When to Use PLS-SEM (and When Not)

Researchers often need to justify why they use partial least squares structural equation modeling (PLS-SEM) instead of another structural equation modeling (SEM) approach, such as covariance-based structural equation modeling (CB-SEM). A strong justification does not claim that one SEM method is always superior. It explains why the chosen method fits the study's research objective, theoretical model, measurement model, data situation, prediction goals, and reporting requirements.

SmartPLS (Ringle et al., 2024) supports both PLS-SEM and CB-SEM. This means researchers can estimate prediction-oriented composite-based SEM models and covariance-based common factor models in one software environment. SmartPLS also supports related methods and algorithms, including bootstrapping, model fit assessment, HTMT, PLSpredict, CVPAT, confirmatory factor analysis (CFA), mediation, moderation, multigroup analysis, higher-order models, FIMIX-PLS, PLS-POS, IPMA, NCA, nonlinear relationships, and Gaussian copula-based endogeneity assessment.

Quick Answer

Use PLS-SEM when the study focuses on prediction, explanation, or identifying key driver constructs in a theoretically developed model. PLS-SEM is also useful when the model is complex, includes formative measurement, requires construct scores, uses secondary or archival data, or needs out-of-sample predictive assessment.

Do not use PLS-SEM only because the sample is small, the data are nonnormal, or the software is convenient. These can support the method choice, but they are not sufficient as stand-alone justifications.

Use CB-SEM when the main goal is theory confirmation, CFA-based measurement model assessment, covariance structure assessment, or global model fit under common factor assumptions.

Use a multimethod SEM perspective when the study benefits from both methods: CB-SEM for model fit and common factor-based assessment, and PLS-SEM for driver analysis, construct scores, and prediction-oriented evaluation.

Key Takeaways

PLS-SEM is appropriate for prediction-oriented and explanation-oriented research, especially when researchers want to identify important drivers of target constructs.
CB-SEM is appropriate for theory confirmation and model fit assessment, especially when the model is conceptualized under common factor assumptions.
Small sample size and nonnormal data are supporting reasons, not sufficient primary reasons for choosing PLS-SEM.
Reflective measurement does not automatically require CB-SEM. Reflective measurement is a conceptual decision; common factor and composite modeling are statistical estimation choices.
PLS-SEM and CB-SEM are complementary. A multimethod SEM workflow can increase confidence when key findings converge across methods and improve transparency when they diverge.
Prediction must be assessed directly. Do not infer predictive power only from R², path significance, or model fit. Use out-of-sample methods such as PLSpredict and CVPAT.

PLS-SEM in One Sentence

PLS-SEM is a composite-based SEM method that estimates construct scores and structural relationships with a causal-predictive orientation, making it especially useful for explaining and predicting key target constructs in complex models.

PLS-SEM vs. CB-SEM in One Sentence

PLS-SEM is typically used when prediction, explanation, composite-based modeling, formative measurement, or construct scores are central; CB-SEM is typically used when theory confirmation, common factor modeling, CFA, and global model fit are central.

When Is PLS-SEM Appropriate?

PLS-SEM is often appropriate when:

The analysis tests a theoretical framework from a prediction perspective.
The goal is to explain important target constructs and identify their key antecedents.
The model is complex and includes many constructs, indicators, relationships, mediation effects, moderation effects, nonlinear effects, or higher-order constructs.
The research extends established theory in new or complex ways.
The model includes one or more formatively measured constructs.
The research uses financial ratios, index values, secondary data, archival data, or other data artifacts.
A small population restricts the achievable sample size, for example in business-to-business research.
Distributional issues are a concern, such as nonnormal data.
The research requires construct scores for follow-up analyses, such as segmentation, IPMA, NCA, prediction-oriented analysis, or further statistical modeling.
The researcher wants to evaluate the model's out-of-sample predictive performance using PLSpredict or CVPAT.

These arguments build on the widely cited guidance by Hair, Risher, Sarstedt, and Ringle (2019) and the PLS-SEM primer by Hair, Hult, Ringle, and Sarstedt (2027). Recent research emphasizes that some reasons should be used carefully. Small sample size or nonnormal data, for example, may support the use of PLS-SEM, but they should not be the only justification.

