Dear Stephen,
> Hi - I have a simple question regarding the RFX theory and practice. It
> has probably been asked before - sorry. In the original paper and
> poster, it looks like the input to the RFX equation is a group of
> parameter estimate images (or constrasts of parameter estimate images).
> A simple t-test is then carried out on this group of images.
Absolutely correct.
> However, I
> think I am right in saying that in practice SPM uses t images as input
> to the RFX equation, and not parameter estimate images. Therefore, is
> there a difference because:
In SPM99, you would implement a second level analysis (SPM96 is
conceptually the same, but in practice not as simple to implement as in
SPM99) by choosing some contrast images (weighted parameter estimates
images) as input into one of the basic models. This is valid both in
theory and in practice...
> a) it doesn't make any difference in practice
>
> OR
>
> b) using t values instead of parameter estimates (i.e., including
> scan-scan variance information) is an improvement, as the higher
> variance inputs are then de-weighted?
Choosing t-images as input for a second analysis is different from doing
a 2nd level analysis based on the weighted parameter estimates images
(again in theory and practice). There was once a helpful discussion
about this on the maillist, a good handle on these messages is given by:
http://www.mailbase.ac.uk/lists/spm/1999-01/0122.html
> Is using t values theoretically as valid as using parameter estimates?
In theory, both are valid, it just depends on the question you have. In
practice: Assuming that you are after the average population effect, you
should use the contrast images for your analysis...
Stefan
--
Stefan Kiebel
Functional Imaging Laboratory
Wellcome Dept. of Cognitive Neurology
12 Queen Square
WC1N 3BG London, UK
Tel.: +44-171-833-7490
FAX : -813-1420
email: [log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|