 |
 |
Dear Peter, thank you very much! That helps a lot! All the best, Philipp > On 18. Feb 2020, at 10:24, Zeidman, Peter <[log in to unmask]> wrote: > > Dear Philipp > >> I have a question regarding the interpretation of DCM results under mean centering. I understand that when mean centering, the A-matrix reflects the mean connectivity across experimental conditions (as opposed to connectivity during implicit baseline when not mean centering). But does mean centering also change the interpretation of the B-matrix, i.e. is a connectivity change under experimental condition 1 now relative to the other conditions? This seems crucial to be able to say whether a connection is excitatory or inhibitory under different conditions. > > If you mean-center the inputs, then the B-matrix parameters will be relative to the mean connectivity over conditions (encoded by the A-matrix). If you don't mean-center, then they will be relative to the baseline connectivity (encoded by the A-matrix). If it's easier, you can think of this like a general linear model, in which the design matrix has column of ones (the intercept), followed by some other covariates. You'll get the same difference in interpretation depending on whether or not you mean-center the covariates. > >> Specifically, using mean centering, I found that a connection is negatively modulated under experimental condition 1. When I apply formula 3 in Zeidman et al. (2019 - tutorial 1) to calculate the resulting Hz change, I find that the normally positive connection indeed becomes negative. Can we now conclude that this normally excitatory connection becomes inhibitory (region A inhibits region B) during experimental condition 1? Or can we only say that the connectivity is lower during condition 1 than during (the average of) the other conditions? > > You can conclude that the connection is more negative for condition 1 than the average connectivity across experimental conditions (i.e., more inhibitory or less excitatory, if it's the between-region connections, in the default DCM for fMRI model). > > Hope that helps, > Peter