Structural Equation Modeling (SEM) is a widely used technique in the behavioral and social sciences to study the relationships between latent constructs (such as depression, personality traits, attitudes, ...). In the world of SEM, of model typically consists of two parts: a measurement part (relating the latent variables to observed indicators), and a structural part (relating the latent variables to each other). In practice, the primary interest of researchers using SEM is in the structural part. However, the traditional way to estimate the (many) parameters in these models is to consider to full model and estimate all parameters simultaneously. This leads to many estimation problems, in particular when some parts of the model are not entirely correct. By contrast, in the structural-after-measurement (SAM) approach, we estimate the parameters in two steps: first the measurement part, and then the structural part. This works equally well as the traditional method if the model is correct, but is more robust if some parts are slightly misspecified. This project will further evaluate and extend the SAM approach in order to make it faster, more scalable, and capable of fitting a wider range of models that are of interest to researchers interested in correct and robust models of human functioning.