Dear Martin Greetings. The problem is in spm_spm_Bayes.m when one uses Matlab R14 or greater. If you go to approx lines 294 and 301 this code : tmp = tmp + NaN*(~tmp); produces the NaN's. I have updated this in the FIL version of the file and hopefully it will make its way to wider distribution at a future update (or it may have already). In any case replace the above with this line in both places. tmp(~tmp) = NaN; You will have to re-estimate the dataset, but then all should be fine. Attached is also an updated version of spm_spm_Bayes.m Regards, Darren > -----Original Message----- > From: SPM (Statistical Parametric Mapping) > [mailto:[log in to unmask]] On Behalf Of Martin Kronbichler > Sent: Tuesday, January 30, 2007 4:12 AM > To: [log in to unmask] > Subject: [SPM] spm5 bayesian problems > > Dear SPM Experts, > > i have recently started to use variational bayesian first > level analyses for analysing presurgical language and motor > experiments because i had the feeling that the posterior > probalities are easier to communicate and easier to > understand (especially when making inferences on how likely > increased activatity for speech or motor is in regions near > to the lesions). > Generally, the bayesian analyses worked extremly well. > However i have stumpled across a strange result i do not > understand when analysing a block-design language experiment > (2 blockedc onditions: simple tones vs. auditory words > presented in 2 sessions). Whatever effect size threshold > (even completerly unrealistic values like > 1000) or > posterior probality i choose the PPM shows the whole brains > as exhibiting posterior prob. of 1 (i.e. the whole brain is > "activated"). SPM5 displays the following error message when > is estimate the contrasts: > > spm{P} image 1 : ...computing > Warning: Returning NaN for out of range arguments > > In /opt/spm5/spm_Ncdf.m at line 80 > In /opt/spm5/spm_contrasts.m at line 214 > In /opt/spm5/spm_getSPM.m at line 474 > In /opt/spm5/spm_results_ui.m at line 264 > ...written spmP_0001.img > SPM computation : > ...initialising > SPM computation : > ...done > SPM computation : > ...done > > I have tried to remove and add motion parameters, analyse > only one session, use different basis functions etc... but > whatever i do the problem persits. > Inspection of the files generated by SPM5 shows that the beta > and SDbeta images look reasonable. The spmP and con_SD images > however contain only NaNs.. > I might add that a conventional analysis results in > stastically highly signficant voxels... > Anybody has any idea what is causing this problem? > > Greetings from Salzburg, > > Martin > > > Martin Kronbichler, M.Sc. > --------------------------------------------- > Department of Psychology > > > Center for Neurocognitive Research > University of Salzburg > > Hellbrunnerstr.34, A 5020 Salzburg, Austria > e-mail:[log in to unmask] > Tel.:+43/(0)/662/8044-5162 > > Fax:+43/(0)/662/8044-5126 > > --------------------------------------------- > Department of Neurology > Center for Neurocognitive Research Christian-Doppler Clinic, > Paracelsus Private Medical University Ignaz Harrerstr.79, A > 5020 Salzburg, Austria e-mail:[log in to unmask] > Tel.:+43/(0)/662/4483-3966 > --------------------------------------------- >