Hi
I am currently looking to set up the right design and contrast for a longitudinal study. And I was wondering if I could use a similar design and contrast as described here.
I have more experience with analysis using SPM, but I am now analyzing data with MRtrix which relies on randomise function of the FSL toolbox for its statistical analysis. I have been reading the manual and different papers, but I am still quite unsure how to correctly set up the design and contrast.
I have the following study design:
2 groups (males and female songbirds)
DTI was measured repeatedly at 6 different time points to monitor how their brain changes structurally over time.
My research question has 2 parts:
-where there is an overall difference between male and female songbirds. It is quite known that there are sex differences in the songbird brain in brain regions involved in song.
-How the brain develops over time. Over different seasons, brain regions involved in song increase in size.
This is what I have figured out so far: (correct me if I am doing something wrong)
To assess the group difference I will use a separate model, where I average over time, using a two-sample t-test. (based on the paper of McFarquhar 2019). And this would look like example 6 of Winkler et al. 2014
To assess the changes over time, and possible interactions between male and females things are more complicated. I have come up with the following design. But have some questions.
Subj EB VG Time ยต EV1 EV2 EV3 EV4 EV5 EV6 EV7 EV8 EV9 EV10 EV11 EV12 EV13 EV14
1 1 1 t1 1 1 0 0 0 0 0 0 0 0 0 1 0 0 0
1 1 2 t2 1 0 1 0 0 0 0 0 0 0 0 1 0 0 0
1 1 3 t3 1 0 0 1 0 0 0 0 0 0 0 1 0 0 0
1 1 4 t4 1 0 0 0 1 0 0 0 0 0 0 1 0 0 0
1 1 5 t5 1 0 0 0 0 1 0 0 0 0 0 1 0 0 0
1 1 6 t6 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0
2 2 1 t1 1 1 0 0 0 0 0 0 0 0 0 0 1 0 0
2 2 2 t2 1 0 1 0 0 0 0 0 0 0 0 0 1 0 0
2 2 3 t3 1 0 0 1 0 0 0 0 0 0 0 0 1 0 0
2 2 4 t4 1 0 0 0 1 0 0 0 0 0 0 0 1 0 0
2 2 5 t5 1 0 0 0 0 1 0 0 0 0 0 0 1 0 0
2 2 6 t6 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0
3 3 1 t1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0
3 3 2 t2 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0
3 3 3 t3 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0
3 3 4 t4 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0
3 3 5 t5 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0
3 3 6 t6 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
4 4 1 t1 1 0 0 0 0 0 1 0 0 0 0 0 0 1 0
4 4 2 t2 1 0 0 0 0 0 0 1 0 0 0 0 0 1 0
4 4 3 t3 1 0 0 0 0 0 0 0 1 0 0 0 0 1 0
4 4 4 t4 1 0 0 0 0 0 0 0 0 1 0 0 0 1 0
4 4 5 t5 1 0 0 0 0 0 0 0 0 0 1 0 0 1 0
4 4 6 t6 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0
5 5 1 t1 1 0 0 0 0 0 1 0 0 0 0 0 0 0 1
5 5 2 t2 1 0 0 0 0 0 0 1 0 0 0 0 0 0 1
5 5 3 t3 1 0 0 0 0 0 0 0 1 0 0 0 0 0 1
5 5 4 t4 1 0 0 0 0 0 0 0 0 1 0 0 0 0 1
5 5 5 t5 1 0 0 0 0 0 0 0 0 0 1 0 0 0 1
5 5 6 t6 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1
6 6 1 t1 1 0 0 0 0 0 1 0 0 0 0 0 0 0 0
6 6 2 t2 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0
6 6 3 t3 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0
6 6 4 t4 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0
6 6 5 t5 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0
6 6 6 t6 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
with the following contrast
C1 overall time effect 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 F1
C2 overall time effect 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 F1
C3 overall time effect 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 F1
C4 overall time effect 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 F1
C5 overall time effect 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 F1
C6 Interaction 0 1 0 0 0 0 -1 0 0 0 0 0 0 0 0 F2
C7 Interaction 0 0 1 0 0 0 0 -1 0 0 0 0 0 0 0 F2
C8 Interaction 0 0 0 1 0 0 0 0 -1 0 0 0 0 0 0 F2
C9 Interaction 0 0 0 0 1 0 0 0 0 -1 0 0 0 0 0 F2
C10 Interaction 0 0 0 0 0 1 0 0 0 0 -1 0 0 0 0 F2
I have repeated measures, so I need to include subject measure in the model (EV11, EV12, EV13, EV14), so that the subject error is included.
How should the subject factor be included using ones and zero's. I do not understand where the 1/3 and -1/3 in the design_yawu.ods come from.
EV7 EV8 EV9 EV10
1/3 0 0 0
1/3 0 0 0
1/3 0 0 0
- 1/3 1/3 0 0
- 1/3 1/3 0 0
- 1/3 1/3 0 0
0 - 1/2 1/2 0
0 - 1/2 1/2 0
0 0 - 1/2 1/2
0 0 - 1/2 1/2
0 0 0 - 1/3
0 0 0 - 1/3
0 0 0 - 1/3
And I have to make use of exchangeability blocks.
Since I am interested in an effect that occurs over time within subjects, I should use the -within option.
What kind of error option should I use?
Can I use the default exchangeable errors (-ee) option? But this may violate the assumption of compound symmetry?
Or do you recommend to use the Independent and symmetric errors (-ise) option?
Do I need to specify the variance matrix similar to the one I would use in the group analysis?
Is the matrix constructed in the right way? I use 'treatment coding', dropping the last indicator variable of the last time point and of the last subject in each group.
I want to check with the experts if I am doing the right thing, before proceeding with the analysis.
Kind regards
Jasmien
########################################################################
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
|