My answer is more a question actually. The others may know it better.
Wouldn't it be better to covary out RTs (also) at the first level
analysis (i.e. parametric modulations)?
2009/7/8 Guillaume Sescousse <[log in to unmask]>:
> Dear SPMers,
> I'm comparing patients and controls on a simple contrast of a cognitive
> task. Region A is hypoactivated in patients, however I want to make sure
> that this hypoactivation is not due to the fact that patients are slower.
> For this purpose I want to covary out the effect of response times by
> modelling them at the second level (mean response time for each subject).
> I'm using the multiple regression model in SPM2, but I have several
> - should I model the 2 groups with 1 covariate (i.e. 1 1... -1 -1) and do a
> contrast of  OR with 2 covariates (i.e. 1 1... 0 0 and 0 0... 1 1) and
> then do a contrast of [1 -1] ?
> - same question for the RTs: should I model the RTs of all subjects
> (controls and patients) in 1 covariate or in 2 separate covariates ?
> I have tried these different ways and observed that the results were pretty
> different, but I'm not expert enough in statistics to decide whether one of
> them is "more correct". Moreover I found conflicting comments on the SPM
> Any help will be greatly appreciated.
> Thank you very much.
> Guillaume Sescousse, PhD student
> 'Reward and decision making' group
> Centre de Neuroscience Cognitive
> CNRS UMR5229 - UCB Lyon 1
> 67 Bd Pinel, 69675 Bron, France
> tel: 00 33 (0)4 37 91 12 44
> fax: 00 33 (0)4 37 91 12 10