# Second level analysis from 3dDeconvolve output

Dear AFNI Experts,
We’ve performed a first level fmri analysis for 48 subjects by means of the 3dDeconvolve command as following:

3dDeconvolve
-input \${tfMRI_input} -num_stimts 3
-stim_times 1 \${EVsFold}/cue.1D ‘SIN(0,20,10)’ -stim_label 1 \${fMRIName}_cue
-stim_times 2 \${EVsFold}/math.1D ‘SIN(0,20,10)’ -stim_label 2 \${fMRIName}_math
-stim_times 3 \${EVsFold}/qmath.1D ‘SIN(0,20,10)’ -stim_label 3
-polort 1 -fout -tout -bucket \${OutputF1}/\${fMRIName}_stats_SIN.nii.gz
-fitts \${OutputF1}/\${fMRIName}_fitt_SIN.nii.gz
-iresp 1 \${OutputF1}/\${fMRIName}_pos_iresp_SIN_cue.nii
-iresp 2 \${OutputF1}/\${fMRIName}_pos_iresp_SIN_math.nii
-iresp 3 \${OutputF1}/\${fMRIName}_pos_iresp_SIN_qmath.nii

Thus obtaining in the output bucket: Full_Fstat, _Tstat, _Coeff and _Fstat for each stim_lab.
We’d like to perform a second level group analysis using 3dttest. Which of the contents of the bucket we have to use? Which sub-brick of the statistical bucket dataset contains the contrast file for the t statistic and f statistic (thus for F the extra sum of squares)? Do we have to produce additional output using -gltsym?

Usually you should take the effect estimates (*_Coeff) to group analysis.

Do we have to produce additional output using -gltsym?

That is optional, which may simplify the group analysis.

Just curious: why are you modeling the hemodynamic response with ‘SIN(0,20,10)’? How long is your task duration?

Dear Gang

Fabio

Another thing…

of course we’ve tried to use the effect estimates, thus (let be C1 Coeff of the hrf term, C2 and C3 coeffs for 1st and 2nd derivative) we produced 1/3(C1+C2+C3) and tested it at the second level. However, we wonder whether is it more correct to bring to the second level the results of an F test. What do you think?

Thanks again

Fabio

Fabio,

I’m not so sure why you modeled the hemodynamic response with ‘SIN(0,20,10)’. It seems to me that ‘BLOCK(30,1)’ would make more sense to me.

we wonder whether is it more correct to bring to the second level the results of an F test. What do you think?

No, the F-stat shows the significance for a specific null hypothesis for each subject, and would not be a reasonable candidate for group analysis. You want to make some statistical inference based on the physical measurement or effect estimate.