Dear all,
after plenty of archive trawling and web-crawling I still need some
reassurance regarding a number of issues around doing a 2nd level "basic
models" (SPM 99) paired t-test. (in fact some authors seem to even suggest
that one should avoid it at all cost!)
First of all, as far as I understand, this does not strictly constitute an
RFX analysis, as there are more than 1 con.img per subject. Would this be
referred to as a mixed-effects analysis?
On a more practical note, I have difficulty interpreting the design matrix,
or rather how to define the contrasts: When I pick e.g. 12 pairs of con.img,
the design matrix will look as follows: First 2 columns represent the 2
conditions, then there is one column per subject, with the associated gray
rectangles descending diagonally across the matrix. Is it correct that in
this case, my contrast of interest (i.e. paired t-tests of cond A versus B)
would simply be [1, -1, 0, 0, 0, ...0]? (Intuitively I would have thought
I'd need to enter a [1, -1] for each subject, but given the matrix only
displays none column per subject this must be patently wrong).
Another slightly confusing issue concerns the use of global scaling. While
some SPM "how-to's" warn adamantly about the dangers of scaling, from
archive discussions it appears that one MUST use global scaling at the first
level of analysis in order to use the con.img for 2nd level analysis, is
this correct? From some of the documentation and discussions it would appear
that this is implemented automatically in SPM 99, but is that true even if I
chose "global scaling: none" at the model estimation stage?
Many thanks for any helpful comments,
Tobias.
Tobias Egner, PhD
Cognitive Neuroscience & Behaviour
Faculty of Medicine
Imperial College London
St Dunstan's Road
London W6 8RF
UK
Ph: (+44) 208 846 7281
Fax: (+44) 208 846 1670
|