Strong Reasons to Use PLS-SEM

Prediction and Driver Identification Are Central

PLS-SEM is well suited for studies that predict and explain key target constructs. Typical examples include customer satisfaction, loyalty, technology adoption, brand performance, innovation success, employee engagement, and organizational performance.

When prediction is central, researchers should not only report path coefficients and R² values. They should also assess predictive performance with PLSpredict or CVPAT. This is important because managerial recommendations are often predictive: they imply that changing one driver should improve a future outcome.

The Model Is Complex

PLS-SEM can estimate complex models with many constructs, indicators, structural paths, mediating relationships, moderating effects, nonlinear effects, and higher-order constructs. This makes the method attractive in management, marketing, information systems, human resource management, strategy, tourism, operations, and other applied research fields.

SmartPLS offers dedicated functionality for advanced model setups, including higher-order models, mediation, moderation, nonlinear relationships, and endogeneity assessment using Gaussian copulas.

The Model Includes Formative Measurement

PLS-SEM is frequently used when the model includes formatively measured constructs. Formative measurement is appropriate when indicators jointly form or define a construct. Examples include capability indexes, socioeconomic status, service quality dimensions, digital readiness, innovation capability, and composite performance indicators.

When using formative measurement, researchers should assess convergent validity, indicator collinearity, indicator weights, and the relevance and significance of indicators. SmartPLS supports the required assessment steps, including bootstrapping and variance inflation factor (VIF) diagnostics.

Construct Scores Are Needed

PLS-SEM provides construct scores that can be used in follow-up analyses. These scores are useful for segmentation, prediction-oriented analyses, importance-performance map analysis (IPMA), necessary condition analysis (NCA), multigroup comparisons, and other post-estimation procedures.

Secondary or Archival Data Are Used

PLS-SEM can be useful when studies rely on secondary data, archival data, financial ratios, index values, or other data artifacts. In such situations, measurement theory may be less fully developed than in survey-based scale measurement, and composite-based modeling can be appropriate.

Reasons That Support PLS-SEM but Are Not Enough on Their Own

Small Sample Size

PLS-SEM can be helpful when the sample size is restricted by the population size, for example in business-to-business research or studies of senior managers, specialized firms, or niche populations. However, a small sample size does not automatically justify PLS-SEM.

Researchers should determine whether the available sample size is sufficient for the model and research objective. Appropriate approaches include power analysis, the inverse square root method, the gamma-exponential method, or Monte Carlo-based power analysis.

Nonnormal Data

PLS-SEM is often described as robust to nonnormal data. This can support the method choice, but researchers should still assess skewness, kurtosis, outliers, and the potential influence of extreme observations.

Exploratory Research

PLS-SEM is often used in exploratory or theory-development research. This is acceptable when the study clearly labels exploratory elements and avoids presenting data-driven model changes as if they were fully confirmatory. A better justification is that the study explores theoretically meaningful extensions of established theory, especially in complex or prediction-oriented settings.

When Not to Use PLS-SEM

PLS-SEM is not always the best choice. Researchers should consider another SEM approach, especially CB-SEM, when:

The primary goal is strict theory confirmation under common factor assumptions.
Global model fit is central to the study's contribution.
The research requires a detailed CFA-based measurement model evaluation.
The study compares alternative covariance-based theoretical models.
The model includes circular or nonrecursive relationships that require specific covariance-based modeling.
The researcher does not need construct scores, prediction-oriented assessment, or composite-based estimation.
PLS-SEM is selected only because it is easy to use or because another method produced less favorable results.

In these situations, CB-SEM may be more appropriate. SmartPLS supports CB-SEM and CFA, making it possible to evaluate covariance-based models in the same software environment.

PLS-SEM or CB-SEM? Practical Decision Table

Research situation	Recommended SEM approach
Prediction is the main goal	PLS-SEM
Driver identification is central	PLS-SEM
Construct scores are needed	PLS-SEM
The model includes formative constructs	PLS-SEM
The model is complex	PLS-SEM
The sample size is restricted by a small population	PLS-SEM, with sample size justification
The data are nonnormal	PLS-SEM, with data diagnostics
Theory confirmation is the main goal	CB-SEM
Global model fit is central	CB-SEM
CFA is the main task	CB-SEM / CFA
Competing covariance-based theories are compared	CB-SEM
Robustness across SEM approaches is desired	Multimethod SEM
The study needs both explanation and prediction	Multimethod SEM or PLS-SEM with predictive assessment

These guidelines should not be applied mechanically. In many research situations, both PLS-SEM and CB-SEM can provide useful insights, but they answer different methodological questions.

