Esa -
> We have an activation study with three conditions (neutral (no emotion),
> emotion 1, emotion 2) and four replications of each condition. Totally
> there are 12 scans per subject. We have also two different groups of
> subjects (10 and 11 subjects). The sequence of emotions is the same
> across all subjects. The total amount of scans is 252.
>
> The hypotheses are following:
> 1) To test the differences between conditions within both groups
> separately (neutral vs emotion 1, neutral vs emotion 2)
> 2) To test the interaction between group and differences between
> conditions (previous contrasts)
>
> We analysed the data using The Full Monty and treated each replication
> as a separate condition. However, we are not sure whether this is the
> right way to draw population inferences.
No, between-scan and between-subject variability are confounded.
> Do we need a random effects analysis for testing our hypotheses?
Yes, particularly when comparing across groups as in hypothesis 2).
> If the random effects analysis is appropriate, how to implement it ?
For 1), create separate contrast images for each subject in one group
(from subject-specific contrasts in your Full Monty SPM, collapsing
across replications). Then enter these into a one-sample t-test for
within-group RandFX tests.
For 2), put the contrast images from both groups into an unpaired
two-sample t-test. A [1 -1] F-contrast (or [1 -1] and [-1 1]
t-contrasts)
will then test for group x condition interactions.
---------------------------8-{)}-------------------------
DR R HENSON EMAIL [log in to unmask]
Wellcome Department of
Cognitive Neurology TEL (work1) +44 (0)20 7833 7483
12 Queen Square TEL (work2) +44 (0)20 7833 7472
London, WC1N 3BG FAX +44 (0)20 7813 1420
URL: http://www.psychol.ucl.ac.uk/rik.henson/index.html
---------------------------------------------------------
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|