Structural equation modeling

Abstract This chapter describes structural equation modeling (SEM), which represents a probabilistic modeling framework for studying causal hypotheses about systems. SEM relies on interconnected series of equations to represent networks as complex hypotheses. As a general modeling methodology, SEM potentially permits any type of functional response (error distributions) and linkage form (linear, non-linear, or non-parametric). The methodology also includes procedures for evaluating proposed models against data, permitting the discovery of unsuspected mechanisms leading to an understanding of how multiple processes collectively control systems. A key element of SEM is the use of graphical models to represent the causal logic implied by the equations. In this treatment the chapter concisely describes SEM fundamentals, incorporating the latest advances in the core methodology. Worked examples are used to illustrate procedures.