Question on group-level analysis with subject-specific covariate

AFNI Message Board
1

Hi AFNI experts,

I’m running an analysis and wanted to make sure I am correctly setting up the group-level model that includes a subject-specific covariate.

Subject-level model (3dDeconvolve)

For each subject, I run a GLM that includes trial-type regressors and motion parameters. The relevant portion of the script is:

3dDeconvolve \
  -overwrite \
  -input sub-${subj}_task-TASK_allrun_space-MNI152NLin2009cAsym_SI.nii.gz \
  -mask sub-${subj}_space-MNI152NLin2009cAsym_GM_probseg_resampled.nii.gz \
  -concat '1D: 0 343 686' \
  -polort A \
  -nobout \
  -noFDR \
  -num_stimts 6 \
  -local_times \
  -stim_times 1 Stim_Onset.txt 'GAM(8.6,0.547,6)' -stim_label Stim \
  -stim_times 2 Determ_Onset.txt 'GAM(8.6,0.547,2)' -stim_label Determ \
  -stim_times 3 PE_Pos_Onset.txt 'GAM(8.6,0.547,2)' -stim_label PE_Pos \
  -stim_times 4 PE_Neg_Onset.txt 'GAM(8.6,0.547,2)' -stim_label PE_Neg \
  -stim_times 5 Error_Stim_Onset.txt 'GAM(8.6,0.547,6)' -stim_label StimError \
  -stim_times 6 Error_Feedback_Onset.txt 'GAM(8.6,0.547,2)' -stim_label FeedbackError \
  -ortvec Motion.1D'[0..5]' MotionParam \
  -ortvec Motion_deriv.1D'[0..5]' MotionParamDerv \
  -num_glt 2 \
  -gltsym "SYM: Determ +PE_Pos +PE_Neg" -glt_label General_Task \
  -gltsym "SYM: 0.5*PE_Pos +0.5*PE_Neg -Determ" -glt_label Unsigned_PE_Effect \
  -cbucket sub-${subj}_Trial_Type_betas.nii.gz \
  -bucket sub-${subj}_Trial_Type.nii.gz \
  -fitts sub-${subj}_fitts.nii.gz \
  -errts sub-${subj}_errts.nii.gz

The GLT I’m interested in at the group level is the Unsigned_PE_Effect contrast.

For each subject, I also have a behavioral model-derived Asymmetry Index (AI).
AI is computed from subject-wise learning rates: (α_pos − α_neg) / (α_pos + α_neg).

I want to test, at the group level, whether the Unsigned PE contrast varies systematically as a function of AI.

What I think I should do

My current understanding is:

  1. Extract the GLT brick for Unsigned_PE_Effect from each subject’s bucket.
  2. Create a list file (A_list.txt) containing one image per subject.
  3. Create a covariates file (ai.txt) containing subject IDs and their AI.
  4. Use 3dttest++ something like:
3dttest++ -prefix group_UnsignedPE_AI \
          -setA A_list.txt \
          -covariates ai.txt

My questions

  1. Is 3dttest++ the correct tool, or should I be using 3dMVM or 3dLME, given that this is a single contrast per subject but with a subject-level covariate?
  2. Is the covariate treated correctly by listing AI in a separate covariate table (one row per subject matching setA order)?

I want to make sure I’m setting up the parametric modulation by AI properly and not missing anything important.

Thank you in advance for any guidance — I really appreciate the help!

-- sahithyan

2

Sahithyan,

Your 3dttest++ script looks appropriate for the simple group-level regression model. You can also run 3dMVM to confirm the results if you’d like.

Gang Chen