In SEM, Does it make sense to evaluate direct, indirect, and total effects among latent variables?

Question

I am doing an SEM model with has both latent and measured components. When I read textbooks and online examples, almost all examples that involve direct/indirect/total effects use path models with no latent variables (like the one in the figure below). I am guessing that is because the interpretation of the coefficients becomes less useful when the variables cannot be measured? But does it still make sense for me to estimate indirect effects like ab and total effects like ab+c' and examine whether they are significant, if my IV, DV, and M are all latent variables? Thank you very much.

Answer 1

There is nothing wrong with performing mediation with latent variables. The reason examples use path models with no latent variables is that they help communicate the idea of mediation without introducing the additional complexity latent variables bring. But mediation is a causal concept that describes the relationship among variables. Whether they are latent or not is simply a matter of circumstance, not an intrinsic feature of the variables, and not related to their role in a causal system that involves mediation. See Muthén and Asparouhov (2015) for a clear paper introducing mediation, including its use with latent variables.