Dear Nicolas,
At 04:28 PM 6/21/99 +0200, Nicolas Costes wrote:
| What is the signification of the 'Dogdy population main effect' model in
| PET/SPECT section ?
| In what is it different from the 'Population main effect' model ?
There are three models in the population comparisons section of SPM99b
PET/SPECT:
i) Population main effect: 2 cond's, 1 scan/cond (paired t-test)
ii) Dodgy population main effect: >2 cond's, 1 scan/cond
iii) Compare-populations: 1 scan/subject (two sample t-test)
These are all for the case when there is only one scan per subject per
condition, where, provided the model assumptions are met and appropriate
contrasts used, inference extends to the population. That is, fixed effects
of interest in these models are the population parameters.
The first two models are for assessing the population main effect from a
single group of subjects each with one scan per condition. Model (i) is for
the case when there are only two conditions. The only valid contrast
compares the two conditions, and the test is essentially a paired t-test.
(There is a "Paired t-test" model in the "Basic Stats" section, but the
PET/SPECT one here also offers grand mean scaling and global intensity
normalisation options).
The second model allows you to enter more than two conditions. This is
therefore a blocked Anova design. The assumption that the residuals (data
less subject effect and condition effect) are independent, is an assumption
of variance sphericity. This is the dodgy aspect, as it is pretty unlikely
in practice that the subject and condition effects tell the whole story up
to IID errors. By contrasting two conditions here when there are three or
more conditions (3+ scans/subject) you are essentially doing a paired
t-test with the variance estimate pooled with that of the third, untested,
condition. Further, contrasts more complicated than simple comparisons of
conditions probably won't be valid for population inference.
The third model is for comparing two populations when you have one scan per
subject. This is essentially a two-sample t-test.
Hope this helps,
-andrew
+- Dr Andrew Holmes [log in to unmask]
| ___ __ __ Wellcome Department of Cognitive Neurology |
| ( _)( )( ) Functional Imaging Laboratory, Stats & |
| ) _) )( )(__ 12 Queen Square, Systems |
| (_) (__)(____) London. WC1N 3BG. England, UK |
+------------------------------------- http://www.fil.ion.ucl.ac.uk/ -+
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|