1) The document proposes a penalized likelihood method using a penalty function like SCAD to perform simultaneous variable selection and parameter estimation in structural equation models (SEMs).
2) The method considers a general SEM where latent variables are linearly regressed on themselves with a coefficient matrix, avoiding the need to specify outcome and explanatory latent variables. Selecting nonzero coefficients in the matrix identifies the structure of the latent variable model.
3) Under regularity conditions, the consistency and oracle properties of the proposed penalized maximum likelihood estimators are established. An expectation-conditional maximization algorithm is developed for computation.