PLS-SEM vs. CB-SEM: When to Use Which SEM Method
PLS-SEM (partial least squares structural equation modeling) is a composite-based SEM method that maximizes the explained variance of endogenous constructs and is best suited for prediction, explanation, formative measurement, and complex models. CB-SEM (covariance-based structural equation modeling) is a common factor-based SEM method that reproduces the empirical covariance matrix and is best suited for theory testing, theory confirmation, and global model fit assessment.
Researchers who estimate structural equation models often need to decide between these two approaches. Both methods are valuable, but they serve different analytical purposes and rely on different statistical estimation principles.
SmartPLS (Ringle et al., 2024) supports researchers in applying both PLS-SEM and CB-SEM. The choice between PLS-SEM and CB-SEM should therefore not be seen as a choice between different software packages. Instead, researchers should select the SEM method that best fits the study's research objective, theoretical model, measurement model specification, data characteristics, prediction goals, and reporting requirements.
PLS-SEM vs. CB-SEM at a Glance
| PLS-SEM | CB-SEM | |
|---|---|---|
| Approach | Composite-based (variance-based) SEM | Common factor-based SEM |
| Statistical objective | Maximize explained variance of endogenous constructs | Reproduce the empirical covariance matrix |
| Primary research goal | Prediction and explanation (causal-predictive) | Theory testing and confirmation |
| Model evaluation focus | Out-of-sample prediction (PLSpredict, CVPAT), R², path coefficients | Global goodness-of-fit (e.g., chi-square, CFI, RMSEA, SRMR) |
| Formative measurement | Directly supported | Possible, but requires additional specification |
| Model complexity | Well suited for complex models with many constructs and relationships | Increasing complexity can create estimation and convergence issues |
| Latent variable scores | Directly available for follow-up analyses (e.g., IPMA) | Not determinate |
| Typical use cases | Driver analyses, success factor research, predictive studies | Confirmatory factor analysis, testing established theories |
| In SmartPLS | PLS-SEM algorithm | CB-SEM module |
In short: Choose PLS-SEM when prediction, explanation, formative constructs, or model complexity are central. Choose CB-SEM when theory confirmation and global model fit drive the study. Both methods are available in SmartPLS and can be combined in a multimethod SEM strategy.
PLS-SEM vs. CB-SEM: The Main Difference
The main difference between PLS-SEM and CB-SEM lies in their estimation logic: PLS-SEM estimates constructs as composites and maximizes explained variance, whereas CB-SEM estimates common factor models and minimizes the difference between the observed and the model-implied covariance matrix.
PLS-SEM estimates construct proxies as composites, usually as weighted linear combinations of indicators. Its statistical objective is to reduce unexplained variance in the model's endogenous constructs and indicators. This makes PLS-SEM particularly suitable for causal-predictive research, prediction-oriented model assessment, and explaining key target constructs. Because of this estimation logic, PLS-SEM is also referred to as variance-based SEM.
CB-SEM estimates common factor models and focuses on reproducing the empirical covariance matrix. Its statistical objective is to minimize the difference between the observed covariance matrix and the model-implied covariance matrix. This makes CB-SEM especially useful when the primary objective is theory testing, theory confirmation, confirmatory factor analysis, and global model fit assessment.
The decision should not be framed as "PLS-SEM is better" or "CB-SEM is better." Instead, the guiding question should be:
Which SEM method best supports the study's research objective, theoretical assumptions, measurement model, and intended conclusions?
When to Use PLS-SEM
Use PLS-SEM when your primary goal is prediction, explanation of key target constructs, or when your model includes formative constructs or high complexity. More specifically, PLS-SEM is appropriate when one or more of the following conditions apply:
- The main objective is prediction, explanation, or identifying key driver constructs.
- The research focuses on explaining variance in important target constructs.
- The model includes formatively measured constructs or composite-based measurement.
- The model is complex, with many constructs, indicators, paths, mediating effects, moderating effects, or higher-order constructs.
- The researcher wants to assess the model's out-of-sample predictive power.
