I believe that SPM5 DCM is slightly different from SPM2 DCM in terms of
the output function, translating activity into BOLD signal based on the
Balloon model. SPM5 benefited from the lates developments but SPM2 was
left "as is" when the developers focused on SPM5.
The general answer is thus to believe the latest software version which
should be an improvement over the last one. Unless you've got a good
reason to believe your old results.
If you used exactly the same data with SPM2 and SPM5 DCM, you could
compare the models to decide which one is better. The free energy
approximation wasn't implemented in SPM2 for sure but AIC and BIC should
be available(*) and you could estimate the (log) Bayes Factor,
indicating which model, SPM2 or SPM5, is optimal for your data.
(*) I never tried SPM2 DCM, so can't guarantee that AIC/BIC were
estimated along the parameters as this was the early days of DCM. Still,
if not provided, those numbers should be easily estimated from the
resulting DCM structure using the formulas in Will Penny's 2004 paper.
Tai-Li Chou a écrit :
> Dear SPM,
> We have used the same DCM models in both SPM5 and SPM2, but got
> different results. In the same connections, we got negative modulatory
> effects in SPM5 and got positive modulatory effects in SPM2.
> By the way, we used the same VOIs for the same participants in these two
> SPM versions. Do people know how to solve this inconsistency between SPM5
> and SPM2?