Dear Helka
Your design matrix isn't quite right - it's rank deficient, or in other words, columns two and three are perfectly (anti-)correlated. This means it cannot be estimated efficiently.
Here, you just need columns one and two - encoding the mean over groups and the difference between groups. You should mean-correct the regressors (except for the constant) to ensure that the first regressor can be interpreted as the mean over subjects. The change in scale due to mean-centring will slightly alter the results (it will alter the complexity part of the model evidence), but hopefully will give the same overall conclusion.
Best
Peter
-----Original Message-----
From: SPM (Statistical Parametric Mapping) <[log in to unmask]> On Behalf Of Helka Ozelo
Sent: 05 March 2020 22:48
To: [log in to unmask]
Subject: [SPM] PEB - design matrix
Dear SPM experts,
Sorry for my huge e-mail, but I have some doubts about DCM analysis. I have ran PEB analysis for 2 groups (groupA with 6 subjects and groupB with 4 subjects). In my design matrix I did 2 tests:
Test 1: Matrix Design with 1s and -1s
Mean A>B B>A
Subject1 1 1 -1
Subject2 1 1 -1
Subject3 1 1 -1
Subject4 1 1 -1
Subject5 1 1 -1
Subject6 1 1 -1
Subject1 1 -1 1
Subject2 1 -1 1
Subject3 1 -1 1
Subject4 1 -1 1
Test 2: Matrix Design with fractional values (to get mean zero)
Mean A>B B>A
Subject1 1 1/6 -1/6
Subject2 1 1/6 -1/6
Subject3 1 1/6 -1/6
Subject4 1 1/6 -1/6
Subject5 1 1/6 -1/6
Subject6 1 1/6 -1/6
Subject1 1 -1/4 ¼
Subject2 1 -1/4 ¼
Subject3 1 -1/4 ¼
Subject4 1 -1/4 ¼
I have got two distinct results:
BMA.Ep (Test1) – P>0.95
Mean group (10subj) groupA>groupB groupB>groupA
PCC – PCC -0,4030 -0,0031 0,0031
PCC – mPFC 0,0567 0,0169 -0,0169
PCC – LIPC 0,0370 0,0065 -0,0065
PCC – RIPC -0,0203 0,0014 -0,0014
mPFC – PCC 0,0569 0,0227 -0,0227
mPFC - mPFC -0,3733 -0,0050 0,0050
mPFC – LIPC 0,0268 0,0104 -0,0104
mPFC – RIPC 0,0131 -0,0125 0,0125
LIPC – PCC 0,0706 0,0097 -0,0097
LIPC – mPFC 0,0395 0,0148 -0,0148
LIPC – LIPC -0,2405 -0,0016 0,0016
LIPC – RIPC 0,2676 -0,0011 0,0011
RIPC – PCC -0,0333 -0,0031 0,0031
RIPC – mPFC 0,0153 -0,0142 0,0142
RICP – LIPC 0,1959 0,0040 -0,0040
RIPC - RIPC -0,3059 -0,0023 0,0023
Test 1: There are no differences between A and B for strong evidence (P>0.95).
BMA.Ep (Test2) – P>0.95
Mean group (10subj) groupA>groupB groupB>groupA
PCC – PCC -0,4030 -0,0279 0,0279
PCC – mPFC 0,0567 0,2020 -0,2020
PCC – LIPC 0,0370 0,0773 -0,0773
PCC – RIPC -0,0203 0,0168 -0,0168
mPFC – PCC 0,0569 0,2719 -0,2719
mPFC - mPFC -0,3733 -0,0519 0,0519
mPFC – LIPC 0,0268 0,1244 -0,1244
mPFC – RIPC 0,0131 -0,1502 0,1502
LIPC – PCC 0,0706 0,1154 -0,1154
LIPC – mPFC 0,0395 0,1770 -0,1770
LIPC – LIPC -0,2405 -0,0124 0,0124
LIPC – RIPC 0,2676 -0,0168 0,0168
RIPC – PCC -0,0333 -0,0370 0,0370
RIPC – mPFC 0,0153 -0,1703 0,1703
RICP – LIPC 0,1959 0,0452 -0,0452
RIPC - RIPC -0,3059 -0,0194 0,0194
Test 2: There a lot of differences between them. Furthermore, groupA>groupB and groupB>groupA are symmetrical. Is it right, please?
Thank you so much.
Helka
|