- The analysis requires latent variable scores for subsequent analyses, such as importance-performance map analysis, segmentation, or predictive assessment.
- The available sample size is limited because the population itself is small or difficult to access.
- The data are nonnormally distributed.
- The study uses secondary data, archival data, or large-scale data sources.
- The research is theory-development oriented or extends established theories in complex ways.
According to Hair et al. (2027), PLS-SEM is particularly appropriate when researchers focus on prediction, complex models, formative measurement, and the explanation of key target constructs.
Recent methodological guidance further clarifies that traditional arguments such as small sample size, nonnormal data, or exploratory research should not be used as the only justification for PLS-SEM. These reasons can support the decision, but the primary justification should be linked to the research objective, the conceptual nature of the constructs, and the suitability of composite-based estimation for the model.
When to Use CB-SEM
Use CB-SEM when your primary goal is to test or confirm an established theoretical model and when global model fit is central to the study. More specifically, CB-SEM is appropriate when one or more of the following conditions apply:
- The main objective is theory testing or theory confirmation.
- The study compares competing theoretical models.
- The model is strongly grounded in established theory.
- The research requires global goodness-of-fit assessment.
- The researcher wants to evaluate how well the theoretical model reproduces the observed covariance structure.
- The analysis requires confirmatory factor analysis (CFA) as part of the measurement model assessment.
- The model requires detailed specification of error term covariances.
- The model includes circular or nonrecursive relationships.
- The research design relies on common factor measurement assumptions.
CB-SEM is often preferred when researchers want to confirm a theoretically derived model and when model fit is central to the study's contribution. In SmartPLS, CB-SEM is available as a dedicated method, so researchers can estimate and evaluate CB-SEM models in the same software environment used for PLS-SEM.
SmartPLS Supports Both PLS-SEM and CB-SEM
SmartPLS is not limited to PLS-SEM. Researchers can use SmartPLS for CB-SEM and CFA, including model setup, estimation, and results evaluation. This is important for users who want to compare PLS-SEM and CB-SEM results, conduct robustness checks, or follow a multimethod SEM strategy.
The SmartPLS CB-SEM module is particularly useful when researchers want to:
- Estimate common factor-based SEM models.
- Conduct CFA before evaluating structural relationships.
- Assess global model fit.
- Compare CB-SEM results with PLS-SEM results.
- Use a common software environment for alternative SEM methods.
For practical guidance, see the SmartPLS documentation on CB-SEM and the SmartPLS 4 CB-SEM software tutorial by Hair et al. (2025). Bido and Souza (2026) also illustrate the continued relevance of learning structural equation modeling and demonstrate the role of CB-SEM in model re-estimation and critical model evaluation.
Reflective Measurement Does Not Automatically Mean CB-SEM
A common misunderstanding is that reflectively measured constructs must always be estimated with CB-SEM. Recent methodological discussions clarify that reflective measurement is a conceptual specification, whereas common factor modeling and composite modeling are statistical estimation approaches.
In other words, the decision to specify a construct reflectively or formatively is made at the theoretical and measurement level. The decision to estimate the model with CB-SEM, PLS-SEM, consistent PLS-SEM (PLSc-SEM), GSCA, or another SEM method is made at the statistical estimation level.
This distinction matters because empirical research never observes theoretical constructs directly. Instead, every SEM method creates statistical proxies for conceptual variables. These proxies approximate the theoretical constructs, but they are not identical to them. Therefore, researchers should avoid equating reflective measurement automatically with common factor estimation.
PLS-SEM can be used with reflectively measured constructs when the research objective and model context support composite-based estimation. Researchers should, however, carefully assess reliability, convergent validity, discriminant validity, and the robustness of their findings. When researchers want composite-based estimation that mimics common factor model results for reflectively specified constructs, consistent PLS-SEM (PLSc-SEM) offers a bridge between both worlds.
Conceptual and Empirical Reasons for Choosing PLS-SEM
Recent guidance suggests that researchers should ask two important questions when considering PLS-SEM:
First, could the indicator residual variances have meaning for the construct or for other constructs in the model?
