A Flexible Methodology for Model Validation in Structural Equation Modeling

Structural Equation Modeling (SEM) is a popular methodology where relations between latent and manifest variables are described by equations simultaneously estimated using methods such as Maximum Likelihood, Least Squares, or Weighted Least Squares. Traditionally, researchers must specify all hypothesized relations a priori, which requires strong theoretical knowledge. This paper presents an exploratory approach to model validation in SEM. It consists in exploring all relevant relations in structural and measurement models guided by the observed data. An application using educational data demonstrates the methodology.