Print

Print
Dear Joy,

Yes, that's right. No need for site, sex or education if you have subject-specific regressors as nuisance.

Alternatively, and perhaps easier to see, you can just subtract one timepoint from another, in which case you'll have 1 image per subject, and things should look more familiar then.

All the best,

Anderson


On 2 December 2013 02:10, Joy Matsui <[log in to unmask]> wrote:
Hi FSL Experts.

I've been reading all the emails about creating design matrices and contrasts for doing a longitudinal analysis on TBSS outputs. For my purposes, I'm trying to evaluate DTI scalars along tract skeletons in 4 groups (1 control, 3 disease) where each subject has 2 time points. There are a few variables I want to control for: age, gender (1=male, 0=female), education, and site of collection. Based on one of the emails from the list (https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1302&L=fsl&D=0&1=fsl&9=A&J=on&d=No+Match%3BMatch%3BMatches&z=4&P=461525), I was thinking of making my design matrix as follows. I simplified my example to two groups, age (which is time), gender, education, and two different sites.

(Data)  (Grp)   (Time,controls) (Time,patients) (Gender,controls)       (Gender,patients)       (Educ,controls) (Educ,patients) (Site1,controls)        (Site1patients) (Site2,controls)        (Site2,patients)        ((Intercept,subj1)      (Intercept,subj2)       (Intercept,subj3)       (Intercept,subj4)
(Subj1,Visit1)  1       a11     0       1       0       9       0       1       0       0       0       1       0       0       0
(Subj1,Visit2)  1       a12     0       1       0       9       0       1       0       0       0       1       0       0       0
(Subj2,Visit1)  2       a21     0       0       0       7       0       0       1       0       0       0       1       0       0
(Subj2,Visit2)  2       a22     0       0       0       7       0       0       1       0       0       0       1       0       0
(Subj3,Visit1)  3       0       a31     0       1       0       12      0       0       1       0       0       0       1       0
(Subj3,Visit2)  3       0       a32     0       1       0       12      0       0       1       0       0       0       1       0
(Subj4,Visit1)  4       0       a41     0       0       0       6       0       0       0       1       0       0       0       1
(Subj4,Visit2)  4       0       a42     0       0       0       6       0       0       0       1       0       0       0       1

But then I came across this other email (https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1111&L=fsl&P=R28722&1=fsl&9=A&J=on&d=No+Match%3BMatch%3BMatches&z=4) that suggested that I don't need the gender, education, and site columns because the intercept subject columns account for gender, education, and site. Each subject's gender, education, and site also do not change between time point 1 and 2 (not time varying), which sort of helps me understand why columns for those variables aren't needed in a longitudinal analysis. However, site has always had to be accounted for in our past cross-sectional analyses so I'm a little nervous to drop site.

If anyone has suggestions, I'd be happy to hear them.


Thank you,

Joy