Covariate interaction in two-sample 3dMEMA

AFNI Message Board
1

Hello,

I have been trying to test a difference of covariate effect between two groups. I got stat for that labeled “GroupA-GoupB_cov1_Tstat” using 3dttest++ as follows:


3dttest++ \
    -setA   GroupA \
        A01     beta.nii.gz \
        A02     beta.nii.gz \
        A03     beta.nii.gz \
        A04     beta.nii.gz \
    -setB   GroupB \
        B01     beta.nii.gz \
        B02     beta.nii.gz \
        B03     beta.nii.gz \
    -covariate \
        cov.txt

where ‘cov.txt’ is something like that:


name    cov1
A01       -4
A02      -22
A03       -6
A04       32
B01       32
B02      -28
B03       -4

Values of “cov1” were manually centered within each group.

How can we get the same stat using 3dMEMA? I tried like below, but a result looked different from that of 3dttest++. Is this a matter of coding covariates?


3dMEMA \
    -groups GroupA GroupB \
    -set    GroupA \
        A01     beta.nii.gz    tstat.nii.gz \
        A02     beta.nii.gz    tstat.nii.gz \
        A03     beta.nii.gz    tstat.nii.gz \
        A04     beta.nii.gz    tstat.nii.gz \
    -set    GroupB \
        B01     beta.nii.gz    tstat.nii.gz \
        B02     beta.nii.gz    tstat.nii.gz \
        B03     beta.nii.gz    tstat.nii.gz \
    -covariate          cov.txt \
    -covariate_center   cov1 = 0 0

Thanks,

Kosuke

2

How can we get the same stat using 3dMEMA? I tried like below, but a result looked different from that of 3dttest++.

These are two different models, so they are not expected to have the same results. How different between the two?

I’m not sure if it makes any difference, but these two lines

-covariate cov.txt
-covariate_center cov1 = 0 0

should be

-covariates cov.txt
-covariates_center cov1 = 0 0

Depending on your prior information, you may consider adding

-covariates_model slope=different

3

Thanks Gang,

“cov1X:t” reported by “-covariates_model slope=different” seems to be what I expected.

Kosuke