Dear Pablo
I'm going to suggest how I would model this, in the hope it makes the more detailed question you've asked resolve itself :-)
You could create a separate PEB for each group, with two columns: mean and pre vs post (ensuring that the pre vs post regressor is mean-centred, so that the constant can be interpreted as the mean). Then, the PEBs-of-PEBs would include all three groups. The design matrix will include three columns. You could either have: 1) mean over groups, 2) controls vs patients, 3) patients group 1 vs patients group 2. For example:
X = [1 1 0;
1 -0.5 1;
1 -0.5 -1 ];
X(:,2:3) = X(:,2:3) - mean(X(:,2:3));
Or you could have the following three regressors in your peb-of-pebs: 1) control group, 2) patient group 1, 3) patient group 2:
X = [1 0 0
1 1 0
1 0 1]
Notice that I'm not centring the second option, so that the first column is interpretable as the controls.
I hope that helps,
Peter
-----Original Message-----
From: SPM (Statistical Parametric Mapping) <[log in to unmask]> On Behalf Of Pablo Maturana
Sent: 22 July 2022 12:30
To: [log in to unmask]
Subject: [SPM] PEBxPEBs question
⚠ Caution: External sender
Dear Peter and DCM experts,
I am doing a PEB analysis in a sample with 3 groups (2 groups of patients and a control group) and each subject takes two measurements (PRE-POST).
I did a PEBxPEBs analysis to analyze the interaction of TIME and GROUPS. I have formed 6 GCMs divided by group and session, then a first level PEB for each group as follows:
1 1
1 -1
the ones column by number of subjects
The result are 3 PEBs representing PRE-POST of each group, which allows me to see the connections over time.
The next analysis were two PEBxPEBs to compare the control group vs the patient groups separately and the patients with each other to avoid rank deficient. The design represents row 1 the controls and rows 2 and 3 patients, so the design of my matrix was as follows:
Group_1 Group_2
1 1 0 1 1 0
1 -1 1 1 0 -1
1 0 -1 1 -1 1
My results for column 3 in both groups are not the same, although I am just inverting the sign. I think this is because column two is not the same and according to the PEB model which corresponds to a GLM I understand that:
z = Xb + e
Where b are the regression parameters (PEB parameters). So for a single subject s:
z(s) = b1 + b2 * x2(s)
Where x2 is the second column of the design matrix. The b1 parameters are commonalities and b2 are the group difference parameters. I think that the reason for the change in the sign of the third column may be due to the fact that the second column differs between group 1 and group 2.
I did several tests trying to model the change in both groups of patients and I used this matrix model:
Group_1 Group_2
1 1 1 1 0 0
1 -1 0 1 1 -1
1 0 -1 1 -1 1
In the Group_1 it works fine but in the Group_2, I expected to find that the column 2 and 3 have the same connections but with reversed sign and it does not happen.
Could you give me any suggestion on how to model the group difference in a PEB model?
|