Hi,
>> I've probably lost the plot in fumbling through various stats papers and
>> fMRI textbooks, but for the sake of my mental health (among other things),
>> any insights would be greatly appreciated!
>>
>> As the con images which form the data input into the second level are simply
>> weighted first-level parameter estimates, then it would seem that the
>> residual variance (in ResMS, which is not brought to the second level) is
>> used only for FFX inference and does not play a role in RFX as implemented
>> in SPM.
>>
>
> I don't think this is _literally_ true. See for example equation (19) in
> http://www.fil.ion.ucl.ac.uk/spm/doc/books/hbf2/pdfs/Ch12.pdf
> ("Random-Effects Analysis," W.D. Penny and A.J. Holmes). The accompanying
> text states, "Thus, the variance of the estimate of the population mean
> contains contributions from both the within-subject and between-subject
> variances."
>
> It _is_ true that the subject-level variance isn't explicitly taken to the second
> (group) level.
>
The subject specific ResMS is not taken to the second level. However,
the error in the estimation of the parameters is taken to the second
level, and will contribute to the error in estimation of the group mean
parameter. The parameter estimates from the first level are after all
just approximate estimates, not the "true" subject's parameters.
bw
James
--
--------------------------------------
Dr James Rowe
Wellcome Trust Senior Research Fellow in Clinical Science,
and Consultant Neurologist,
Cambridge University Department of Clinical Neuroscience,
Box 83, Addenbrooke’s Hospital,
Cambridge,
CB2 2QQ
UK
Tel: +44 (0)1223 273630
|