Here is the tests that I have run. I am summarizing the afni commands to help with clarity.

3dMEMA -groups Controls Subjects -mask mask_group+tlrc \
-set Subjects \
Subject1 Subject1_REML+tlrc'[Cond1#0_Coef]' Subject1_REML+tlrc'[Cond1#0_Tstat]' \
Subject2 Subject2_REML+tlrc'[Cond1#0_Coef]' Subject2_REML+tlrc'[Cond1#0_Tstat]' \
Subject3 Subject3_REML+tlrc'[Cond1#0_Coef]' Subject3_REML+tlrc'[Cond1#0_Tstat]' \
Subject4 Subject4_REML+tlrc'[Cond1#0_Coef]' Subject4_REML+tlrc'[Cond1#0_Tstat]' \
-set Controls \
Control1 Control1_REML+tlrc'[Cond1#0_Coef]' Control1_REML+tlrc'[Cond1#0_Tstat]' \
Control2 Control2_REML+tlrc'[Cond1#0_Coef]' Control2_REML+tlrc'[Cond1#0_Tstat]' \
Control3 Control3_REML+tlrc'[Cond1#0_Coef]' Control3_REML+tlrc'[Cond1#0_Tstat]' \
Control4 Control4_REML+tlrc'[Cond1#0_Coef]' Control4_REML+tlrc'[Cond1#0_Tstat]'

The output BRIK for this command has the following sub-briks:
0) Controls:b
1) Controls:t
2) Subjects:b
3) Subjects:t
4) Subjects-Controls:b
5) Subjects-Controls:t
6) Controls:tau^2
7) Controls:QE
8) Subjects:tau^2
9) Subjects:QE
10) tau1^2/tau2^2
11) tau2^2/tau1^2

The resuls in general show that the differences between Subjects-Controls are negative. This is also reflected in the beta values with the beta values for the Control group larger than the Subject group.


I also ran 3dttest++ on the same datasets as follows

3dttest++ -setA Subjects \
Subject1 Subject1_REML+tlrc'[Cond1#0_Coef]' Subject1_REML+tlrc'[Cond1#0_Tstat]' \
Subject2 Subject2_REML+tlrc'[Cond1#0_Coef]' Subject2_REML+tlrc'[Cond1#0_Tstat]' \
Subject3 Subject3_REML+tlrc'[Cond1#0_Coef]' Subject3_REML+tlrc'[Cond1#0_Tstat]' \
Subject4 Subject4_REML+tlrc'[Cond1#0_Coef]' Subject4_REML+tlrc'[Cond1#0_Tstat]' \
-setB Controls
Control1 Control1_REML+tlrc'[Cond1#0_Coef]' Control1_REML+tlrc'[Cond1#0_Tstat]' \
Control2 Control2_REML+tlrc'[Cond1#0_Coef]' Control2_REML+tlrc'[Cond1#0_Tstat]' \
Control3 Control3_REML+tlrc'[Cond1#0_Coef]' Control3_REML+tlrc'[Cond1#0_Tstat]' \
Control4 Control4_REML+tlrc'[Cond1#0_Coef]' Control4_REML+tlrc'[Cond1#0_Tstat]'

The output of this command produces a BRIK with the following sub-briks:
0) Subjects-Controls_mean
1) Subjects-Controls_Tstat
2) Subjects_mean
3) Subjects_Tstat
4) Controls_mean
5) Controls_Tstat

The differences between Subjects-Controls generated here are largerly positive and the Subject means are largerly greater than the Control means.

I am trying to reconcile the differences between these two sets of results.