Hi Anderson,
Thanks a bunch for this. I tried both approaches with success!
Three follow-up questions:
1. When I am including nuisance EV's in design2, should they be demeaned? (I know in general they would be, but since they are listed twice (once for each time-point), would this be an issue?
2. PPheights... Are they all just 1's here for each column in both design1 and design2 .mat and .con files?
3. I am having trouble understanding what each contrast represents in design1. In the example that you provided, what is the meaning of the first contrast listed? (I'm guessing it is g2>g1? )
C1 : t1>t2 8 8 0 0 0 0 0 0 0 0 F1
C2 g1>g2 t1>t2 8 -8 0 0 0 0 0 0 0 0 F2
...Sneaking a 4th question in... Would the other two possible contrasts be this(?):
C1 g1>g2 t2>t1 -8 -8 0 0 0 0 0 0 0 0 F1
C2 g2>g1 t2>t1 -8 8 0 0 0 0 0 0 0 0 F2
Thanks again. Hopefully others out there looking to do repeated-measures ANCOVA find assistance from this thread!
-Derek
_______________________________________
From: FSL - FMRIB's Software Library [[log in to unmask]] on behalf of Anderson M. Winkler [[log in to unmask]]
Sent: Wednesday, June 22, 2016 11:19 PM
To: [log in to unmask]
Subject: Re: [FSL] 2 X 2 repeated measures ANCOVA for TBSS
Hi Derek,
Please, see below:
On 22 June 2016 at 23:14, Derek Pisner <[log in to unmask]<mailto:[log in to unmask]>> wrote:
Hi Anderson,
Thanks for getting back to me and sorry for the slightly delayed response on my part!
The four additional covariates are Gender (i.e. 0's and 1's), Age, Time Since Head Injury (measured in days), and Compliance with Treatment (measured in days adhering to treatment regimen). For each of these covariates, we have only 1 measurement per subject (Age, Gender, and Time Since Injury are collected at Timepoint 1; Compliance with Treatment is collected at Timepoint 2)
Please, see this example from an earlier thread: https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;3cee56eb.1510
There are 2 designs inside: one for within-subject effects (design1, for the effect of timepoint across groups, and interaction group by timepoint), and one for between-subject effects (design 2, effect of group across timepoints).
For design1, which is likely the most interesting, you can use the design as is: there is no need to enter any of the nuisance variables, as the effect of these is already captured by the subject-specific intercepts (EV3 onwards).
For design2, if you decide to run it, include the 4 nuisance variables as extra EVs, repeating the values for the two timepoints for each subject.
Define one exchangeability block per subject and, in randomise, use the option "-e design.grp" for design1, and further, include also "--permuteBlocks" for design2. If you use PALM, it's possible to run both designs in sync, with the options "-eb design.grp, and "-whole" together with "-within", and "-corrcon" to correct across all contrasts.
There is no need to run any of the F-tests. They are shown more as curiosities -- the "-corrcon" supersedes it.
I should also note that we have 25 subjects in total (n=12 in Group A, and n=12 in Group B).
Who is missing? At any rate, the groups can be unbalanced, so different N for the groups aren't a problem.
Curious to see what you have up your sleeve. Many thanks in advance and hopefully this thread will aid others out there trying to piece together more complex repeated measures ANCOVAs using FSL's Glm...
Thanks for the feedback. Hope it helps!
All the best,
Anderson
-Derek
________________________________________
From: FSL - FMRIB's Software Library [[log in to unmask]<mailto:[log in to unmask]>] on behalf of Anderson M. Winkler [[log in to unmask]<mailto:[log in to unmask]>]
Sent: Thursday, June 16, 2016 11:54 PM
To: [log in to unmask]<mailto:[log in to unmask]>
Subject: Re: [FSL] 2 X 2 repeated measures ANCOVA for TBSS
Hi Derek,
Could you please tell a bit more about the four additional variables? For each of these, do you have 1 measurement per subject or 2 measurements per subject? If it's the second, in your hypothesis the covariate would affect the difference between timepoints, or the average of the two timepoints?
Depending on your answers, the design will be different.
All the best,
Anderson
On 16 June 2016 at 19:59, Derek <[log in to unmask]<mailto:[log in to unmask]><mailto:[log in to unmask]<mailto:[log in to unmask]>>> wrote:
Hi everyone,
Would someone be able to provide an example of a design matrix for a repeated-measures analysis comparing groups A and B across timepoints 1 and 2, with 4 additional covariates?
Assume the FA files are in the following order: Group A (Subject _1 tp1, Subject_1 tp2; Subject_2 tp1, Subject_2 tp2, etc.) and then Group B (Subject_3 tp1, Subject_3 tp2; Subject_4 tp1, Subject_4 tp2, etc.)
Is this still not doable in the Glm without having to use fslmaths, etc.?
Extra kudos to anyone who can solve this.
-Derek
|