Explanation, Prediction, and Replicability

Many empirical studies want to explain why relationships occur and predict relevant outcomes. These are related but not identical goals.

Explanation focuses on whether the model is theoretically meaningful and whether relationships are supported in the observed sample. Typical criteria include path coefficients, confidence intervals, effect sizes, R², and model fit.

Prediction focuses on whether the model can generalize to new or unseen observations. Typical criteria include out-of-sample prediction, cross-validation, PLSpredict, and CVPAT.

A model can explain the current sample well but perform poorly when predicting new observations. Likewise, a construct can improve prediction even when its theoretical role needs further development. For this reason, researchers should avoid making strong predictive claims based only on in-sample evidence such as R² or significant path coefficients.

The EP-mixed perspective distinguishes exploratory, confirmatory, and mixed research modes and combines them with explanatory and predictive assessments. This helps researchers develop more transparent, replicable, and practically relevant SEM studies.

SmartPLS Also Supports CB-SEM

SmartPLS is not limited to PLS-SEM. It also supports covariance-based structural equation modeling (CB-SEM), including model setup, estimation, and results evaluation. The SmartPLS CB-SEM documentation explains CB-SEM in SmartPLS as a covariance-based approach in which constructs are treated as common factors.

The CB-SEM tutorial by Hair, Babin, Ringle, Sarstedt, and Becker (2025) demonstrates how to conduct a CB-SEM analysis in SmartPLS 4, including model setup, measurement model assessment with CFA, and structural model assessment based on CB-SEM results. Bido and Souza (2026) further illustrate why SEM remains worth learning and show the value of CB-SEM for critical reading, re-estimation, and alternative model assessment.

This means researchers can use SmartPLS for both major SEM traditions:

PLS-SEM for prediction-oriented, composite-based modeling.
CB-SEM for covariance-based theory testing, CFA, model fit, and common factor models.

This dual capability is important because modern SEM research increasingly recommends a multimethod SEM perspective.

A Multimethod SEM Perspective

A multimethod SEM perspective treats PLS-SEM and CB-SEM as complementary tools rather than competing camps. The goal is not to defend one method against the other, but to evaluate whether a model's conclusions are robust under different but theoretically meaningful estimation assumptions.

In this perspective, the model is the central product of the research. The question is not only Which method should I use? but also Do the key theoretical relationships remain meaningful when assessed with different SEM approaches?

A multimethod approach is particularly useful when:

The study has both explanatory and predictive goals.
The theoretical model is important for managerial recommendations.
The researcher wants to reduce methodological uncertainty.
There is uncertainty about whether constructs are better approximated as common factors or composites.
A reviewer, editor, or supervisor asks for robustness checks across SEM methods.
The study wants to show that key conclusions are not artifacts of one estimation method.

Using both CB-SEM and PLS-SEM can strengthen a study in two ways. First, CB-SEM can assess how well the theoretical model fits the observed covariance structure. Second, PLS-SEM can assess driver relationships, construct scores, and out-of-sample predictive performance. When both approaches lead to similar substantive conclusions, confidence in the model increases. When the conclusions differ, the difference becomes informative and should be examined rather than ignored.

A Practical Multimethod SEM Workflow in SmartPLS

A practical multimethod SEM workflow can follow these steps:

Specify the model based on theory. Define the constructs, indicators, and structural paths before estimating the model.
Check the data. Examine missing values, distributional characteristics, outliers, and the sample size requirements for the intended analyses.
Assess measurement model quality. Use criteria appropriate for reflective and formative measurement. For CB-SEM, use CFA-based diagnostics where relevant. For PLS-SEM, assess reliability, convergent validity, discriminant validity, and formative measurement quality.
Estimate the model with CB-SEM. Use CB-SEM to evaluate global model fit, CFA results, and path coefficients under common factor assumptions.
Estimate the model with PLS-SEM. Use PLS-SEM to evaluate path coefficients, R², f², construct scores, and prediction-oriented results under composite-based assumptions.
Compare the structural paths. Identify paths that are supported in both methods, supported only in one method, or not supported in either method.
Assess out-of-sample prediction. Use PLSpredict or CVPAT to evaluate predictive performance.
Interpret convergence and divergence. Convergent results strengthen confidence. Divergent results may point to measurement issues, construct specification problems, data characteristics, or method-specific assumptions.
Report the workflow transparently. Document software version, algorithm settings, bootstrapping settings, estimator choice, model modifications, and all robustness checks.

