Dear Leann,
> I have been modeling a study in a particular way, but am wondering if a
> different model may perhaps be more relevant. I have a group of 10
> subjects, each of which have received drug and placebo on different
> occasions. They also, however, differ in regard to previously-assessed
> mood effects in response to drug (i.e. 6 subjects reacted in one way
> and 4 reacted in another way when their mood was assessed in response
> to a dose of the drug given many months prior to their being scanned).
> What I had previously done was a multi-study, different conditions
> analysis with eight contrasts. After entering drug and placebo scans
> for the group of 6, then drug and placebo for the group of 4, the
> contrasts were:
>
> 1 -1 0 0 drug - placebo in group of 6
> -1 1 0 0 placebo - drug in same group
> 0 0 1 -1 drug - placebo in group of 4
> 0 0 -1 1 placebo - drug in same group
> 1 -1 -1 1 drug - placebo in group of 6 vs. group of 4
> -1 1 1 -1 placebo - drug in group of 6 vs. group of 4
> 1 -1 1 -1 overall drug - placebo
> -1 1 -1 1 overall placebo - drug
>
> Now, however, after reading past messages which have been posted, I
> wonder if I wouldn't be better off doing a multi-subject, conditions
> and covariates design using mood effect as my covariate of interest. Is
> this a more-appropriate design than the one I had previously chosen?
> Could anyone tell me how I would input this in SPM? How many
> covariates of interest should I choose? Just one? And how would this be
> entered in order to contrast the mood effects which seem to
> differentiate the 6 subjects from the other 4 subjects?
I think your analysis is very good as it stands. Effectively you have
looked for a mood score x drug interaction but have assumed that the
mood factor is categorical (i.e. group of 4 or group of 6). Using
covariates would give a similar model but allowing for parametric
variation, in the mood factor, within each group. To do this analysis
you would model just two conditions and use a covariate of interest to
model the interaction explicitly. This covariate would be a vector of
[mean corrected] mood scores (one for each scan) multiplied by 1 for
all drug scans and -1 for placebo scans. You would then test for this
single effect with a contrast like [0 0 1] or [0 0 -1].
With best wishes,
Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|