Hi Marc
Including the PPI in your model should indeed affect the estimate of the main effects. The reason is that the PPI regressor and the PSY regressor are partially correlated. In the GLM, variance will only be assigned to one regressor if it is uniquely explained by that regressor. When two regressors are partially correlated, like PSY and PPI, the bit of the variance previously assigned to PSY, which is now also correlated with PPI, will not be assigned to either regressor. Hence the estimate for PSY changes when you add PPI to the model.
The same explanation applied to the changing results when you include more than one PPI in the analysis. The PPI regressors will be partially correlated with each other - especially if they are drawn from the same ROIs or task regressors - and each one will affect the estimability of the others. For this reason it is advisable to run separate analyses for the separate PPIs you are interested in.
Jill
-----------
Dr Jill O'Reilly
FMRIB Centre, Dept Clinical Neurology, Oxford University
Phone +44 1865 222466
-----------------------------
Begin forwarded message:
From: Marc Seal <[log in to unmask]<mailto:[log in to unmask]>>
Date: 5 February 2010 01:13:35 GMT
To: [log in to unmask]<mailto:[log in to unmask]>
Subject: Re: [FSL] PPI queries & quirks
Reply-To: FSL - FMRIB's Software Library <[log in to unmask]<mailto:[log in to unmask]>>
hello Jill and PPI experts,
We have been running series of PPI analyses in FSL following the useful
guidelines on Jill O'Reilly's webpage.
However, we have noticed a couple of quirks with the PPI output vs. the
default FEAT 1st Level output. That is, the inclusion of the Physiological
Regressor (PHYS) and PPI in the EVs will change the observed output of the
"Psychological" statistical maps (the EV of interest) compared to the output
produced in the default FEAT analysis.
We had assumed that the inclusion of the Physiological Regressor (PHYS) and
PPI Intercation EV would make no difference to the modeling of the
"Psychological" EVs of interest. But it does!
So does this difference matter and does it impact on the PPI output map?
Finally, as a shortcut we were trying to include multiple Physiological
Regressor (PHYS) and PPI's in a 1st level analysis (ie trying to model > 2
PPI's int he same analysis) and that produces very different output maps
compared to running only one PPI in the analysis.
Any thoughts or suggestions?
cheers,
Marc
---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask]<mailto:[log in to unmask]> http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------
|