Dear Dr Ruff,

The results should be identical.  The parameter estimates are not affected by the other subjects, and the con*.img images are images of parameter estimates.  A covariate for a given subject will just include a column of zeros for every other subject, so this covariate can contribute nothing to modelling any other subject's data.

(The t and z statistics are different in these two cases, however, since in the multi-subjects fixed-effects analysis the within-subjects residual variance is estimated for all of the subjects together.  Provided all of the subjects have reasonably similar variance, though, the results will be tend to be fairly similar.)

Best wishes,

Richard Perry.

Dear SPMers and staff,
 
I wonder what is better for a random effect analysis :
 
1. To use con*.img images from each single-subject analysis (little SPM.mat easuy to load)
 
or
 
2. To use con*.img images from single-subject contrasts of a global multi-subjects analysis (big SPM.mat not so easy to load...)
 
 
Thanks a lot
 
Serge RUFF
INSERM U455
 

--
from: Dr Richard Perry,
Clinical Lecturer, Wellcome Department of Cognitive Neurology, Institute of Neurology, Darwin Building, University College London, Gower Street, London WC1E 6BT.
Tel: 0207 679 2187;  e mail: [log in to unmask]