Dear Steven,
> While looking through the code of spm_spm_ui, I noticed an option for
> Multivariate Analysis. There was very little documentation for this
> design. Furthermore, I don't recall seeing this option on any of the
> GUI menus.
> Does anyone know how this analysis option works and how to access it ?
This is a hidden feature of SPM that is accessed by invoking spm_spm_ui
from the command wondow (as oposed to the GUI). There is documentation
in spm_spm_ui.m and spm_spm
% Multivariate analyses
% =====================
%
% Mulitvariate analyses with n-variate response variables are supported
% and automatically invoke a ManCova and CVA in spm_spm. Multivariate
% designs are, at the moment limited to Basic and PET designs.
Effectively, if you select the multivariate option SPM performs a
ManCova (as opposed to an AnCova) at each voxel and reduces the results
to a single dimension (the first Canonical Variate). It has not been
written up and is, therefore, not endorsed by peer review (this is why
it is not an explicit feature). Its future is under consideration.
The next release of SPM will have a non-sphericity option which allows
univariate analyses of data of different types. This provides
exactly the same functionality as ManCova but retains exactly the same
formalism for correcting p values and expressing the results as SPM{F}
and SPM{T} as conventional mass univariate approaches.
We might therefore expand the non-sphericity option and remove the
multivariate one.
I hope this clarifies things.
Karl
|