \
s2 NS NA t2 'S2/SoA_S2_GAM.results/stats.SoA_S2_GAM+tlrc[22]'

OK, you have a 2 x 2 x 2, not 2 x 2, data structure. Indeed, 3dMVM is the way to go.

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.