If your thesis tests a model with several latent constructs, such as trust, satisfaction and loyalty measured through survey items, you will almost certainly use structural equation modelling (SEM). The next question is which kind: covariance-based SEM, usually run in SPSS AMOS, or partial least squares SEM, usually run in SmartPLS. This guide explains the difference in plain language and helps you choose and justify the right approach.
The short answer
- Choose CB-SEM (AMOS) when your main aim is to confirm an established theory, your sample is reasonably large, your constructs are reflective and you want to report overall model fit.
- Choose PLS-SEM (SmartPLS) when your aim is prediction or theory development, your model is complex, your sample is smaller, your data are not normally distributed or your model includes formative constructs.
What is the difference?
CB-SEM estimates the model by trying to reproduce the covariance matrix of your observed data. It asks: how well does the whole model fit the data? That is why it produces global fit indices such as CFI and RMSEA.
PLS-SEM works more like a series of regressions. It focuses on maximising the variance explained in the dependent constructs. It asks: how well does the model explain and predict the outcomes? It does not rely on the same global fit logic, which is why fit indices play a much smaller role in PLS-SEM reporting.
Comparison at a glance
| CB-SEM (AMOS) | PLS-SEM (SmartPLS) | |
|---|---|---|
| Main goal | Theory testing and confirmation | Prediction and theory development |
| Sample size | Larger samples, often 200 or more | Works with smaller samples, though power still matters |
| Data distribution | Maximum likelihood assumes multivariate normality | No normality assumption; bootstrapping for significance |
| Construct types | Mainly reflective | Reflective and formative |
| Model complexity | Can struggle with very complex models | Handles many constructs and indicators well |
| Model fit | Central: CFI, TLI, RMSEA, SRMR | Limited role; focus on R², Q² and prediction |
What to report in CB-SEM (AMOS)
A typical CB-SEM analysis has two stages. First, a confirmatory factor analysis (CFA) tests the measurement model. Then the structural model tests your hypotheses.
- Measurement model: standardised factor loadings (ideally 0.70 or higher), composite reliability (0.70 or higher), average variance extracted (AVE of 0.50 or higher) and discriminant validity, for example using the HTMT ratio.
- Model fit: chi-square with degrees of freedom, CFI and TLI (above 0.90 is acceptable, 0.95 or higher is good), RMSEA (0.06 to 0.08 or lower) and SRMR (0.08 or lower), following Hu and Bentler (1999).
- Structural model: standardised path coefficients, p values and the variance explained in each endogenous construct.
What to report in PLS-SEM (SmartPLS)
PLS-SEM reporting usually follows the guidelines of Hair and colleagues and is also done in two stages.
- Reflective measurement model: indicator loadings (0.708 or higher), internal consistency reliability through composite reliability or rho_A (between 0.70 and 0.95), convergent validity through AVE (0.50 or higher) and discriminant validity through HTMT (below 0.85 or 0.90).
- Formative measurement model, if any: collinearity (VIF) and the significance and relevance of indicator weights.
- Structural model: collinearity between predictors (VIF below about 3), path coefficients with significance from bootstrapping (commonly 5,000 subsamples), R², effect sizes (f²) and predictive power using PLSpredict or Q².
How to justify your choice in the methodology chapter
Examiners do not expect one method to be “correct”. They expect you to explain why your choice fits your study. A good justification links the method to your aim, your model and your data, for example:
PLS-SEM was selected because the study aims to predict behavioural intention, the model includes a formative construct and the data deviate from multivariate normality (Hair et al., 2022).
Avoid justifying PLS-SEM only by saying your sample is small. Reviewers increasingly see that as a weak reason on its own.
Common mistakes
- Reporting CB-SEM fit indices as the main evidence in a PLS-SEM study
- Deleting many items to reach thresholds without explaining the effect on content validity
- Skipping discriminant validity or relying only on the Fornell-Larcker criterion
- Forgetting to check common method bias in single-source survey data
- Interpreting non-significant paths as proof that no relationship exists
Choosing and running the right SEM approach can make or break your results chapter. Our data analysis team works daily with SmartPLS 4 and AMOS and can run your analysis, produce APA-formatted tables and explain every result so you can defend it.