Dear Louise, it's hard to say what is valid and/or sufficient with regard to the specification of a model space because it depends on the persuasiveness of your reasons to define a specific space... However, if you specify only models with endogenous connections and direct inputs of C1 only it is definitely not ok to claim that the obtained dynamics reflect those only during C1. Endogenous connections always reflect the average "dynamics" among the ROIs independent from any condition(s). If you are interested in the dynamics that are specific to C1, you are (by definition) interested in a modulatory input of C1, i.e. which connections are affected by condition C1. In your case I would suggest to setup new first-level SPM.mat files (without the need to really estimate these models). These new models also have two regressors where the first includes all (!) dynamic scenes and the second only those from condition C1. This approach is analogous to the attention-to-motion task by Friston and Büchel (see SPM manual). When you use these SPM.mats in the DCM model specification you can define direct inputs of all dynamic scenes (e.g. an early visual region). Then you specify the connections that you assume to be modulated by C1. By switching some of these connections on and off you define successively your model space and you will see which model(s) among those tested actually best describe your data. The new specification of the inputs in new SPM.mats has the advantage that you "simultaneously" (or better: implicitly) test the modulatory input of C1 against all dynamic scenes. If you simultaneously switch the modulatory inputs of all dynamic scenes according to those specified for C1, the modulatory inputs of C1 will reflect the additional rate of change attributable to C1 compared to all other scences. In other words, if the parameter is positive you can conclude that the respective connection is enhanced during C1 compared to all other scenes (or suppressed during C2) whereas it is suppressed when the parameter is negative (or the connection is enhanced during C2). Hope this helps although it might sound a bit complicated... Thilo On 01/24/2014 03:22 PM, Louise Kauffmann wrote: > Dear SPM-ers, > > I am currently trying to use DCM on fMRI data. My design includes three > conditions (modeled as three regressors) : > - C1: visual perception of dynamic scenes (short movies) > - C2: visual perception of the same dynamic scenes as in C1, but > time-reversed > - C3: Fixation periods > > the contrast of interest [C1>C2] reveals significant activation in 3 > regions, and I am interested in unerstanding the interactions between > those 3 regions, in the C1 condition. The reverse contrast revealed no > significant activation. My question is the following: > > Is it valid (and sufficient) to compare models varying only in their > endogenous connections (with no modulation) and specifying only C1 as > driving input? (and is it OK to think that the winning model would > reflect the dynamics between the 3 ROI, only during C1 condition?). If > it is correct, is there a published study using the same rationale? > Also, would it be valid to do the same with only C2 as driving input and > then show that the winning model in C1 is different from the winning > model in C2? > I know that it would be more appropriate to compare models with > connections modulated by C1 or C2, but in this case, driving and > modulatory inputs would be the same and I believe that it is not correct. > > Any help would be really appreciated. > Thanks, > > Louise > > > > -- > > Louise KAUFFMANN > > > Laboratoire de Psychologie et Neurocognition > > CNRS UMR 5105 > > Université Pierre Mendès-France > > BP47, 38040 Grenoble Cedex 9 France > > Tel: 04 76 82 54 00 > > http://web.upmf-grenoble.fr/lpnc/membre_louise_kauffmann > -- Thilo Kellermann RWTH Aachen University Department of Psychiatry, Psychotherapy and Psychosomatics JARA Translational Brain Medicine Pauwelsstr. 30 52074 Aachen Germany Tel.: +49 (0)241 / 8089977 Fax.: +49 (0)241 / 8082401 E-Mail: [log in to unmask]