Dear Arkan, I feel that the what you mentioned in your first email is classical validation of an identified model [ref 1]. However I understood the second email as follows: 1-you have two groups of subjects. 2- and you would like to use DCM and estimate the connectivity parameters in the two groups. 3- Then you labeled these groups and construct parameters of the classification machine (train phase). 4- You fit a DCM to a new data set (test data) and based upon the estimated connectivity parameters you would like to know which label can explain the test data. I was wondering if you can inform us is the above steps are what you have in mind, please? Many thanks, Best Regards, Amir [ref 1] i.e., first identify some parameters of a model using train data sets. Then you can use this model and compare its output EEG/fmri with your test data (which seems here that its predicted DCM signal) and calculate mean squared error between these signal. On Wednesday, 20 June 2018, 11:13:26 BST, Arkan Al-Zubaidi <[log in to unmask]> wrote: Dear Peter, I would like to to know the accuracy of spectral DCM model to predict the connectivity and hemodynamic parameters for a subject from DCM parameters of group of subjects. Can I do something similar to the following lines? If this code is correct, how can I setup the hemodynamic parameter in field cell array? Mean_effect=ones(20,1); % one group of 20 subjects CP_4_5 = [GCM(1,1).Ep.A(4,5); GCM(2,1).Ep.A(4,5); ........; GCM(20,1).Ep.A(4,5)]; % connectivity parameters from connection 5 to 4 field = {'A(4,5)'}; M.X = [Mean_effect,CP_4_5]; [qE,qC,Q]=spm_dcm_loo(GCM,M,field); Best regards, Arkan ________________________________________ From: Zeidman, Peter <[log in to unmask]> Sent: 20 June 2018 11:13 To: Arkan Al-Zubaidi; [log in to unmask] Subject: RE: [SPM] Accuracy of DCM model Dear Arkan The Leave-One-Out (LOO) function is intended to check whether your effect sizes (DCM parameters) are large enough to predict subject-level covariates (e.g. group membership or behavioural scores). However, you say you want to predict unseen subjects' effective connectivity parameters, in order to validate DCM. I don't understand that. Please could you clarify the validation you are trying to perform? All the best Peter -----Original Message----- From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Arkan Sent: 19 June 2018 16:55 To: [log in to unmask] Subject: [SPM] Accuracy of DCM model Dear SPM list, I know that the spm_dcm_loo function is for distinguishing (classify) between two groups if the subject is unseen. Here, I would like to validate the spectral DCM model to predict the unseen subject parameters (effective connectivity and hemodynamic) for one group. How can I use the spm_dcm_loo function in this case? I would like to plot the expected parameter and the real parameter (not the classification). Regards, Arkan