Interpreting Multimethod Results

Result pattern	Interpretation	Reporting recommendation
CB-SEM and PLS-SEM support the path, and prediction is supported	Strong theoretical and practical relevance	Highlight as a robust key relationship
CB-SEM and PLS-SEM support the path, but prediction is not supported	The path may be theoretically relevant but has limited predictive value	Avoid strong predictive claims
Only CB-SEM supports the path	Support depends on common factor-based estimation	Report as method-sensitive and discuss why
Only PLS-SEM supports the path	Support depends on composite-based estimation	Report as method-sensitive and discuss why
Prediction is supported but in-sample evidence is weak	The path or construct may have practical value, but the theoretical mechanism may need refinement	Treat as theory-development evidence
Neither method supports the path and prediction is not supported	The path is not supported in this study	Consider removing, revising, or testing in another context

The strongest results are those that are theoretically meaningful, empirically supported, predictively useful, and stable across plausible estimation methods. Method-sensitive findings are not necessarily wrong, but they require careful interpretation and transparent reporting.

Reflective Measurement Does Not Automatically Require CB-SEM

A common misunderstanding is that reflectively measured constructs automatically require CB-SEM. Recent methodological discussions clarify that reflective measurement is a conceptual specification, whereas common factor modeling and composite modeling are statistical estimation approaches.

Researchers first define their constructs and measurement models based on theory. They then choose a statistical method to estimate proxies for these conceptual variables. Since theoretical constructs are never observed directly, every SEM method produces method-specific approximations of the constructs.

This means that reflectively measured constructs can be used in PLS-SEM when the research objective, model design, and data situation support composite-based estimation. Researchers should still carefully assess reflective measurement models using indicator loadings, internal consistency reliability, convergent validity, and discriminant validity.

For discriminant validity, SmartPLS provides the heterotrait-monotrait ratio of correlations (HTMT), which is generally preferred over relying only on the Fornell-Larcker criterion.

Exploratory, Confirmatory, and Mixed SEM Studies

Researchers should clearly distinguish the theorization mode and the analytical mode of their study.

The theorization mode concerns how the study develops and tests theory:

Confirmatory mode: The study tests hypotheses derived before analyzing the data.
Exploratory mode: The study develops ideas, patterns, or hypotheses based on the data.
Confirmatory-first mixed mode: The study first tests pre-specified hypotheses and then explores additional model extensions or boundary conditions.
Exploratory-first mixed mode: The study first explores a model or phenomenon and then tests the generated hypotheses in a separate sample or confirmatory stage.

The analytical mode concerns what the model is expected to achieve:

Explanatory mode: The model should explain observed relationships.
Predictive mode: The model should predict new or unseen observations.
Explanatory-predictive mode: The model should both explain and predict.

Many applied SEM studies are best described as explanatory-predictive. These studies should combine in-sample explanatory assessments with out-of-sample predictive assessments. If exploratory and confirmatory elements are combined, they should be clearly labeled, and confirmatory claims should ideally be tested with a separate sample or a preregistered analysis plan.

Modern Reporting Recommendations for PLS-SEM

When reporting PLS-SEM results, researchers should provide transparent evidence for measurement model quality, structural model results, predictive performance, and robustness.

For reflective measurement models, report:

Indicator loadings.
Internal consistency reliability, including reliability coefficient rho_A where appropriate.
Composite reliability.
Average variance extracted (AVE).
Discriminant validity, preferably using HTMT and bootstrap confidence intervals.

For formative measurement models, report:

Convergent validity using redundancy analysis.
Indicator collinearity using VIF values.
Indicator weights.
Indicator relevance and significance.

For the structural model, report:

Path coefficients.
Bootstrap confidence intervals and significance levels.
R² and adjusted R².
Effect sizes f².
Predictive assessment using PLSpredict and, where appropriate, CVPAT.
Model fit criteria such as SRMR when relevant.
Robustness checks for advanced models and key conclusions.

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.

