Will Penny wrote:
> Dear Donald,
>
> MCLAREN, Donald wrote:
> > Dear Will,
> >
> > I am reading Ch.12 on RFX analyses and am running into difficulty
> > understanding one part of it.
> >
> > Section 2.2: "In Step 3, we have speciļ¬ed that only one contrast per
> > subject be taken to
> > the second level. "
> >
> > If you are feeding only one image into the second-level, then you are
> > able to estimate the between-subject variance. However, your equations
> > list a within-subject variance term as well. In the case when you enter
> > only one contrast per subject, does the within-subject variance term
> > equal 0? If not, how is estimated?
>
>
> The Summary Statistic (SS) approach to RFX does not explicitly
> manipulate within subject variances. But because what is taken up to
> the second level are *sample means*, these samples will contain
> variability. Just the right amount of variability, it turns out, such
> that *on average* the SS approach is equivalent to the corresponding
> maximum likelihood estimate, as is shown mathematically in the summary
> statistics section of:
>
> http://www.fil.ion.ucl.ac.uk/~wpenny/publications/spm-book/rfx.pdf
>
> This equivalence assumes the within-subject variances are equal. It also
> turns out that the within-subject variances have to be very different
> (order of magnitude) to throw off the SS approach ie. it is robust.
>
> Best wishes,
>
> Will.
>
>
--
William D. Penny
Wellcome Trust Centre for Neuroimaging
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/
|