Date: Sat, 5 Aug 2017 00:05:40 -0300
From: "Anderson M. Winkler" <[log in to unmask]>
Subject: Re: Help with GLM for testing Correlation Difference (single-group repeated measures for TBSS)
Hi Yuwen,
Could you give more details? In particular, could you clarify the following:
- How many FA images each subject has? One or two? If just one, this
doesn't need a repeated measures design?
=> Hi Anderson,
Each subject only has 1 FA image, but each subject has two different types of symptoms (Symptom A and Symptom B).
- What were the designs used for the analyses that lead to the different
results for Symptom A and Symptom B? This helps to understand what
dissociation you are talking about.
=> It was a simple correlation matrix ran between FA and Symptom A, and then ran again between FA and Symptom B, within the same group (with the same FA images).
- Are Symptoms A and B: (a) actual symptoms that can be present, absent, or
graded into some scale, or (b) are these *sets* of symptoms (hence the
count)? This would clarify why in your design you have, in addition to the
count, there are two binary regressors.
=>They are symptom count only.
- Can a given subject present symptoms A and B simultaneously? This also
relates to the binary EVs in your design.
=>Yes, any subject can present the two types of symptoms simultaneously.
- Are these symptoms associated to some extent? (e.g., if A and B are sets
of symptoms, are some of them, e.g., pain and nausea, present in both
sets?) This wouldn't necessarily change the design, but might allow
suggesting other approaches.
=>Symptom A and B are different sets of symptoms and they are not correlated. None of the symptoms are present in both sets, and the symptom counts of A and B are not statistically correlated either.
Now I guess if I want to generate a design matrix that can test the differences of the slope of FA in relation to symptom A versus the slope of FA in relation to symptom B, can I use the following matrix?
SubjectID EV1 EV2 EV3
SymA SymB Covariate (e.g., Age)
Subject1 2 1 21
Subject2 0 3 25
Subject3 6 2 27
Subject4 1 7 45
Subject5 5 0 18
Subject6 0 1 35
Subject7 3 2 40
Subject8 2 0 22
Subject9 9 7 28
Subject10 1 4 37
Contrasts:
Title Ev1 Ev2 EV3
C1 SlopeA>SlopeB 1 -1 0
C2 SlopeB>SlopeA -1 1 0
Thank you very much!
Yuwen
On 3 August 2017 at 22:42, Yuwen Hung <[log in to unmask]> wrote:
> Dear Expert:
> If I have a single group of 30 patients with symptom A and symptom B, in
> these patients TBSS has identified dissociated FA correlation maps with
> symptom A count and symptom B count separately, and now I just need to be
> able to directly test the difference of the correlations (the slope)
> between FA and Symptom A count versus between FA and Symptom B count for
> these 30 subjects. How should I set up the GLM design matrix?
>
> I know if I treat the Symptom A and Symptom B count scores as independent
> measures, I can set it up like below (take 10 subjects for example):
> EV1 EV2 EV3 EV4
> SymA SymB CountA CountB
> 1 0 2 0
> 1 0 0 0
> 1 0 6 0
> 1 0 1 0
> 1 0 5 0
> 0 1 0 1
> 0 1 0 2
> 0 1 0 0
> 0 1 0 7
> 0 1 0 4
>
> How can I make SymptomA and SymptomB into repeated measures? e.g., Should
> I make them in one EV and assign 1 and -1 for different symptoms? And for
> CountA and CountB scores should I keep them as 2 EVs or put them in 1 EV?
> And I also need to control for each subject's repeated measures effect by
> adding 30 EVs with value 1 assigned to the 2 symptom scores for each r,
> right? And what should I put for contrasts if I just need to see the effect
> of slopeA>slopeB and slopeB>slopeA? My last question is that some of the
> symptom count is 0 and I assume this wouldn't mess up the design matrix
> right?
>
> Your help is greatly appreciated,
> Yuwen Hung, PhD
> McGovern Institute for Brain Research, MIT
>
>
|