Comparing Bayesian Path Analysis with BSEM models in BLAVAAN

[Gabe Avakian Orona]

Dear All,

Is the blavCompare function appropriate when comparing two models, one of which is a path analysis, the other a structural equation model (both using BSEM function)? 

• For the path analysis: The mediator and outcome are composites (non-latent variables). 
• For the SEM: The same mediator and same outcome are specified as latent factors. 

Essentially, I want to know if I can compare models to see if specifying the constructs as factors is preferred over specifying them as composites. Or will the output metrics always favor the simpler (e.g., path) model?

Thank you very much.

Best,
Gabe

[Mauricio Garnier-Villarreal]

Gabe

No, the blavcompare need to have the same data/outcomes in the model. Once we use composite variables versus factors, you are changing the structure of the data and the comparasiosns are not longer valid

You can look at the model with factors, and evaluate if the data is very well behaved, might be good enough to use composite variables. But this would required very high overall fit indices, and very high R2 for the factor indicators. When you creaste comosite variables you are assuming that the relation between the factor and the items is perfect (R2 = 1)

In general I would recommend to use the model with factors instead of composite scores

take care