Hi DCM experts,
I found the big difference between Exceedance Probabilities given by Bayes model selection without family comparison and the family level inference.
Firstly, I conducted the BMS without family comparison using the following codes:
matlabbatch{1}.spm.dcm.bms.inference.dir = {pwd};
for s=1:50
for m=1: 100
matlabbatch{1}.spm.dcm.bms.inference.sess_dcm{s}.dcmmat{m,1}=[sub{s} '/DCM_model' num2str(m) '.mat'];
end
end
matlabbatch{1}.spm.dcm.bms.inference.model_sp = {''};
matlabbatch{1}.spm.dcm.bms.inference.load_f = {''};
matlabbatch{1}.spm.dcm.bms.inference.method = 'RFX';
matlabbatch{1}.spm.dcm.bms.inference.family_level.family_file = {''};
matlabbatch{1}.spm.dcm.bms.inference.bma.bma_no =0;
matlabbatch{1}.spm.dcm.bms.inference.verify_id =1;
spm('defaults', 'FMRI');
spm_jobman('run', matlabbatch);
A figure with bars describing the exceedance probability of each model popped up. The exceedance probability of the best model is very low (<0.1) in this figure.
Then I conducted family comparison using the following codes：
matlabbatch{1}.spm.dcm.bms.inference.dir = {pwd};
for s=1:50
for m=1: 100
matlabbatch{1}.spm.dcm.bms.inference.sess_dcm{s}.dcmmat{m,1}=[sub{s} '/DCM_model' num2str(m) '.mat'];
end
end
matlabbatch{1}.spm.dcm.bms.inference.model_sp = {''};
matlabbatch{1}.spm.dcm.bms.inference.load_f = {''};
matlabbatch{1}.spm.dcm.bms.inference.method = 'RFX';
for fa=1:4
matlabbatch{1}.spm.dcm.bms.inference.family_level.family(fa).family_name = name{fa};
matlabbatch{1}.spm.dcm.bms.inference.family_level.family(fa).family_models =sapce{fa}
end
matlabbatch{1}.spm.dcm.bms.inference.bma.bma_yes.bma_famwin = 'famwin';
matlabbatch{1}.spm.dcm.bms.inference.verify_id = 1;
spm('defaults', 'FMRI');
spm_jobman('run', matlabbatch);
A figure of the exceedance probabilities of each model also popped up except the family comparison result. The exceedance probability of the best model is very high(>0.6) in this figure. And the best model in this figure is not the best model mentioned before. The exceedance probabilities of other 49 models are also different.
I am confused. Could you tell me why this happens?
Which kind of exceedance probabilities can be reported in the paper?
Best regards,
Ying