How to Justify PLS-SEM in a Research Paper

A strong PLS-SEM justification should address the following points:

The research objective emphasizes prediction, explanation, or driver identification.
The model is complex or includes formative constructs.
Composite-based estimation is suitable for the conceptual model and data situation.
Construct scores are needed for follow-up analyses or predictive assessment.
The study assesses out-of-sample prediction, not only in-sample explanation.
Sample size and distributional issues are discussed as supporting reasons, not as the only justification.
Measurement model, structural model, prediction, and robustness checks are reported transparently.
If appropriate, CB-SEM is used as a complementary robustness check to assess model fit and method sensitivity.

A weak justification would be:

PLS-SEM was used because the sample size was small.

A stronger justification would be:

PLS-SEM was used because the study focuses on predicting and explaining key target constructs in a complex model that includes formative measurement. The method also provides construct scores for follow-up analyses and allows the assessment of out-of-sample predictive performance. The restricted sample size, which results from the study's limited population, further supports the choice of PLS-SEM. To assess robustness, the study also compares key structural relationships with CB-SEM results.

FAQ: When to Use PLS-SEM

When should I use PLS-SEM?

Use PLS-SEM when the research focuses on prediction, explanation, driver identification, complex models, formative measurement, construct scores, or out-of-sample predictive assessment.

When should I not use PLS-SEM?

Do not use PLS-SEM when the main goal is strict theory confirmation under common factor assumptions, CFA-based model testing, or global covariance-based model fit assessment. In these cases, CB-SEM is often more appropriate.

Is PLS-SEM only for exploratory research?

No. PLS-SEM can be used in confirmatory, exploratory, and mixed research designs. The key is to clearly label the study's purpose and report whether the analysis is explanatory, predictive, or both.

Can PLS-SEM be used with reflective constructs?

Yes. Reflective measurement is a conceptual specification, while PLS-SEM and CB-SEM are statistical estimation approaches. Reflectively measured constructs can be used in PLS-SEM when the research objective and model design support composite-based estimation.

Is CB-SEM available in SmartPLS?

Yes. SmartPLS supports CB-SEM and CFA in addition to PLS-SEM. Researchers can therefore use SmartPLS for both covariance-based and composite-based SEM analyses.

Why use both PLS-SEM and CB-SEM?

Using both methods can strengthen the robustness of SEM conclusions. CB-SEM helps evaluate model fit and common factor-based assumptions, while PLS-SEM helps evaluate driver relationships, construct scores, and out-of-sample prediction.

Recommended Literature

The following sources provide important guidance on when to use PLS-SEM, when to consider CB-SEM, and how to apply a multimethod SEM perspective:

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., Hult, G. T. M., Ringle, C. M., & Sarstedt, M. (2027). A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM) (4th ed.). Sage.
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.
Bido, D. d. S., & Souza, C. A. (2026). Structural equation modeling: Is it still worth learning? Brazilian Administration Review, 23(3), e260020.
Sharma, P. N., Sarstedt, M., Ringle, C. M., Cheah, J.-H., Herfurth, A., & Hair, J. F. (2024). A framework for enhancing the replicability of behavioral MIS research using prediction oriented techniques. International Journal of Information Management, 78, 102805.
Sarstedt, M., Adler, S. J., Ringle, C. M., Cho, G., Diamantopoulos, A., Hwang, H., & Liengaard, B. D. (2024). Same model, same data, but different outcomes: Evaluating the impact of method choices in structural equation modeling. Journal of Product Innovation Management, 41(6), 1100-1117.
Hair, J. F., Sharma, P. N., Chin, W. W., Sarstedt, M., & Ringle, C. M. (2026). A multimethod SEM framework for analyzing models with latent variables. Journal of Global Marketing, 39(2), 167-182.
Hair, J. F., Hult, G. T. M., Ringle, C. M., Sarstedt, 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.
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.
Richter, N. F., Cepeda Carrión, G., Roldán, 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.
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.
Sarstedt, M., Ringle, C. M., & Hair, J. F. (2021). Partial least squares structural equation modeling. In C. Homburg, M. Klarmann, & A. Vomberg (Eds.), Handbook of Market Research. Springer.
Ringle, C. M., Wende, S., & Becker, J.-M. (2024). SmartPLS 4. SmartPLS. https://www.smartpls.com/