Email discussion lists for the UK Education and Research communities

## SPM@JISCMAIL.AC.UK

#### View:

 Message: [ First | Previous | Next | Last ] By Topic: [ First | Previous | Next | Last ] By Author: [ First | Previous | Next | Last ] Font: Proportional Font
 LISTSERV Archives SPM Home SPM July 2013

#### Options

Subject:

Re: VBM methods - group comparisons

From:

Date:

Wed, 3 Jul 2013 12:35:41 +0100

Content-Type:

text/plain

Parts/Attachments:

 text/plain (56 lines)
 ```Hi Oskar > Hi experts, > I am trying to grasp differences in statistical design of VBM analyses. Consider these two cases: > > 1. First, we have two groups with diagnoses A (control) and B (disease 1) that we want to compare in a VBM analysis. The design matrix would be > [1 1 0 > 1 0 1] > and to check for regions where intensity in A is higher than in B we use the contrast [0 1 -1]. yes - except in SPM the constant is at the end of the design matrix X = [1 0 1 ; 0 1 1]; and thus your contrast [1 -1 0] shows A>B > > 2. In this case, we add two (diseased) groups, now having with 4 groups with different diagnoses A (control) B C D (different diseases) and a design matrix > [1 1 0 0 0 > 1 0 1 0 0 > 1 0 0 1 0 > 1 0 0 0 1] > To find for regions where intensity(A) is higher than intensity(B) we use a contrast [0 1 -1 0 0]. yes (but again remember that the constant should be shifted to the right hand side) > > As far as I understand, the main difference between 1 and 2 would be the estimation of the intercept (different now that C and D were added) and the template used (if I use a study specific template, which I probably should?). > > So, here is my question: If I want to compare subjects from different groups, should I make lots of investigations of type 1 above, or just one investigation with different contrasts like 2 above? In this kind of design it doesn't really matter as long as you test pair-wise like eg A vs another group -- now if you have multiple subgroups of patients, it is more elegant to use model 2 (and more numerically more efficient/stable) Here is a little demo A = [9 10 11]; B = [19 20 21]; C =[29 30 31]; Y = [A ; B ; C]; if you model A, B, C altogether ie X = [kron(eye(3), ones(3,1)) ones(9,1)]; the constant is 15 and A=-5 B=5 C = 15 so for A vs B you have a difference of 10 and A vs C a difference of 20 now running twice model 1 X = [kron(eye(2), ones(3,1)) ones(6,1)], would give for Y = [A ; B]; the constant is 10 and A=0 B=10 so for A vs B you have a difference of 10 for Y = [A ; C]; the constant is 13.33 and A=-3.33 B=16.66 so for A vs C you have a difference of ~20 (not in that case it is numerically less accurate) Cyril http://www.sbirc.ed.ac.uk/cyril -- The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336. ```