Dear hector, Looking among some fMRI/SEM studies I realised that the stacked model approach has been used to compare a specific anatomical model under different experimental conditions but most of the time the authors don't say anything about the goodness of fit (p value for example) of the base model I have some problem in understanding why in neuroimaging , SEM has been essentially used to search differences between two conditions while the majority of the studies in other scientific fields have search for models which remain true between different conditions. I think a major problem is the definition of "anatomical" model. If we assume that "anatomical" means that the model is a constant of the brain (without specifying if this means that the connections depicted by the model are physical neuronal linkages?) and that a path coefficient kA_B means that : if activity in area A is increasing two times then activity in area B is increasing of a factor kA_B * 2 , then I can not imagine that a variation of the kA_B coefficient occurs between two tasks that are performed by the same brain with only few seconds delay between the two tasks. For me, and this view is far to be supported by all people which apply SEM to neuroimaging, if we notice a significant difference between the connectivity patterns resulting from the analysis of the covariance matrix corresponding to the two tasks , I think that it means: - either, we have forgotten to include some areas in the model and these areas have an effect on the included areas. A difference (between the two tasks) of the residual variances associated to the areas in the model is a good argument for sustaining this hypothesis. - in one of the two tasks the variations of the input is so low that no fitting is feasible. However, this significant modification of the path coefficient clearly shows that this path is involved in the process sustaining the tasks. So, I feel that the result is more close to a result dealing with functional correlation. My questions are: 1) Is it possible to use SEM to evaluate not only the significant difference across different conditions but also the goodness of fit of the model? It is possible to test separately the model apply to the two sets of data (or to the whole data) and have the corresponding fittings 2) Is it valid to use, in a stacked model approach, an anatomical model that is not statistically significant (p value < 0.05 for example)? If we say "it is a difference between two things that do not exist", it looks strange. I feel it is like when in variance analysis we have a significant interaction without any main effect: it is difficult to conclude. I am not sure to give you some help with this comment and ,one more time, it is only my opinion. Pierre Pierre Fonlupt INSERM - Unité 280 Processus Mentaux et Activation Cérébrale 151 Cours Albert Thomas 69424 LYON CEDEX 03 Tél : 06 60 54 68 29