Getting Started with Structural Equation Modelling
Learn the fundamentals of SEM and how to build your first model using AMOS and R lavaan packages. We cover path diagrams, model fit indices, and common pitfalls.
What is Structural Equation Modeling?
Structural Equation Modeling (SEM) is a multivariate statistical framework that combines factor analysis and path analysis. It allows researchers to test complex relationships between observed and latent (unobserved) variables simultaneously. Unlike standard regression, SEM can model measurement error explicitly, making it a powerful tool for social sciences, psychology, education, and business research.
Key Components of SEM
Every SEM has two parts. The measurement model defines how latent variables are indicated by observed variables — this is essentially a confirmatory factor analysis. The structural model specifies the directional relationships (paths) between latent variables. Together, they let you test whether your theoretical model fits the data. Common fit indices include CFI (Comparative Fit Index, guideline ≥ 0.95), TLI (Tucker-Lewis Index, guideline ≥ 0.95), RMSEA (Root Mean Square Error of Approximation, guideline ≤ 0.06), and SRMR (Standardized Root Mean Square Residual, guideline ≤ 0.08) (Hu & Bentler, 1999); these are guidelines, not strict cut-offs.
Getting Started with lavaan in R
The lavaan package (short for "latent variable analysis") is the most popular open-source tool for SEM in R. It uses an intuitive syntax: the =~ operator defines latent variables (e.g., "motivation =~ item1 + item2 + item3"), the ~ operator specifies regressions, and ~~ defines covariances. After writing your model syntax, you fit it with the sem() function and inspect results with summary(fit, fit.measures=TRUE, standardized=TRUE). lavaan outputs parameter estimates, standard errors, p-values, and a full set of fit indices.
Using AMOS for SEM
IBM SPSS AMOS provides a graphical interface for building SEM models. You draw path diagrams by placing observed variables (rectangles) and latent variables (ovals) on a canvas, then connect them with single-headed arrows (regression paths) or double-headed arrows (covariances). AMOS automatically estimates parameters and provides fit indices. It is particularly popular among researchers who prefer a visual approach over coding.
Common Mistakes to Avoid
Beginners often make several mistakes: specifying too many paths without theoretical justification, ignoring modification indices blindly, relying on a single fit index, and using small sample sizes. As a rough guide, SEM samples of fewer than 100 are usually considered small and about 200 or more more comfortable, but the sample size you need depends on model complexity and the number of parameters. Always ground your model in theory first, then test it with data — not the other way around.
Recommended Resources
For further learning, consider: Kline (2023) "Principles and Practice of Structural Equation Modeling" for a comprehensive textbook; the official lavaan tutorial at lavaan.ugent.be; and UCLA's SEM seminar materials at stats.oarc.ucla.edu for practical R examples with detailed explanations.
References
Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 6(1), 1–55. https://doi.org/10.1080/10705519909540118