Structural Equation Modeling (SEM): A Beginner's Guide
Meet the Expert
Shruti Sharma
Academic Writing Coach & Research Communication Specialist
- Guided 300+ PhD scholars through SEM, CFA, SmartPLS, and AMOS analysis
- Expertise in measurement model assessment, structural model testing, and model fit evaluation
- Specialises in writing advanced quantitative methodology and findings chapters for management theses
Structural Equation Modeling (SEM) is a powerful multivariate technique that simultaneously estimates relationships among multiple latent constructs. It extends regression analysis by incorporating measurement error, testing complex multi-path models, and allowing both direct and indirect (mediated) effects to be tested at once. SEM is the method of choice for testing conceptual frameworks in management, psychology, education, and marketing research.
If your thesis involves a structural model with two or more latent constructs, hypothesised relationships between them, and multi-item scales measured with Likert items — SEM is almost certainly the right analytical technique. This guide walks you through the fundamentals, the key decisions, and how to report SEM results in your thesis.
The Two Components of SEM
Every SEM analysis consists of two sub-models:
- Measurement Model (CFA): Specifies how observed indicator items relate to underlying latent constructs. This is tested first to confirm construct validity before testing structural paths.
- Structural Model (Path Model): Specifies the hypothesised relationships (paths) between latent constructs. This tests your research hypotheses.
CB-SEM vs PLS-SEM — Head to Head
Best for testing established theories with large samples; uses covariance matrix
Best for predictive research, smaller samples, and exploratory frameworks
Requires larger samples; sensitive to non-normality and model misspecification
Works with smaller samples; robust to non-normal data distributions
IBM AMOS is the most popular academic tool for CB-SEM
SmartPLS 4 is the dominant tool for PLS-SEM in management research
SEM Research Process: Step-by-Step
| Step | Activity |
|---|---|
| 1. Model Specification | Define latent constructs, indicators, and hypothesised structural paths based on theory |
| 2. Data Collection | Collect data from an adequate sample (N ≥ 200 for CB-SEM; N ≥ 100 for PLS-SEM) |
| 3. Measurement Model (CFA) | Run CFA to assess reliability (CR ≥ 0.70), convergent validity (AVE ≥ 0.50), and discriminant validity |
| 4. Model Fit Assessment | Evaluate fit indices (CFI, TLI, RMSEA, SRMR, CMIN/DF) |
| 5. Model Modification | If fit is poor, consult modification indices; revise with theoretical justification |
| 6. Structural Model Testing | Test path coefficients, significance levels (t-values or bootstrapping), and R² |
| 7. Hypothesis Testing | Accept or reject each hypothesis based on path significance (p < 0.05) |
| 8. Reporting | Report measurement model results, fit indices, path coefficients, and hypothesis outcomes |
SEM Model Fit Indices — What to Report
| Fit Index | Acceptable Threshold | Preferred Value |
|---|---|---|
| Chi-square (χ²) | p > 0.05 (sensitive to N) | Not used alone; report with CMIN/DF |
| CMIN/DF (χ²/df) | ≤ 5.0 | ≤ 3.0 |
| CFI | ≥ 0.90 | ≥ 0.95 |
| TLI (NNFI) | ≥ 0.90 | ≥ 0.95 |
| RMSEA | ≤ 0.08 | ≤ 0.06 |
| SRMR | ≤ 0.08 | ≤ 0.05 |
Measurement Model: Reliability and Validity Checks
Before testing the structural model, your measurement model must demonstrate:
| Check | Measure | Threshold |
|---|---|---|
| Internal consistency reliability | Cronbach's alpha, Composite Reliability (CR) | ≥ 0.70 |
| Convergent validity | Average Variance Extracted (AVE) | ≥ 0.50 |
| Discriminant validity | HTMT ratio (PLS-SEM); Fornell-Larcker criterion (CB-SEM) | HTMT < 0.85; √AVE > inter-construct correlations |
| Factor loadings | Standardised factor loadings from CFA | ≥ 0.50 (≥ 0.70 preferred) |
Two-Step Approach to SEM (Anderson & Gerbing, 1988)
The widely cited two-step approach recommends: Step 1 — assess and confirm the measurement model (CFA) before testing structural paths; Step 2 — test the structural model (path analysis) once measurement model fit is confirmed. This approach is considered best practice and should be explicitly referenced in your thesis methodology chapter.
