What do you think are the most significant differences between PSEMs as introduced, e.g. by  Conrady and Jouffe (2013, see link 1 and also 2) and a corresponding (pure) Bayesian network (representation)? What are the advantages/drawbacks of the one over the other?

Gupta and Kim (2008, see link 3) provide a related discussion in the light of linking SEMs to Bayesian networks. But how does this relate/differ from PSEMs?

I would also be interested in your opinion concerning the PSEM approach (data-driven) vs. the traditional (theory-guided) SEM approach.

Thanks, Mario

http://library.bayesia.com/download/attachments/4882672/PSEM_v17.pdf?version=1&modificationDate=1368141335000&api=v2

http://library.bayesia.com/display/VW/Probabilistic+Structural+Equation+Models+-+Application+to+Human+Resource+Management

http://www.sciencedirect.com/science/article/pii/S0377221707005590