If the residual variance of indicators may contain meaningful information rather than only measurement error, a composite-based approach such as PLS-SEM can be appropriate.
Second, is the measurement error unlikely to be large?
If the study uses established and refined measures, and if measurement quality is high, PLS-SEM can provide useful construct proxies and valid results, especially in prediction-oriented research.
These questions offer a stronger justification than simply stating that PLS-SEM was used because the sample size was small or because the data were nonnormal.
Practical Decision Rules
| Research situation | Recommended approach |
|---|---|
| Prediction is the main goal | PLS-SEM |
| Explanation of key target constructs is central | PLS-SEM |
| Theory confirmation is the main goal | CB-SEM |
| Global model fit is central to the study | CB-SEM |
| CFA is required for measurement model assessment | CB-SEM |
| The model includes formative constructs | PLS-SEM |
| The model is complex | PLS-SEM |
| The sample size is limited because the population is small | PLS-SEM, with proper power analysis |
| The data are nonnormal | PLS-SEM, with nonnormality and outlier assessment |
| Competing theories are compared | CB-SEM or multimethod SEM |
| Latent variable scores are needed | PLS-SEM |
| Out-of-sample prediction is important | PLS-SEM |
| Robustness across SEM methods is required | Multimethod SEM in SmartPLS |
These rules are practical guidelines, not mechanical rules. Researchers should always justify the choice of method in relation to their research question, theory, model, data, and intended conclusions.
PLS-SEM for Prediction and Explanation
PLS-SEM is frequently selected when the objective is to predict key constructs and explain the relationships that drive them. For example, researchers may want to understand which factors influence customer satisfaction, brand loyalty, technology adoption, employee engagement, innovation success, or organizational performance.
In these situations, PLS-SEM helps identify important driver constructs and assess their relevance for target outcomes. This makes PLS-SEM especially valuable when theoretical knowledge is developing or when managerial implications depend on prediction and explanation. The latent variable scores that PLS-SEM provides can be used directly in follow-up analyses such as the importance-performance map analysis (IPMA).
Because PLS-SEM has a causal-predictive orientation, researchers should not limit their assessment to path coefficients and R-squared values. They should also evaluate the model's out-of-sample predictive power, for example by using PLSpredict or the cross-validated predictive ability test (CVPAT).
CB-SEM for Theory Testing and Model Fit
CB-SEM is frequently selected when the objective is to test a theoretically established model. For example, a researcher may want to confirm whether a measurement model and structural model are consistent with an established theory.
CB-SEM is particularly suitable when global model fit is central to the research design. It allows researchers to assess whether the proposed theoretical model adequately reproduces the observed covariance structure.
However, good model fit does not automatically imply strong predictive power. A model can fit the data well but still perform poorly when predicting new observations. This is why prediction-oriented studies should consider PLS-SEM or a multimethod SEM approach.
Formative Measurement and Model Complexity
A key reason for using PLS-SEM is the inclusion of formatively measured constructs. Formative measurement is common when indicators jointly form or define a construct. Examples include constructs such as service quality dimensions, marketing capability indexes, socioeconomic status, digital transformation readiness, or innovation capability.
While formative measurement can be modeled in CB-SEM, it typically requires additional specification and identification considerations. PLS-SEM offers a more direct and flexible way to estimate models that include formative constructs.
PLS-SEM is also well suited for complex models with many constructs, indicators, structural relationships, mediation effects, moderation effects, higher-order constructs, or heterogeneous data structures. SmartPLS provides additional methods for advanced PLS-SEM analyses, including higher-order models, mediation, moderation, multigroup analysis, and finite mixture partial least squares (FIMIX-PLS).
Sample Size and Data Characteristics
PLS-SEM is often useful when sample sizes are limited or when data do not meet strict distributional assumptions. This is particularly relevant in business-to-business, management, marketing, and organizational research, where large samples are often difficult to obtain.
