I have a design with 2 within-subject factors: Tasks and TypeOfPair. Initially, I was using 3DANOVA3 to analyze the data, with the following command lines (a is TypeOfPair, b is Task, c is subject):
3dANOVA3 -type 4 -alevels 8 -blevels 3 -clevels 17
-mask $output_dir/stimuli/mask_anat.KQu+tlrc \3dANOVA3 -type 4 -alevels 8 -blevels 3 -clevels 17
-dset 1 1 1 $output_dir/stimuli/Subject1_SM_SM_Noise_Averaged+tlrc
-dset 1 1 2 $output_dir/stimuli/Subject2_SM_SM_Noise_Averaged+tlrc
Now I want to include the reaction times as a covariate in a mixed-effect model, using 3DLME. But, before that, I wanted to check whether 3DLME would give me the same results as 3DANOVA3 if I do not include the covariate. So I ran the following lines:
3dLME -prefix $output_dir/LME_922_1 -jobs 1
-gltLabel 1 ‘Action-Material’ -gltCode 1 'Task : 1Action -1Material’
-gltLabel 2 'Action+Material-2Noise’ -gltCode 2 ‘Task : 1Action +1Material -2*Noise’
Subj TypeOfPair Task InputFile
1_SM TP_TP Noise $input_dir/Subject1_TP_TP_Noise_Averaged+tlrc
1_SM TP_TP Action $input_dir/Subject1_TP_TP_Action_Averaged+tlrc
I found results that are qualitatively similar (overall, I find the same clusters), but the clusters have different sizes. I am not sure that I am using 3DLME correctly, though:
- which type of SS should I use? I used type 3 based on the examples in the documentation
- Shouldn’t I include the subjects as a random factor since type 4 in 3DANOVA3 corresponds to define c as a random factor? The program gets stuck when I try to do this, though.
- What is the difference between the glt and glf?
- Should I worry that the two methods give slightly different results?