Just a couple more questions...Should I be using the -D option in randomise,
and should my nuisance variables (age, 1/V-scaling) be demeaned? I'm not too
clear on when and what variables should be demeaned.
Thanks,
Ryan
-----Original Message-----
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
Of Jesper Andersson
Sent: Wednesday, December 16, 2009 4:47 PM
To: [log in to unmask]
Subject: Re: [FSL] randomise design matrix: modeling group interactions
Dear Ryan,
> I have a problem similar to the one entitled, "Re: Help with
> interaction
> matrix in FEAT" from May 2, 2009. I want to model interaction
> effects in
> randomise for a VBM analysis. I have 2 groups (alc, control), 2
> sexes (boys,
> girls), and 2 nuisance variables. I want to test (1) the main
> effects of
> group membership and (2) sex, along with (3) the group*sex
> interaction. The
> interaction effect will show me if one sex is more susceptible to the
> effects of heavy drinking than the other. From what I've read, I
> need to use
> two real EVs for each of my 2 original EVs (group and sex) because
> this
> isn't a repeated measures or paired design. The numbers of
> alcoholics/controls and boys/girls are not exactly balanced.
>
> This is how I set up my design matrix, assuming 1 subject per group.
>
> Design:
>
> alc/boy alc/girl con/boy con/girl nuisance_demeaned
> 1 0 0 0 x
> 0 1 0 0 x
> 0 0 1 0 x
> 0 0 0 1 x
>
> I set up my contrast to include [(con/boy) - (alc/boy)] and [(con/
> girl) -
> (alc/girl)] in an F-test.
>
> Contrast:
>
> alc/boy alc/girl con/boy con/girl nuisance F1
> C1 1 1 -1 -1
> 0 alc>con?
> C2 -1 -1 1 1
> 0 con>alc?
> C3 0 -1 0 1 0 x
> con/girl>alc/girl?
> C4 -1 0 1 0 0 x
> con/boy>alc/boy?
>
> Will F1 will test if the volume loss due to alcohol dependence is
> different
> between boys and girls?
You are practically there already. C3 looks at the effect of alcohol
on girls and C4 at the effect of alcohol on boys. The interaction is
simply the difference between the effect of alcohol on girls and the
effect of alcohol on boys, i.e. C3-C4, which would amount to [1 -1 -1
1]. This is a "directional" contrast and will look at "areas with
greater negative effect on girls than on boys" (if we assume that
anything positive in a con-alc contrast is "negative". If you want to
look at the opposite effect you simply negate that contrast to [-1 1 1
-1].
Good luck Jesper
|