D. Menghini wrote:
> Dear Will Penny, John Ashburner, and other SPM2 experts
>
>
>
> This is a question that relates both to ANOVA/ANCOVA models and VBM.
>
>
>
> I'm running a VBM study to delineate local differences in GM volume between
> a group of dyslexics and a control group.
>
>
>
> For the statistical analysis I’m using the “basic model” in spm2.
>
> I have a question about non-sphericity correction option that gives me
> different estimates (SPM.xVi.h) for the two groups depending on the model I
> use.
>
> First, I applied ONE WAY ANOVA with non sphericity correction option and
> global calculation (mean voxel value (within per image full mean/8 mask )
> to obtain an estimate of the subject-specific over-all GM volume. In this
> analysis the non-sphericity values are similar for the two groups
> (SPM.xVi.h values for dyslexics are 0.9772 and for the control group
> 1.0228).
>
> Then I used the GM volume values obtained in the first analysis
> (SPM.xGX.rg) as covariate in the ANCOVA with non sphericity correction and
> obtained SPM.xVi.h values for dyslexics of 1.2857 and for the control group
> 0.7052.
>
>
>
> I checked the mask images in the ANOVA and ANCOVA analyses: they are
> identical, indicating that the same brain regions are considered.
>
>
>
> 1. I assume that my design matrix in the ANCOVA and ANOVA are identical
> (since the GM-volumes are modelled in both cases). Is this true?
>
No, I don't think so.
Have a look at the design matrices SPM.xX.X.
Because the designs are different, SPM will potentially use a
different subset of voxels to estimate the nonsphericity (see eg. line 472 in spm_spm.m)
Best,
Will.
>
>
> 2. In that case, why are the SPM.xVi.h values so different for the two
> models?
>
> Could it be that the sphericity correction is more effective (the hyper-
> parameters differ more between groups) for the ANCOVA rather than for the
> ANOVA?
>
>
>
> 3. In a VBM analysis, is it plausible that the variability in GM volume is
> so different between groups? Have other people observed similar effects?
>
>
>
> Thanks in advance,
>
> Deny Menghini
>
>
--
William D. Penny
Wellcome Department of Imaging Neuroscience
University College London
12 Queen Square
London WC1N 3BG
Tel: 020 7833 7475
FAX: 020 7813 1420
Email: [log in to unmask]
URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
|