Dear Peter,
> When plotting the response estimate of an arbitrary point of activation, I
> see some unlikely things. The response decreases the first half of the scans
> (when going rest -> ativation) and the response increases on the second half
> of the scans (when going rest -> activation), the strange thing is that the
> increases and decreases happen in perfectly straight lines (which is very
> unlikely) and that the second half of the response is almost symmetrical
> with the first half of the response.
> Has someone a clue what might be the cause of such a strange reponse estimate ?
The estimated response to a measured time-series is a linear combination
of the temporal basis functions specified in the general linear model.
If these basis functions are e.g. solely box-car like, it would be
absolutely ok to observe a fit (estimated response in terms of a linear
combination of the underlying basis functions) which look over time like
straight lines. Different basis functions (e.g. the other choices of
basis functions in the fMRI module or in the event-related plugin) would
yield different fits, which do look smoother...
In practice, remember that the plot in the results section in SPM96
displays the adjusted data i.e. the estimated effects of the covariates
of no interest have been removed from both the measured data and the
associated response estimate.
Hope this helps, Stefan
--
Stefan J. Kiebel
Dept. of Neurology
Friedrich Schiller University
Philosophenweg 3
07740 JENA
Germany
Tel. : +49 3641/9-35285
FAX : -35399
email: [log in to unmask]
WWW : www.neuro.uni-jena.de
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|