Composite vs. Common Factor Models in PLS-SEM
One of the first decisions in structural equation modeling (SEM) is how each construct should be represented: as a composite or as a common factor. In SmartPLS (Ringle et al., 2024), constructs can be modeled as composites, as common factors, or as a combination of both within the same model. This choice is not merely technical. It determines the estimation algorithm, the meaning of the construct scores, and the way results must be interpreted and reported (Rigdon et al., 2017; Sarstedt et al., 2016).
In short:
- A composite model represents a construct as a weighted linear combination of its indicators. The construct is formed by the indicators and carries no separate measurement-error term.
- A common factor model represents a construct as a latent variable that gives rise to its indicators. Each indicator is an error-prone reflection of the underlying construct.
- A mixed model combines both perspectives, specifying some constructs as composites and others as common factors within a single PLS path model.
SmartPLS estimates these model types through the PLS-SEM algorithm, consistent PLS-SEM (PLSc-SEM), and CB-SEM/CFA. The representation should be driven by theory and the research objective, not chosen solely on statistical grounds (Guenther et al., 2023; Rigdon et al., 2017).
Composite vs. Common Factor: Quick Comparison
| Aspect | Composite model | Common factor model |
|---|---|---|
| Conceptual view | Construct is formed as a combination of indicators (an emergent variable) | Construct is a latent variable that explains the indicator responses |
| Measurement error | No separate error term at the construct level | Indicators contain measurement error; the construct is error-free |
| Typical algorithm in SmartPLS | PLS-SEM | PLSc-SEM (approximation) or CB-SEM/CFA |
| Default weighting | Mode A (reflective) or Mode B (formative) | Mode A, then correction for attenuation (PLSc-SEM) |
| Primary objective | Prediction and explanation of target constructs | Testing measurement theory and reproducing the covariance structure |
| Interpretation of scores | Weighted combination of indicators | Estimate of the underlying latent variable |
A key point that is often overlooked: in the standard PLS-SEM algorithm, every construct is estimated as a composite, regardless of whether its indicators are specified as reflective (Mode A) or formative (Mode B). Reflective indicators do not, by themselves, produce a common factor — the measurement specification (reflective or formative) and the statistical estimation (composite or common factor) are two distinct layers of the research process (Guenther et al., 2025; Sarstedt et al., 2016). Approximating a common factor requires PLSc-SEM or a covariance-based approach such as CB-SEM (Dijkstra & Henseler, 2015a; Rigdon et al., 2017).
Composite Models
Composite models are the standard representation in PLS-SEM. The PLS-SEM algorithm estimates each construct as a weighted composite of its indicators. This makes composite models especially useful when the research objective is prediction, the explanation of target constructs, or the estimation of theoretically defined composites such as indices and design constructs (Hair et al., 2019).
The weighting mode determines how the outer weights are computed:
- For reflective measurement models, where the relationships point from the construct to the indicators, SmartPLS uses Mode A by default. Mode A derives each outer weight from the bivariate correlation (covariance) between the construct score and the corresponding indicator.
- For formative measurement models, where the relationships point from the indicators to the construct, SmartPLS uses Mode B by default. Mode B derives the outer weights from a multiple regression in which the indicators predict the construct score.
In both cases the resulting construct is a composite: a weighted linear combination of its indicators. Users can change the construct-score computation by double-clicking a construct in the model. Depending on the model and analysis objective, SmartPLS offers Mode A, Mode B, sum scores, and predefined weights.
Common Factor Models
Common factor models assume that the construct is a latent variable causing the observed indicator responses. In this view, indicators are imperfect measures of an underlying construct and contain random measurement error. This is the classic assumption behind confirmatory factor analysis and covariance-based SEM.