However, researchers should not choose PLS-SEM only because of small sample size. A small sample does not automatically justify the method. Researchers should assess whether the sample size is sufficient for the model and research objective, for example by using power analysis, the inverse square root method, the gamma-exponential method, or Monte Carlo-based power analysis.
Similarly, nonnormal data can support the use of PLS-SEM, but researchers should still assess skewness, kurtosis, outliers, and the potential influence of data problems on the results. When samples need to represent a population more closely, researchers may also consider the weighted PLS algorithm if appropriate sampling weights are available.
Measurement Model Assessment in PLS-SEM
When using PLS-SEM, researchers should carefully assess the measurement models before interpreting the structural model.
For reflectively measured constructs, researchers should assess:
- Indicator loadings.
- Internal consistency reliability, preferably including reliability coefficient rho_A.
- Composite reliability.
- Average variance extracted (AVE).
- Discriminant validity, preferably using the heterotrait-monotrait ratio of correlations (HTMT).
For formatively measured constructs, researchers should assess:
- Convergent validity using redundancy analysis.
- Indicator collinearity using variance inflation factors (VIF).
- Indicator weights.
- Indicator relevance and significance.
Researchers should avoid relying only on older criteria such as the Fornell-Larcker criterion when assessing discriminant validity. HTMT and bootstrap confidence intervals provide stronger evidence. In SmartPLS, statistical inference for many results can be assessed using bootstrapping.
Structural Model Assessment in PLS-SEM
After establishing measurement model quality, researchers should assess the structural model. Important criteria include:
- Path coefficients and their significance.
- Confidence intervals based on bootstrapping.
- Coefficient of determination R-squared.
- Effect size f-squared.
- Predictive relevance and predictive performance.
- Out-of-sample prediction using PLSpredict.
- Predictive model comparison using CVPAT.
- Model fit criteria such as SRMR, where appropriate.
Researchers should avoid using the old PLS goodness-of-fit index as a general model fit criterion, because it does not reliably distinguish valid from invalid models.
A Multimethod Perspective: PLS-SEM and CB-SEM as Complementary Tools
Modern SEM guidance increasingly recommends a multimethod perspective. Instead of treating PLS-SEM and CB-SEM as mutually exclusive alternatives, researchers can use both methods to assess whether their conclusions are robust.
For example, a researcher may use CB-SEM to assess model fit and PLS-SEM to assess prediction. If both methods lead to similar substantive conclusions, the findings are more robust. If the results differ strongly, researchers should examine whether the differences are caused by theoretical model problems, data issues, measurement problems, or method-specific assumptions.
When PLS-SEM and CB-SEM Results Diverge: What to Check
If PLS-SEM and CB-SEM lead to different substantive conclusions, researchers should systematically examine possible causes before interpreting either result:
- Measurement model quality: Weak indicators, low reliability, or misspecified reflective/formative measurement can affect the methods differently.
- Model specification: Omitted paths, error covariances (relevant in CB-SEM), or higher-order structures may drive differences.
- Data issues: Outliers, nonnormality, heterogeneity, or collinearity can influence the estimation approaches to a different degree.
- Method-specific assumptions: Differences between composite-based and common factor-based estimation can produce systematically different parameter estimates (Sarstedt et al., 2016).
This multimethod perspective is especially useful when researchers want to reduce methodological uncertainty and provide stronger support for their conclusions. Since SmartPLS supports both PLS-SEM and CB-SEM, researchers can implement this multimethod strategy within one software environment.
How to Justify the Choice of PLS-SEM or CB-SEM
A strong method justification should explain why the chosen SEM approach fits the study. Researchers should address the following points:
- What is the primary research objective: prediction, explanation, theory testing, or theory confirmation?
- Are the constructs best represented through composite-based or common factor-based estimation?
- Are there formatively measured constructs?
- Is the model complex?
- Is out-of-sample prediction important?
- Is global model fit central to the study?
- Is CFA required?
- Are latent variable scores needed for subsequent analyses?
