Paul -
If your multiple regression (or those in previous recent emails on this topic) was implemented as a parametric modulation (using polynormial expansion option), then this is the one place where SPM does perform orthogonalisation: this page might help:
http://imaging.mrc-cbu.cam.ac.uk/imaging/ParametricModulations
But again, for all other normal regressors, there is no explicit orthogonalisation.
BW,R
-------------------------------------------------------
Professor Richard Henson
Assistant Director for Neuroimaging
MRC Cognition & Brain Sciences Unit
15 Chaucer Road
Cambridge, CB2 7EF
England
EMAIL: [log in to unmask]
URL: http://www.mrc-cbu.cam.ac.uk/people/rik.henson/personal
TEL +44 (0)1223 355 294 x501
FAX +44 (0)1223 359 062
MOB +44 (0)794 1377 345
-------------------------------------------------------
________________________________________
From: SPM (Statistical Parametric Mapping) [[log in to unmask]] on behalf of Jonathan Peelle [[log in to unmask]]
Sent: 27 May 2012 16:08
To: [log in to unmask]
Subject: Re: [SPM] Orthogonalization and Multiple Regression Question
Dear Paul,
> I have been using the multiple regression in SPM8 on some PET data. I want
> to compare two models (linear and exponential). I want to look at separate
> contrasts for the linear and the exponential fits. However when I put both
> of them in the plots do not come out looking exactly like a linear or
> exponential function, they have some strange bends in them. I think this
> has something to do with orthogonalizing the basis functions. Does anyone
> have any suggestions or know the location of the MATLAB code that implements
> the multiple regression? Can the orthogonslization be turned off?
The standard multiple regression shouldn't include any automatic
orthogonalization. However, if you're including both the linear and
exponential (quadratic?) predictors in the same model and they aren't
independent, then the parameter estimates will be reflecting the
independent contribution of each regressor, which may explain your
results.
I think the key thing would be to double-check the design matrix (X,
i.e. SPM.xX.X) and verify that your predictors look as expected, and
that the full model is giving a reasonable fit. If you actually want
to compare the linear vs. exponential fit objectively, then you may
want to construct separate models and look at the residuals, AIC/BIC,
or something similar.
Hope this helps!
Best regards,
Jonathan
--
Dr. Jonathan Peelle
Center for Cognitive Neuroscience and
Department of Neurology
University of Pennsylvania
3 West Gates
3400 Spruce Street
Philadelphia, PA 19104
USA
http://jonathanpeelle.net/
|