Importantly, a reflective measurement specification is not the same as a common factor model. Specifying a construct reflectively or formatively is a conceptual decision grounded in measurement theory, whereas the common factor versus composite distinction concerns the statistical estimation and the assumed data-generation process (Guenther et al., 2025; Hair et al., 2027; Sarstedt et al., 2016). Reflectively specified constructs can therefore be validly estimated as composites; equating reflective measurement with common factor models conflates these two layers of the research process.
When common factor assumptions are appropriate, SmartPLS can approximate common factor results with consistent PLS-SEM (PLSc-SEM). PLSc-SEM first estimates construct scores using Mode A and then corrects the estimates for attenuation using the reliability coefficient rho_A (ρ_A; Dijkstra & Henseler, 2015a, 2015b). This correction disattenuates the indicator loadings, construct correlations, and path coefficients so that they approximate common factor model estimates.
Use PLSc-SEM when the theoretical model assumes common factors and the constructs are measured reflectively. Because PLSc-SEM relies on common factor assumptions, researchers should carefully evaluate measurement model quality, model specification, and any potential inadmissible results (for example, out-of-range correlations or standardized loadings above one).
For covariance-based common factor modeling, SmartPLS also offers CB-SEM and CFA. These are appropriate when the research objective emphasizes testing measurement theory, comparing nested models, or reproducing the observed covariance structure (Rigdon et al., 2017; Sarstedt et al., 2016).
Mixed Composite and Common Factor Models
Mixed models include both composite-based and common factor-based constructs in the same PLS path model. This is useful when theory suggests that some constructs are best represented as composites (for example, formatively measured design or index constructs) while others are best represented as common factors (for example, reflectively measured attitudinal traits).
In a mixed model estimated with PLSc-SEM, SmartPLS applies the disattenuation correction only to the reflectively measured common factor constructs. Constructs specified as composites—especially formative measurement models—are not corrected; they remain weighted composites, typically estimated with Mode B by default.
Mixed models require careful theoretical justification. Researchers should explain why each construct is represented as a composite or a common factor and should avoid selecting the representation on statistical grounds alone (Rigdon et al., 2017; Sarstedt et al., 2016). Interpretation follows the specification: composites are read as weighted combinations of their indicators, whereas common factors are read as latent variables that account for the shared variance among their indicators.
Choosing Between Composite and Common Factor Representations
The decision should follow the conceptual nature of the construct and the goal of the study:
- Choose a composite when the construct is best understood as an emergent combination of its facets, when indicators are formative, or when the objective is prediction and explanation (Guenther et al., 2023; Hair et al., 2019). Because the true data-generation process is unknown in applied research, the composite model estimation is often the safer choice (Sarstedt et al., 2016).
- Choose a common factor when the theory assumes that a latent variable explains the indicators and the objective is to test that measurement theory and the covariance structure (Rigdon et al., 2017).
- Do not let fit statistics alone decide. Composite and common factor models answer different questions, and comparing their results requires care, because the two approaches estimate conceptually different quantities (Rigdon et al., 2017; Sarstedt et al., 2016).
- Consider a multimethod SEM approach. Rather than treating the methods as mutually exclusive, researchers can estimate the same conceptual model with composite-based (PLS-SEM) and factor-based (PLSc-SEM, CB-SEM) methods and assess the robustness of their conclusions across methods (Guenther et al., 2025).
Switching Modes in SmartPLS
By default, SmartPLS uses Mode A with reflective measurement models and Mode B with formative measurement models. To change the construct-score computation, double-click the construct in the model and select the preferred estimation option.
Available options include:
- Mode A — Correlation weights, typically used for reflective measurement models.
- Mode B — Regression weights, typically used for formative measurement models.
- Sum scores — Equal (unit) weights that combine indicators without estimating weights.
- Predefined weights — User-specified weights based on theory, prior studies, or external information.
Changing the mode changes how construct scores are computed and can affect the results. Document the selected mode for each construct and justify any deviation from the default settings.
