Let's suppose that I want to regress A over B and B over C using SEM. Is it possible to also add the variable A square to see whether A has a curvilinear effect on B?
Lorenzo, yes, this is possible. You could either use the product-indicator (PI) approach or latent moderated structural equations (LMS). For LMS you would need to use the Mplus program, while the PI approach can be used with all latent variable programs.
In this article you will find the syntax for LISREL (PI) and Mplus: