Recursive/Bidirectional Paths in SEM


So I’m a complete newbie when it comes to structural equation modelling/path analysis, but I am trying to learn more about it and use the 1dSEM and 1dSVAR commands. One thing that I’ve come across is that 1dSEM seems to estimate separate path coefficients for bidirectional paths. However, from some admittedly basic reading, it seems like these kind of paths cause the model to be unidentified, and the only bidirectional paths allowed simply represent the correlation between the variables. How does the program handle this? Thank you!



Sri, if I still remember it correctly, bidirectional paths are allowed in 1dSEM, but there is an upper bound of n(n-1)/2 for the number of paths (n is the number of regions).

Thanks Gang! So, I’m under the impression that path coefficients for bidirectional paths can only be estimated in specific instances (e.g. with instrumental variables). Are these automatically taken into account when using the data driven search for the path model?

SEM is generally considered as a model validation approach. I’m not an expert on SEM and don’t have a clear answer about this: it seems a little controversial to use it for model search.