Hi Gang,
I ran 3dLME on some files containing area under the curve for responses to pictures (negative or neutral, within subjects), and did a few GLTs to clarify effects and interactions. I have Age and Sex as covariates, and Group (PTSD or HC) as a between subjects variable. Here is my command:
3dLME \
-prefix group_analysis/emo_reg_FACE_AUC_t1t6_LME_noRS.nii \
-jobs 6 \
-mask ${rootPath}/data/emo_reg/rois/PFC_Insula_mask.nii \
-model 'Valence*Group+Age+Sex' \
-qVars 'Age' \
-qVarCenters '13.55' \
-ranEff '~1' \
-SS_type 3 \
-num_glt 6 \
-gltLabel 1 'NEG-NEUT' -gltCode 1 'Valence : 1*NEG -1*NEU' \
-gltLabel 2 'PTSD-HC' -gltCode 2 'Group : 1*PT -1*HC' \
-gltLabel 3 'NEG_PTSD-HC' -gltCode 3 'Valence : 1*NEG Group : 1*PT -1*HC' \
-gltLabel 4 'NEUT_PTSD-HC' -gltCode 4 'Valence : 1*NEU Group : 1*PT -1*HC' \
-gltLabel 5 'PTSD_NEG-NEUT' -gltCode 5 'Valence : 1*NEG -1*NEU Group : 1*PT' \
-gltLabel 6 'HC_NEG-NEUT' -gltCode 6 'Valence : 1*NEG -1*NEU Group : 1*HC' \
-dataTable \
Subj Group Age Sex Valence InputFile \
subj003 HC 8 F NEG /Volumes/Vol6/YouthPTSD/data/emo_reg/_003/tents_censored/FACE_NEG_tent_auc_2t13.nii \
subj005 HC 9 M NEG /Volumes/Vol6/YouthPTSD/data/emo_reg/_005/tents_censored/FACE_NEG_tent_auc_2t13.nii \
...
As expected, my output has subbricks for Valence F, Group F, Age F, Sex F, and Valence:Group F. Because I wanted to see the direction of the effects for Valence and Group, I did GLTs for those. The GLT for PTSD-HC doesn't come close to resembling the Group F results. However, the GLT for PTSD-HC at Negative Valence is almost identical to the Group F result. This makes me think that the Main Effect F tests are actually not collapsing across levels of other factors. Is this possible, and if so, why is it testing at one level of the other factor? Is there a way I could more explicitly force the tests to be done by collapsing across variables?
Thanks!
Rick Wolf