 |
 |
Dear Antanas, Log-evidence integrates model fit (which is related to explained variance) with model complexity i.e. number of active sources and their prior covariance. Therefore, the two measures do not necessarily go together. To compare models log-evidence is a better measure. Best, Vladimir On Tue, Aug 25, 2009 at 10:28 AM, Spokas Antanas<[log in to unmask]> wrote: > > Dear Vladimir, > > Please help me out here. For imaging source reconstruction model, for grand > avarage of each subject group under each condition I get two results: > variance explained and model log evidence . Could you please explain why for > some group/condition I get less log evidence (more negative value) but more > variance explained relative to other group/condition where I get opposite > (even so source estimates are pretty much with the same confidence levels)? > Does log evidence not directly related to variance explained here? Thank you > loads! > > Regards, > Antanas > ________________________________ > From: Vladimir Litvak <[log in to unmask]> > To: Spokas Antanas <[log in to unmask]> > Cc: [log in to unmask] > Sent: Tuesday, 25 August, 2009 6:33:19 > Subject: Re: [SPM] locations of stored ppm's and time courses of estimated > sources files after eeg imaging source reconstruction > > Dear Antanas, > If you know the MNI coordinates of your source of interest you can put them > in the little input box above the MIP button and then press the MIP button. > It will give you the source waveform with confidence intervals for the > nearest mesh vertex. I hope that's what you need. > Best, > Vladimir > On 25 Aug 2009, at 01:38, Spokas Antanas wrote: > > Dear Vladimir, > > I would like to go back to your earlier answer (below this letter). I do > understand its not very clean method to fish for the sources only from eeg > data, but even if I have several source candidates for my paradigm, I can > not put all the sources in 'K.mat' file for restriction of reconstruction > solutions, for at any time point there are some artifact sources (serving > functions not perteining to my study) and in the end it gives me better > evidence for unrestricted solution. Now, if I do choose to use SPM estimated > sources I do need their probabilities or confidence levels at some points of > time. Forgive me for repeating myself but reconstruction results would give > me only the most active source at any point of time and its confidence > level, but if my other source activity does not ever show up as having > highest activity then even if I see it activated pretty clearly in PPM, > there is no way for me to know its estimation likelihood. Please tell me > where I am loosing it here. Thank you for your help. > > Regards, > Antanas > > ________________________________ > From: Vladimir Litvak <[log in to unmask]> > To: Spokas Antanas <[log in to unmask]> > Cc: [log in to unmask] > Sent: Saturday, 15 August, 2009 15:06:49 > Subject: Re: [SPM] locations of stored ppm's and time courses of estimated > sources files after eeg imaging source reconstruction > > Dear Antanas, > Time course and PPMs are not stored but computed on the fly by the function > spm_eeg_invert_display based on output of the inversion which is stored in > the header file. Using the GUI is the simplest way to access them. If you > want to do statistics on activations you should use the standard pathway > described in the manual. > Best, > Vladimir > > On 15 Aug 2009, at 14:55, Spokas Antanas wrote: > > Dear SPM, > > Could you please help me to locate files of stored ppm's and time courses of > estimated sources after eeg imaging source reconstruction. Thank you very > much! > > Regards, > Antanas > > > > > > >