Dear Elizabeth,
SPM.xX.W contains the whitening matrix used to compute weighted least
squares (WLS) estimates (it effectively whitens the data (makes the
noise i.i.d.) to render the ordinary least square (OLS) estimates
maximum likelihood (ML)). See the two-pass strategy in spm_spm.m with ReML.
You will find some more details in:
* slides 18-23 of Klaas' GLM talk:
http://www.fil.ion.ucl.ac.uk/spm/course/slides08-oct/02_General_Linear_Model.ppt
* Variance Components chapter of the HBF book:
http://www.fil.ion.ucl.ac.uk/spm/doc/books/hbf2/pdfs/Ch9.pdf
At last, I'm afraid I'm not familiar with ArtRepair but if SPM.xX.W
contains an identity matrix, it means that SPM will compute OLS estimates.
Best regards,
Guillaume.
Elizabeth Liddle wrote:
> Can someone explain what these weights are, and what they are used for?
>
> The reason I ask is that I have been using ArtRepair, and I notice that
> they are replaced by a much simpler matrix, and I wonder what we are
> losing, if anything.
>
> Thanks
>
> Elizabeth Liddle
--
Guillaume Flandin, PhD
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
|