Constructing t-test from GLTs

Hello AFNI team,

I'm interested in running several t-tests with various combinations of 8 task regressors I've created (3dDeconvolve code included). I'm wondering if there is any major statistical difference between creating GLTs with different combinations of task regressors versus re-running 3dDeconvolve with the stim_time files combined as needed?

For example, I'd like to run a t-test of task versus baseline. Currently I've combined all 8 task regressors into one GLT. (glt_sym1, peak_all_targets_9to13). Would it be preferable to re-run 3dDeconvolve with all of the onsets combined into one stim_time file, and then construct a glt with only that regressor included?

3dDeconvolve -input \
/func_dir/${subject}_ses-ne_task-newsevents_rec-topup_run-1_space-MNI152NLin2009cAsym_desc-preproc_bold_blur4mm_scaled.nii.gz \
/func_dir/${subject}_ses-ne_task-newsevents_rec-topup_run-2_space-MNI152NLin2009cAsym_desc-preproc_bold_blur4mm_scaled.nii.gz \
/func_dir/${subject}_ses-ne_task-newsevents_rec-topup_run-3_space-MNI152NLin2009cAsym_desc-preproc_bold_blur4mm_scaled.nii.gz \
/func_dir/${subject}_ses-ne_task-newsevents_rec-topup_run-4_space-MNI152NLin2009cAsym_desc-preproc_bold_blur4mm_scaled.nii.gz \
/func_dir/${subject}_ses-ne_task-newsevents_rec-topup_run-5_space-MNI152NLin2009cAsym_desc-preproc_bold_blur4mm_scaled.nii.gz \
-polort A \
-num_stimts 37 \
-mask /group_lvl/tpl-MNI152NLin2009aAsym_res-1_desc-brain_mask_dil2mm_2mm_master.nii.gz \
-censor /func_dir/${subject}_censor.txt \
-stim_file 1 /confounds/${subject}_rot_x.txt -stim_base 1 -stim_label 1 roll  \
-stim_file 2 /confounds/${subject}_rot_y.txt -stim_base 2 -stim_label 2 pitch \
-stim_file 3 /confounds/${subject}_rot_z.txt -stim_base 3 -stim_label 3 yaw   \
-stim_file 4 /confounds/${subject}_trans_x.txt -stim_base 4 -stim_label 4 dS    \
-stim_file 5 /confounds/${subject}_trans_y.txt -stim_base 5 -stim_label 5 dL    \
-stim_file 6 /confounds/${subject}_trans_z.txt -stim_base 6 -stim_label 6 dP    \
-stim_file 7 /confounds/${subject}_rot_x_derivative1.txt -stim_base 7 -stim_label 7 rolldx  \
-stim_file 8 /confounds/${subject}_rot_y_derivative1.txt -stim_base 8 -stim_label 8 pitchdx \
-stim_file 9 /confounds/${subject}_rot_z_derivative1.txt -stim_base 9 -stim_label 9 yawdx   \
-stim_file 10 /confounds/${subject}_trans_x_derivative1.txt -stim_base 10 -stim_label 10 dSdx    \
-stim_file 11 /confounds/${subject}_trans_y_derivative1.txt -stim_base 11 -stim_label 11 dLdx    \
-stim_file 12 /confounds/${subject}_trans_z_derivative1.txt -stim_base 12 -stim_label 12 dPdx    \
-stim_file 13 /confounds/${subject}_rot_x_power2.txt -stim_base 13 -stim_label 13 roll_sq  \
-stim_file 14 /confounds/${subject}_rot_y_power2.txt -stim_base 14 -stim_label 14 pitch_sq \
-stim_file 15 /confounds/${subject}_rot_z_power2.txt -stim_base 15 -stim_label 15 yaw_sq   \
-stim_file 16 /confounds/${subject}_trans_x_power2.txt -stim_base 16 -stim_label 16 dS_sq   \
-stim_file 17 /confounds/${subject}_trans_y_power2.txt -stim_base 17 -stim_label 17 dL_sq    \
-stim_file 18 /confounds/${subject}_trans_z_power2.txt -stim_base 18 -stim_label 18 dP_sq    \
-stim_file 19 /confounds/${subject}_rot_x_derivative1_power2.txt -stim_base 19 -stim_label 19 rolldx_sq  \
-stim_file 20 /confounds/${subject}_rot_y_derivative1_power2.txt -stim_base 20 -stim_label 20 pitchdx_sq \
-stim_file 21 /confounds/${subject}_rot_z_derivative1_power2.txt -stim_base 21 -stim_label 21 yawdx_sq   \
-stim_file 22 /confounds/${subject}_trans_x_derivative1_power2.txt -stim_base 22 -stim_label 22 dSdx_sq    \
-stim_file 23 /confounds/${subject}_trans_y_derivative1_power2.txt -stim_base 23 -stim_label 23 dLdx_sq    \
-stim_file 24 /confounds/${subject}_trans_z_derivative1_power2.txt -stim_base 24 -stim_label 24 dPdx_sq    \
-stim_file 25 /confounds/${subject}_a_comp_cor_00.txt -stim_base 25 -stim_label 25 a_comp_cor_00   \
-stim_file 26 /confounds/${subject}_a_comp_cor_01.txt -stim_base 26 -stim_label 26 a_comp_cor_01    \
-stim_file 27 /confounds/${subject}_a_comp_cor_02.txt -stim_base 27 -stim_label 27 a_comp_cor_02    \
-stim_file 28 /confounds/${subject}_a_comp_cor_03.txt -stim_base 28 -stim_label 28 a_comp_cor_03    \
-stim_file 29 /confounds/${subject}_a_comp_cor_04.txt -stim_base 29 -stim_label 29 a_comp_cor_04 \
-stim_times 30 /rand/$t2017to2015 'TENT(0,22.4,29)' -stim_label 30 2017to2015_targets \
-stim_times 31 /rand/$t2014to2012 'TENT(0,22.4,29)' -stim_label 31 2014to2012_targets \
-stim_times 32 /rand/$t2011to2009 'TENT(0,22.4,29)' -stim_label 32 2011to2009_targets \
-stim_times 33 /rand/$t2008to2006 'TENT(0,22.4,29)' -stim_label 33 2008to2006_targets \
-stim_times 34 /rand/$t2005to2003 'TENT(0,22.4,29)' -stim_label 34 2005to2003_targets \
-stim_times 35 /rand/$t2002to1998 'TENT(0,22.4,29)' -stim_label 35 2002to1998_targets \
-stim_times 36 /rand/$t1997to1993 'TENT(0,22.4,29)' -stim_label 36 1997to1993_targets \
-stim_times 37 /rand/$t1992to1988 'TENT(0,22.4,29)' -stim_label 37 1992to1988_targets \
-num_glt 15 \
-gltsym 'SYM: +0.125*1992to1988_targets[9..13] +0.125*1997to1993_targets[9..13] +0.125*2002to1998_targets[9..13] +0.125*2005to2003_targets[9..13] +0.125*2002to1998_targets[9..13] +0.125*1997to1993_targets[9..13] +0.125*1992to1988_targets[9..13] 
-glt_label 1 "peak_all_targets_9to13" \

Would it be preferable to re-run 3dDeconvolve with all of the onsets combined into one stim_time file, and then construct a glt with only that regressor included?

The model implemented through 3dDeconvolve should accurately reflect the original data generation process. Therefore, the assessment of derived effects (e.g., averaging across 8 tasks) should be independent of the modeling process. In other words, the model dictates the derived effects, not the other way around.

Gang