Maybe I can explain the analysis that I am hoping to do based on behavioral results for better clarification on how to model the imaging data with 3dMEMA or another program if necessary (I have used 3dMVM and 3dLME but would rather use 3dMEMA if possible).
I have two groups of subjects (BN and HC). They differ behaviorally on the task with respect to age (so I am looking for an interaction) and they differ continuously with respect to another variable (STAI). There is no main effect of subject group. I am interested in examining all of these effects together in imaging to parallel the behavioral results. Would you recommend all subjects as one set with the following covariate matrix (example):
subj Dx STAI Age
BN122 -1 0.304146175737179 -1.93625024866797
BN125 -1 0.032355976142253 -0.152259975679931
SC01717 1 -0.511224423047599 -0.890688183913078
SC01792 1 -0.171486673553941 -1.00047220071234
Where group is dummy coded in the Dx column, and ages for group -1 (BN) are all negative Z-scored and with option -covariates_center Age=0 STAI=0 Dx=0?
Or should the groups be treated as two separate sets with a matrix like this:
subj STAI Age
BN122 46 30.8
BN125 42 24.3
SC01717 34 20.5
SC01792 39 20.1
and with options -covariates_center Age=25.16316 22.58696 -covariates_model center=different slope=different?
Or some other option I haven't thought of?
Essentially, Is the difference between treating the groups as dummy coded 1 and -1 and as two separate sets? (I have run both and the results seem almost the same).
For reference the behavioral model in R lookes like this:
glmer(Approach ~
Group * Age.Z + STAI.Z + (1|Subject),data.set)
where Group:Age.Z and STAI.Z have significant effects on the dependent variable.
Also, I am confused by the 3dMEMA help page, which says : "Option -unequal_variance may not be used in the presence of covariates with two groups." but example 2 clearly uses both options.
Thank you very much