Measurement and structural models built around the research question
Use EFA, CFA, and SEM for the decisions they can support, with specification, identification, estimation, fit, and interpretation treated as one connected workflow.
Different tools, different decisions
EFA explores plausible structure; CFA evaluates a specified measurement model; SEM connects measurement and structural relations. Moving between them should follow the design and evidence, not a convenient fit statistic.
- Define constructs and indicators
- Check identification and estimation
- Separate measurement from structural claims
Fit is evidence, not a target to chase
Residuals, loading patterns, theory, data quality, estimation warnings, and plausible alternatives matter alongside global fit indices. Modification indices are not instructions and can overfit a model to one sample.
- Inspect local and global diagnostics
- Record theoretically justified revisions
- Use sensitivity or validation where feasible
Software follows the model
AMOS, R, and SmartPLS can support different workflows. The choice should reflect estimand, data, estimator, diagnostics, reproducibility, and reporting needs.
- Specification diagram
- Fit and diagnostic summary
- Direct, indirect, and limitation notes
Review my measurement or structural model
Bring the research question, design, or existing output. A first review can identify the next statistical decision and the information needed to scope it.