Griselda,
As Jesper implied, your covariate of interest is linearly
dependent with subject effects, preventing the unique estimation of either.
The only way to estimate the partial slope on your covariate of interest
with the type of design you have (which I am presuming is one scan per subject)
would be to not include subject effects in your model. This would
effectively treat subject effects as error components. I am not precisely
sure how to implement this in SPM, but my guess would be to "pretend"
(in the context of SPM) that all of the data was from a single subject.
Sincerely,
Eric
Eric Zarahn
University of Pennsylvania
>
> Dear SPM'ers
>
> We have tried to use SPM99 to investigate linear correlations between
> regional brain uptake of SPECT tracer in our psychiatric samples (OCD
> and major depressive patients) and the severity of symptoms as assessed
> using standardized rating scales.
>
> We have attempted a design that used to work with SPM96, which is
> “multi-subject: covariates only” but this did not work now (error
> message: “unestimable block effects”). We have also attempted the
> ”single subject: covariates only” option with no success either
> (“segmentation fault”).
>
> I wonder if we could have some help on that. Many thanks...
>
> Griselda Jara Garrido
>
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|