Summary

Structural Equation Modeling (SEM) is a powerful quantitative tool for theory testing with the ability to generate latent variables that more closely approximate theoretical constructs and parsing out the causal effects (both direct and indirect) versus spurious relationships between them. The roots of SEM date back to the early 20th century, with the developments of Factor Analysis and Path Analysis, but it was not until the second half of the 20th century that CFA developed, and not until the late 20th century that these methods were extended to analyses of both continuous and categorical independent and dependent variables. Data requirements are more demanding for SEM applications relative to single-equation models with observed variables, and users must understand the limitations of both their data and conceptual model (the latter related to model “identification”). SEM applications in criminology have expanded considerably during the 21st century, although its use remains rare compared to other fields. A substantive understanding of the topic is critical for proper model building and model refinement in SEM, and the ability to assess both global fit (for the entire model) and local fit (for specific parameter estimates) is essential for a full evaluation of multi-equation systems. Aside from the more general applications of SEM, it provides one of the most useful methods for studying individual change over time through Latent Growth Curves. SEM is a rigorous method for both theory testing and policy evaluation.