dear will and list, i have been facing the same problem as thio kellerman ( see post below). i am trying to use Bayesian estimation for a two-sample t-test at the group level. the OLS estimation works fine but the bayesian estimation gives the following error message with the debugger. (i am using matlab 6.5) --------------------- PEB Iteration : 3 ...2.825956e-04 Non-sphericity : ...REML estimation ??? Conversion to unknown from double is not possible. Error in ==> /app/spm2/spm_spm_Bayes.m On line 338 ==> PPM.Cb(v,v) = P{2}.C(j,j); K>> ------------------ i would be extremely grateful for any help/suggestion, bhisma On Thu, 9 Sep 2004 12:15:02 +0100, Will Penny <[log in to unmask]> wrote: >Thilo Kellermann wrote: > >> Dear SPMers, >> >> I would like to claculate PPMs for a 2nd level group analysis and the function >> spm_spm_Bayes.m seems to work fine until fitting the model and writing the >> parameter images. I guess it is right after the beginning of the post >> estimation when the function crashes and returns the error message pasted >> below. Does anybody know what might have gone wrong and how to proceed with >> post estimation? >> >> Thanks a lot, >> Thilo >> >> ------------------------------------ >> >> PEB Iteration : 2 ...2.564839e-05 >> PEB Iteration : 1 ...2.969170e-01 >> PEB Iteration : 2 ...1.684169e-05 >> PEB Iteration : 1 ...9.235810e-02 >> PEB Iteration : 2 ...4.617268e-06 >> >> Non-sphericity : ...REML estimation >> ??? Error while evaluating uicontrol Callback. >> >> >> ??? Conversion to unknown from double is not possible. >> >> Error in ==> /nfsusr/local/SPM/spm2/spm_spm_Bayes.m >> On line 338 ==> PPM.Cb(v,v) = P{2}.C(j,j); >> >> > > >Is this the first time you've used PPM analysis ? Did it work before on >other data ? > >What version of matlab are you running ? > >This is the first time PPM.Cb is referred to so I'm wondering if later >versions of matlab require you to declare variables in some way - only a guess. > >What is your second level design matrix ? Just a column of 1's ? > >To really find out what's happening you'll need to use the debugger. >Type 'dbstop if error' in the command window before pressing the Bayesian button. >Then see if there are problems with any of the variables. > >Best, Will.