Hi,
> Model 2:
> Set up 3 EVs for the levels of ratings (binary 0/1 for off/on) and set up
> contrasts using the mean ratings for each level as weights.
Model 2 is the way to go. As you noted with model 1, having a single EV
restricts the relationship between ratings and stimuli. With 3 EVs and
using different contrasts you can ask a wider range of questions.
> Problem: How? Say, to test the hypothesis that brain activation is
> modulated linearly with the mean ratings (e.g. 162, 248, 356, for S1, S2,
> S3 respectively). Any F test needed?
No F test needed.
The best you can do with only three rating levels is:
EV1(S1) EV2(S2) EV3(S3)
positive linear -1 0 1
negative linear 1 0 -1
This can be setup similarly if you had more rating levels, e.g. 5:
EV1(S1) EV2(S2) EV3(S3) EV4(S4) EV5(S5)
positive linear -2 -1 0 1 2
However, note that this does NOT exclude higher order relationships
(you could set up contrasts to look for these as well - if there are
enough rating levels) and is only looking "locally" at the ratings
strengths used.
Cheers, Mark.
Mark Woolrich.
Oxford University Centre for Functional MRI of the Brain (FMRIB),
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK.
Tel: (+44)1865-222782 Homepage: http://www.fmrib.ox.ac.uk/~woolrich
|