Reporting Composite, Common Factor, and Mixed Models
When reporting results, clearly state whether the model was estimated as a composite model, a common factor model, or a mixed model. Describe the algorithm used (PLS-SEM, PLSc-SEM, or CB-SEM) and report the construct-score settings for the relevant constructs (Hair et al., 2019).
For transparent reporting, include the following:
- The theoretical rationale for modeling each construct as a composite or a common factor.
- The estimation algorithm used (PLS-SEM, PLSc-SEM, or CB-SEM).
- The weighting mode for each construct whenever it differs from the default.
- The assessment criteria applied to the reflective and formative measurement models.
- Any inadmissible results, estimation issues, or changes to the model specification.
Frequently Asked Questions
What is the difference between a composite and a common factor?
A composite is a weighted linear combination of indicators that forms the construct, with no construct-level measurement error. A common factor is a latent variable assumed to explain its indicators, each of which contains measurement error. Composites are the default in PLS-SEM; common factors are approximated with PLSc-SEM or estimated with CB-SEM/CFA.
Does PLS-SEM estimate common factors?
Not directly. The standard PLS-SEM algorithm estimates every construct as a composite, even when the indicators are reflective. To approximate common factor results, use PLSc-SEM (consistent PLS-SEM) or a covariance-based method such as CB-SEM.
When should I use PLSc-SEM instead of PLS-SEM?
Use PLSc-SEM when your theory assumes common factors and your constructs are measured reflectively, and you want disattenuated (common-factor-consistent) estimates. Keep the standard PLS-SEM algorithm when your constructs are conceptualized as composites or when your indicators are formative.
Can one model contain both composites and common factors?
Yes. In a mixed model estimated with PLSc-SEM, the disattenuation correction is applied only to reflectively measured common factor constructs; composites (including formative constructs) are left uncorrected.
Does a reflective measurement model require a common factor model?
No. The reflective versus formative specification is a measurement-theoretic decision, while the common factor versus composite distinction concerns the statistical estimation and the assumed data-generation process. Reflectively specified constructs can be validly estimated as composites with PLS-SEM (Guenther et al., 2025; Sarstedt et al., 2016).
Is the composite-versus-common-factor choice a statistical decision?
No. It is primarily a measurement-theory decision that should be justified conceptually. Statistical criteria can support the assessment but should not be the sole basis for the choice (Guenther et al., 2025; Rigdon et al., 2017; Sarstedt et al., 2016).
Related SmartPLS Resources
References
- Dijkstra, T. K., & Henseler, J. (2015a). Consistent Partial Least Squares Path Modeling. MIS Quarterly, 39(2), 297–316.
- Dijkstra, T. K., & Henseler, J. (2015b). Consistent and Asymptotically Normal PLS Estimators for Linear Structural Equations. Computational Statistics & Data Analysis, 81, 10–23.
- 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., Hult, G. T. M., Ringle, C. M., & Sarstedt, M. (2022). A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM) (3rd ed.). Sage.
- 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., 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., Ringle, C. M., Sarstedt, M., & Gudergan, S. P. (2018). Advanced Issues in Partial Least Squares Structural Equation Modeling. Sage.
- Hair, J. F., Sarstedt, M., Ringle, C. M., & Gudergan, S. P. (2024). Advanced Issues in Partial Least Squares Structural Equation Modeling (2nd ed.). Sage.
- 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, 616–632.
- Henseler, J., Dijkstra, T. K., Sarstedt, M., Ringle, C. M., Diamantopoulos, A., Straub, D. W., Ketchen, D. J., Hair, J. F., Hult, G. T. M., & Calantone, R. J. (2014). Common Beliefs and Reality about Partial Least Squares: Comments on Rönkkö and Evermann (2013). Organizational Research Methods, 17(2), 182–209.
- 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 – Journal of Research and Management, 39(3), 4–16.
- Ringle, C. M., Wende, S., & Becker, J.-M. (2024). SmartPLS 4. SmartPLS GmbH. 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.
