Dear Peter,
| I try to analyze a very simple data set consisting of three (3)
| PET-images acquired in three conditions. I try to find those voxel
| whose intensity increases with the image number (1 < 2 < 3). I use
| SPM96.
|
| So I choose single-ss conditions & covariates:
| 3 conditions
| 1 covariate of interest (1 2 3)
| 0 confounding covariates
| AnCova
| GrandMeanScaling
| 2 contrasts (1 -1)
| and harvest the following error messages:
|
| ??? Error using ==> spm_invFcdf
| df must be strictly positive
|
| Error in ==> /data/sw/spm96/spm_spm.m
| On line 166 ==> UF = spm_invFcdf(1 - UFp,Fdf);
|
| Error in ==> /data/sw/spm96/spm_spm_ui.m
| On line 703 ==>
| spm_spm(V,H,C,B,G,CONTRAST,ORIGIN,THRESH*GX,HCBGnames,P,0,[])
|
| ??? Error using ==> error
| Error while evaluating callback string.
|
| Please tell me what the heck went wrong here.
Despite the confusing error messages, the problem is very simple: Your
design has more parameters than there are images, and is therefore
completely overdetermined and unestimable.
Note also that the covariate (1 2 3) is linearly dependent with the
condition effects. With three scans you can't really do much
statistics, since you can't estimate a variance for any model with more
than two parameters.
Even a model with just the covariate (1 2 3), and no global
normalisation (or proportional scaling) will only have one residual
degree of freedom - hardly enough! Basically, you need more scans to
make any statistical inference.
Hope this helps,
-andrew
+- Dr Andrew Holmes [log in to unmask]
| -___ __ __ Wellcome Department of Cognitive Neurology - |
| ( _)( )( ) Functional Imaging Laboratory, Stats & |
| ) _) )( )(__ 12 Queen Square, Systems |
| (_) (__)(____) London. WC1N 3BG. England, UK |
+---------------------------------------http://www.fil.ion.ucl.ac.uk/-+
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|