- Are the sample size and data characteristics suitable for the selected method?
- Would a multimethod SEM approach strengthen the robustness of the conclusions?
Weak justifications should be avoided. For example, "PLS-SEM was used because the sample size was small" is not sufficient by itself. A stronger justification links PLS-SEM to prediction, explanation, composite-based estimation, formative measurement, model complexity, latent variable scores, and predictive assessment.
Example Justification for PLS-SEM
Researchers can adapt the following wording for their method section:
"We used PLS-SEM because our study follows a causal-predictive research objective: we aim to explain and predict target construct and to identify its key driver constructs (Hair et al., 2027). Our model includes formatively specified constructs and a complex structure with mediating and moderating effects, for which composite-based estimation is well suited. In line with recent guidance (Guenther et al., 2023), we assessed the measurement models, the structural model, and the model's out-of-sample predictive power using PLSpredict."
Example Justification for CB-SEM
"We used CB-SEM because our study tests an established theoretical model and global model fit is central to our contribution. CB-SEM allows us to assess how well the hypothesized common factor model reproduces the empirical covariance structure and to conduct a confirmatory factor analysis of the reflectively specified constructs (Hair et al., 2025)."
Frequently Asked Questions
Is PLS-SEM better than CB-SEM?
Neither method is universally better. PLS-SEM and CB-SEM pursue different statistical objectives: PLS-SEM maximizes explained variance and supports causal-predictive research, while CB-SEM reproduces the covariance matrix and supports theory confirmation. The right choice depends on the research objective, the measurement models, and the intended conclusions — not on general superiority claims.
Can I use PLS-SEM with reflectively measured constructs?
Yes. Reflective measurement is a conceptual specification, whereas composite and common factor modeling are statistical estimation approaches. PLS-SEM can estimate models with reflectively specified constructs when the research objective supports composite-based estimation (Guenther et al., 2025). Researchers should then carefully assess reliability and validity.
Does SmartPLS support CB-SEM?
Yes. SmartPLS includes a dedicated CB-SEM module for estimating common factor-based models, running confirmatory factor analysis (CFA), and assessing global model fit — in the same environment used for PLS-SEM (Hair et al., 2025).
Is a small sample size a valid reason to use PLS-SEM?
Not on its own. A limited sample can support the decision when the population is small or hard to access, but the primary justification should link PLS-SEM to the research objective, composite-based estimation, formative measurement, or model complexity. Researchers should verify sample size adequacy, for example with power analysis or the inverse square root method.
Can I use both PLS-SEM and CB-SEM in one study?
Yes. A multimethod SEM strategy uses CB-SEM for model fit assessment and PLS-SEM for predictive assessment. Consistent conclusions across methods strengthen the robustness of the findings. SmartPLS supports both methods in one software environment.
Recommended Literature
A useful starting point for understanding the choice between PLS-SEM and CB-SEM is the article by Rigdon, Sarstedt, and Ringle (2017), which discusses five perspectives and recommendations for comparing results from CB-SEM and PLS-SEM.
Sarstedt et al. (2016) explain where estimation bias can occur when PLS-SEM and CB-SEM are applied under different model assumptions. Hair et al. (2027) provide practical rules of thumb for using PLS-SEM. Hair et al. (2019) offer guidance on when to use PLS-SEM and how to report the results.
Guenther et al. (2023) provide current best-practice guidance on improving PLS-SEM use in business marketing research, including justifications for using PLS-SEM, common assessment routines, predictive assessment, and advanced analysis techniques.
Guenther et al. (2025) further clarify that reflective measurement does not automatically equal common factor modeling and recommend viewing SEM methods as complementary tools in a multimethod SEM framework.
Hair et al. (2025) provide a SmartPLS 4 software tutorial for CB-SEM, showing how researchers can set up, estimate, and evaluate covariance-based structural equation models in SmartPLS. Bido and Souza (2026) reinforce the continued relevance of SEM and illustrate how CB-SEM can support critical model reading, re-estimation, and alternative model evaluation.
