> From [log in to unmask] Wed Jan 5 20:37:53 2000
> Date: Wed, 5 Jan 2000 14:33:14 -0500
> Subject: laterality as a covariate
Dear Jennifer and Lara,
> We have done an experiment in which we administered a drug during PET
> scans. Prior to scanning, we compiled indices of disease severity for
> each patient on both the left and right sides so that each patient has
> a "right" score and a "left" score. The hypothesis is that we will see
> greater activation with administration of the drug in the hemisphere
> contralateral to the most severe indices. For our analyses we used
> SPM95 and selected multisubjects with replications (on and off drug)
> and covariates (2- right and left). We then entered the appropriate
> scores for each patient as covariates of interest and then as a
> separate analyses as covariates of non-interest. Is this correct?
It depends on how the covariates entered into the design matrix. Your
hypothesis is about a condition x score interaction (and further a
condition x score x hemisphere interaction that would have to be
assessed separately). In other words you are asking if the condition
effect (drug - placebo) depends on the indices of disease severity (and
further is this dependence in the appropriate hemisphere). This is
assessed with a condition x score interaction using a single covariate
of interest for each of the two scores. For one score this should
comprise the subject-specifc score times plus one or minus one
depending on whether the scan was drug or baseline. The scores should
be mean corrected over subjects first. This covariate will model
differences in activation from subject to subject that can be predicted
by deviations from the mean severity.
One final point; it may be that your severity scores from each
hemisphere are correlated. To ensure efficient estimates (and
straightforward interpretation) I would take the sum and differences of
the scores before constructing the interaction covariates as described
above. This will orthogonalize the effects, giving a mean severity and
a laterization index that could be used to address your laterization
hypothesis above.
I hope this helps - Karl
|