3dLMEr with missing longitudinal data and several covariates

AFNI Message Board
1

Dear AFNI experts,

I'm interested in looking at group effects (3 groups) and time effects (baseline and follow-up) while covarying for the effect of sex, age, and site (the data is from 3 research sites). I have an unequal number of subjects in each of the 3 groups, and I'm also missing follow-up data for some of the subjects. My understanding is that 3dMVM can handle unequal groups, while 3dLMEr can handle both the unequal groups and the missing follow-up data. Is this correct?

I've also tried to set up the 3dLMEr model but I'm not sure it is correct. Any advice would be greatly appreciated.

-model  'group*time+sex+age+site'    \
-SS_type 3    \
-qVars 'age'    \
-gltCode  group1-group2    'group : 1*group1 -1*group2'    \
-gltCode  group1-group2_grp_by_time    'group : 1*group1 -1*group2  time : 1*baseline -1*followup'    \
-gltCode  group1-group3    'group : 1*group1 -1*group3'    \
-gltCode  group1-group2_grp_by_time    'group : 1*group1 -1*group3  time : 1*baseline -1*followup'    \
-gltCode  group2-group3    'group : 1*group2 -1*group3'    \
-gltCode  group2-group3_grp_by_time    'group : 1*group2 -1*group3  time : 1*baseline -1*followup'    \
-dataTable    \

Thank you,
Dave

2

Hi Dave,

My understanding is that 3dMVM can handle unequal groups, while 3dLMEr can handle both the unequal groups and the missing follow-up data. Is this correct?

Yes, your understanding of the usage of 3dMVM and 3dLMEr is correct. Here are some suggestions to enhance your model specification:

  1. Within-Individual Factor (time): If the variable time is a within-individual factor, consider specifying the model as follows:

    -model 'group*time+sex+age+site+(1|Subj)'

  2. Centering the Variable age: Properly centering continuous variables can improve result interpretability. I recommend centering the variable age. For more details on centering, refer to this document.

  3. Interactions: Consider exploring interactions between sex, age, and site with group and time. These interactions can provide valuable insights into the relationships among these variables.

Gang Chen