Dear Alexandra,
> The message below discusses using the beta images to calculate signal
> change. My question is: When you say multiply the design matrix by
> the parameter vector does that mean multiply the contrast (e.g., 1 0
> 0 -1) vector times the parameter vector (in this case .4097, .3342,
> .4533., .1301)? And then divide the result by the fifth element in
> the vector (the estimate of the session effect)?
The assumption here is that the session effect or constant basis
function estimates the mean of a time-series taking into account the
effects modelled by the other basis functions. One could then express
any changes fitted by basis functions by reference to this estimated
voxel-specific mean. Alternatively, one can also express signal changes
in a given voxel with reference to the global mean over voxels which is
automatically scaled session-specific to 100 by default (grand mean
scaling). That is why plots in SPM are by default in percent with
respect to the estimated global mean over voxels. Maybe the best way to
think about the first approach is to compute a re-scaling of the y-axis
in a voxel-specific adjusted data plot by multiplying the y-axis with
100/(parameter estimate of the session effect) (where I assume that the
session effect basis function is a vector of ones and that the global
mean is scaled to 100. Both assumptions should be the case in a default
SPM analysis).
> Also, with regards to negative beta values. While I understand that
> negative values don't necessarily mean a deactivation (just that the
> value is less than the mean parameter estimate), doesn't that mean
> that the condition is relatively deactivated compared to the mean of
> the parameter estimates?
Sorry, I'm a bit confused... What do you mean by the 'mean parameter
estimate'? The mean parameter estimate over space?
> On a related note: I have been mystified by plots of
> event/epoch-related activity in which ALL of the conditions are below
> the baseline. How can this be? I thought the baseline was the mean
> of the parameter estimates.
Still slightly confused. All parameters (one parameter for each basis
function, which is a column in the design matrix) are estimated
simultaneously.
Hope this helps, Stefan
--
Stefan Kiebel
Functional Imaging Laboratory
Wellcome Dept. of Cognitive Neurology
12 Queen Square
WC1N 3BG London, UK
Tel.: +44-(0)20-7833-7478
FAX : -7813-1420
email: [log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|