Dear Michael,
> I have a study where seven
> subjects were scanned 8 times, twice in each of 4 mood conditions: rest,
> induced negative mood, induced positive mood, and induced neutral mood.
> The order of presentation of the mood conditions was randomized across
> subjects (with the caveat that the two presentations of each mood
> condition
> were adjacent). In each condition, subjects rated their mood
> (essentiallly
> how positive or negative it was).
>
> I have already done the straightforward subtractions comparing the
> different mood conditions. However, this approach implicitly assumes that
> each subject reacted identically to the mood manipulation conditions,
> which
> isn't the case. What I would like to do is, instead of treating condition
> as a categorical proxy for mood, to correlate activity with the mood
> ratings directly.
>
> To do this, I have used "Multiple subjects: covariates only", selecting
> the
> images of each subject, and then entering the 56 ratings (7 subjects X 8
> conditions) as a single covariate vector (with the ratings corresponding
> to
> the appropriate image, of course). But from here I'm a bit flummoxed as
> to
> what would be the appropriate settings for the other options.
>
I'm assuming you are referring to the questions relating to the type
of interaction and centering of your covariate, because the other options
(global normalisation and thresholding) are the same for both analyses.
Given the usually large variability in these subjective one-item ratings of
clinically similar mood states, you might consider the 'cov x subject'
interaction option, and choose 'centering around subject mean', even though
that uses up 6 more d.f.'s.
> I'm also
> assuming that specifying 1 for the contrast to get positive correlations
> and -1 to get negative correlations is what I want, but I'm not sure.
>
That is correct if you select 'no interaction'. When you model
subject x cov interactions, group effects are defined by '1 1 1 1 1 1 1' and
'-1 -1 -1 -1 -1 -1 -1', respectively.
Good luck
Dick Veltman
|