Related SmartPLS Methods and Algorithms
Estimation methods
Measurement model assessment
Prediction and model evaluation
- PLSpredict
- Cross-validated predictive ability test (CVPAT)
- Prediction-oriented model selection
- Model fit
Advanced analyses
- Mediation
- Moderation
- Multigroup analysis
- Endogeneity and Gaussian copulas
- Necessary condition analysis (NCA)
References
Bido, D. d. S., & Souza, C. A. (2026). Structural equation modeling: Is it still worth learning? Brazilian Administration Review, 23(3), e260020.
Fornell, C. G., & Bookstein, F. L. (1982). Two structural equation models: LISREL and PLS applied to consumer exit-voice theory. Journal of Marketing Research, 19(4), 440-452.
Guenther, P., Guenther, M., Ringle, C. M., Zaefarian, G., & Cartwright, S. (2023). Improving PLS-SEM use for business marketing research. Industrial Marketing Management, 111, 127-142.
Guenther, P., Guenther, M., Ringle, C. M., Zaefarian, G., & Cartwright, S. (2025). PLS-SEM and reflective constructs: A response to recent criticism and a constructive path forward. Industrial Marketing Management, 128, 1-9.
Hair, J. F., Babin, B. J., Ringle, C. M., Sarstedt, M., & Becker, J.-M. (2025). Covariance-based structural equation modeling - CB-SEM: A SmartPLS 4 software tutorial. Journal of Marketing Analytics, 13, 709-724.
Hair, J. F., Hult, G. T. M., Ringle, C. M., & Sarstedt, M. (2027). A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM) (4th ed.). Sage.
Hair, J. F., Ringle, C. M., & Sarstedt, M. (2011). PLS-SEM: Indeed a silver bullet. Journal of Marketing Theory and Practice, 19(2), 139-151.
Hair, J. F., Ringle, C. M., & Sarstedt, M. (2012). Partial least squares: The better approach to structural equation modeling? Long Range Planning, 45(5-6), 312-319.
Hair, J. F., Risher, J. J., Sarstedt, M., & Ringle, C. M. (2019). When to use and how to report the results of PLS-SEM. European Business Review, 31(1), 2-24.
Hair, J. F., Sarstedt, M., Ringle, C. M., & Thiele, K. O. (2017). Mirror, mirror on the wall: A comparative evaluation of composite-based structural equation modeling methods. Journal of the Academy of Marketing Science, 45(5), 616-632.
Jöreskog, K. G., & Wold, H. O. A. (1982). The ML and PLS techniques for modeling with latent variables: Historical and comparative aspects. In H. O. A. Wold & K. G. Jöreskog (Eds.), Systems under indirect observation: Part I (pp. 263-270). North-Holland.
Lohmöller, J.-B. (1989). Latent variable path modeling with partial least squares. Physica.
Reinartz, W. J., Haenlein, M., & Henseler, J. (2009). An empirical comparison of the efficacy of covariance-based and variance-based SEM. International Journal of Research in Marketing, 26(4), 332-344.
Richter, N. F., Cepeda Carrion, G., Roldan, J. L., & Ringle, C. M. (2016). European management research using partial least squares structural equation modeling (PLS-SEM): Editorial. European Management Journal, 34(6), 589-597.
Rigdon, E. E. (2016). Choosing PLS path modeling as analytical method in European management research: A realist perspective. European Management Journal, 34(6), 598-605.
Rigdon, E. E., Sarstedt, M., & Ringle, C. M. (2017). On comparing results from CB-SEM and PLS-SEM: Five perspectives and five recommendations. Marketing ZFP, 39(3), 4-16.
Ringle, C. M., Wende, S., & Becker, J.-M. (2024). SmartPLS 4. SmartPLS. https://www.smartpls.com/
Sarstedt, M., Hair, J. F., Ringle, C. M., Thiele, K. O., & Gudergan, S. P. (2016). Estimation issues with PLS and CBSEM: Where the bias lies! Journal of Business Research, 69(10), 3998-4010.