Building an SEM model for your thesis and need help with AMOS, SmartPLS, or writing up results? Our SEM specialists at Thesis Ace Writers can guide you from model specification to final write-up.
Mediation and Moderation in SEM
SEM is particularly powerful for testing mediation (indirect effects) and moderation (interaction effects):
- Mediation (indirect effect): Variable M mediates the relationship between X and Y. In SEM, test using bootstrapping (5,000 resamples) to get confidence intervals for indirect effects. If the CI does not include zero, mediation is confirmed.
- Moderation (interaction effect): Variable Z moderates the X → Y relationship. In PLS-SEM, use the product indicator or latent moderation scaling approach in SmartPLS.
Reporting SEM Results in Your Thesis
A complete SEM results section in your thesis should include:
- Table of standardised factor loadings, AVE, CR, and Cronbach's alpha for all constructs
- Discriminant validity matrix (HTMT or Fornell-Larcker)
- Model fit summary table (CFI, TLI, RMSEA, SRMR, CMIN/DF)
- Structural path diagram with path coefficients and significance levels
- Hypothesis testing summary table (path coefficient, t-value, p-value, supported/not supported)
Related Reading from Thesis Ace Writers
Stuck on your SEM model, fit indices, or hypothesis testing write-up? Book a session with Thesis Ace Writers today.
Frequently Asked Questions
Click a question to expand the answer.
Structural Equation Modeling (SEM) is a multivariate statistical technique that simultaneously tests the relationships between multiple independent and dependent variables, including latent (unobserved) constructs. It combines confirmatory factor analysis (CFA) — which tests the measurement model (how observed items relate to latent constructs) — and path analysis (which tests the structural model: the hypothesised relationships between constructs). SEM is widely used in management, psychology, marketing, and social science research to test complex theoretical frameworks.
CB-SEM (Covariance-Based SEM), implemented in AMOS or R (lavaan), is suited for testing and confirming well-established theories with large samples (N ≥ 200) and normally distributed data. PLS-SEM (Partial Least Squares SEM), implemented in SmartPLS, is suited for predictive research, smaller samples, non-normal data, and exploratory theory building. PLS-SEM is increasingly popular in management and information systems research; CB-SEM is preferred in psychology and organisational behaviour.
Model fit in SEM is assessed using multiple indices: CFI (Comparative Fit Index) ≥ 0.90 (≥ 0.95 preferred); TLI (Tucker-Lewis Index) ≥ 0.90; RMSEA ≤ 0.08 (≤ 0.06 preferred); SRMR ≤ 0.08; Chi-square/df ratio (CMIN/DF) ≤ 3.0 (≤ 5.0 acceptable). No single index is definitive — report at least four fit indices and explain the pattern.
For CB-SEM, the minimum recommended sample size is 200, with 10–20 observations per free parameter as a rough rule of thumb. For PLS-SEM, the sample size requirement is lower — a common guideline is 10 times the maximum number of paths pointing to any single construct (the '10× rule'). For most management studies, a sample of 150–250 is adequate for PLS-SEM. Conduct a power analysis (e.g., using G*Power or Monte Carlo simulation) to justify your sample size formally.
Popular SEM software options include: AMOS (IBM) — for CB-SEM; graphical interface, widely used in academic research; SmartPLS — for PLS-SEM; popular in management, IS, and marketing research; R (lavaan, semPLS, plspm packages) — free, flexible, for both CB-SEM and PLS-SEM; Mplus — advanced, handles non-normal data, multilevel SEM, mixture models; LISREL — one of the original SEM programs; EQS — alternative to AMOS. Most Indian and international PhD students in management use AMOS or SmartPLS.