Description
A stage-by-stage checklist for planning, reporting or reviewing a structural equation model so that its specification, fit, modifications and validity can be judged and replicated.
Reviews of published SEMs found recurring faults: a single fit index, no cross-validation, weak theoretical grounding, missing data and no consideration of rival models. The checklist responds by asking specific questions at each stage. Data preparation: population, measurement level, descriptives, multivariate normality, missing data handling (FIML or imputation), outliers, transformations, the correlation, mean and SD table, and the matrix analysed. Specification: rationale, latent variable definitions, theory for measurement and structural parts, alternative models, exploratory or confirmatory status, sample size and power, expected signs and a diagram. Identification: distinct values, free parameters, order and rank conditions. Estimation: estimator choice, convergence, Heywood cases, scaling, software version and start values. Testing: several fit indices, composite reliability, validity, invariance, unstandardised paths with standard errors and standardised values. Modification: one change at a time with justification and equivalent models. Validation: replication, split-sample cross-validation and bootstrap.
Attributes
- Discipline
- Research Design & Rigor
- Loss type
- Siloed
- Stage
- Evaluation
- Data
- Analytical / expert
- Effort
- Hours
Steps
- Data preparation: describe the population, measurement levels, descriptives, normality, amount and handling of missing data, outliers, transformations and the matrix analysed; share data and syntax where possible.
- Specification: state the rationale, latent variable definitions, theory, alternative models, exploratory or confirmatory status, sample size justification, hypotheses with expected signs, and an initial diagram.
- Identification: report the number of distinct values and free parameters and how order and rank conditions are met.
- Estimation: justify the estimator, and report convergence problems, Heywood cases, scaling choices, software version and R-squared values.
- Testing: report at least two fit indices with their criteria, reliability and validity evidence, invariance tests where groups are compared, and unstandardised paths with standard errors or intervals alongside standardised values.
- Modification and validation: document each change one at a time with its justification, consider equivalent models, show initial and final diagrams, and cross-validate on a split or new sample.
Sources
- A Beginner's Guide to Structural Equation Modeling (Schumacker & Lomax, 2004)
- Latent Variable Models (Loehlin & Beaujean, 2017)
- How to Perform and Report an Impactful Analysis Using Partial Least Squares (Benitez et al., 2020)
Also known as
- Structural equation model checklist
- SEM reporting standards
- Thompson's ten commandments of SEM
- Communicating about latent variable models (reporting guidance)
Categories
- Research Operations
- Research Design & Rigor