Summary of the Article
“A Multimethod SEM Framework for Analyzing Models with Latent Variables”
by Hair, Sharma, Chin, Sarstedt & Ringle (2026) published in Journal of Global Marketing
1. Background and Purpose of the Study
The article addresses an important issue in social science research: how researchers analyze complex relationships between concepts that cannot be directly measured, such as satisfaction, trust, loyalty, or motivation. These concepts are called latent variables.
To study relationships between such variables, researchers commonly use a statistical technique called Structural Equation Modeling (SEM). However, there are two main approaches to SEM:
- Factor-based SEM (CB-SEM): focuses mainly on explaining theoretical relationships.
- Composite-based SEM (PLS-SEM): focuses more on predicting outcomes.
Traditionally, researchers tend to choose only one of these methods and use it exclusively in their analysis. The authors argue that this practice is problematic because each method has strengths and weaknesses, and relying on only one method may lead to incomplete or misleading conclusions.
A Multimethod SEM Framework for…
The main goal of the article is therefore to introduce a new multimethod SEM framework that uses both approaches together to analyze research models more reliably.
2. Key Idea of the Article
The authors propose that instead of debating which SEM method is better, researchers should use multiple methods together to test the same model.
They compare this idea to medical diagnosis: Doctors do not rely on a single test (like an X-ray); they combine multiple tests (blood tests, scans, etc.) to reach a more accurate diagnosis. Similarly, researchers should combine different analytical methods to better understand relationships between variables.
Thus, the paper promotes a shift from “method loyalty” (defending a preferred statistical method) to “model evaluation” (testing whether the theoretical model actually works).
3. Differences Between the Two SEM Approaches
The article explains why both SEM methods are useful but different.
A. Factor-based SEM (CB-SEM)
This method is mainly used to:
- Test theoretical explanations
- Evaluate how well a model fits the data
- Confirm whether hypothesized relationships are supported by theory
It focuses on explanatory power, meaning how well the model explains relationships among variables.
B. Composite-based SEM (PLS-SEM)
This method is more suitable for:
- Prediction
- Identifying variables that help forecast outcomes
- Evaluating how well the model predicts new data
It emphasizes predictive power, meaning whether the model can accurately predict future observations.
4. The Main Problem Identified in the Literature
The authors highlight an important issue in research practice:
- Many studies focus only on explanation (testing theory)
- Few studies examine whether the model can predict outcomes
However, good research should ideally do both.
A model that explains relationships but cannot predict outcomes has limited practical value. On the other hand, a model that predicts well but lacks theoretical explanation may produce misleading conclusions.
Therefore, the authors argue that explanation and prediction should be evaluated together.
5. The Proposed Multimethod SEM Framework
The article introduces a step-by-step workflow for applying both SEM approaches to the same model.
Step 1 – Develop the theoretical model: Researchers define variables and hypotheses based on existing theory.
Step 2 – Check the data: Researchers examine issues such as missing data or unusual values.
Step 3 – Validate the measurement model: They ensure that survey questions or indicators measure the constructs properly.
Step 4 – Evaluate the model overall: Two aspects are tested:
· Explanatory validity (how well the model fits the data)
· Predictive validity (how well it predicts new data)
Step 5 – Estimate the model using CB-SEM: This evaluates whether the hypothesized relationships are statistically significant.
Step 6 – Estimate the same model using PLS-SEM: This evaluates the model using a different statistical logic.
Step 7 – Compare the results: Researchers check whether the results from both methods are similar or different.
Step 8 – Test predictive ability: The model is tested on new or unseen data to see whether it predicts outcomes.
Step 9 – Integrate the results: Researchers interpret the findings by combining explanatory and predictive evidence.
6. Four Possible Outcomes for Each Relationship in a Model
The authors explain that each relationship between variables can fall into four categories:
1. No theoretical or predictive relevance: The relationship is not supported by theory and does not improve predictions.
2. Theoretical relevance only: The relationship is statistically significant but does not improve prediction. This may indicate overfitting or limited practical usefulness.
3. Predictive relevance only: The relationship improves prediction but lacks theoretical support. This suggests that unknown mechanisms or missing theory may exist.
4. Both theoretical and predictive relevance (ideal case): The relationship is statistically supported and improves prediction. This provides strong evidence for both theory and practice.
7. Importance of Convergence and Divergence
A central concept in the framework is comparing results from the two SEM methods.
Convergence
When both methods produce similar results:
- Confidence in the findings increases
- The relationship is likely robust
Divergence
When the methods produce different results:
- The relationship may depend on modeling assumptions
- The measurement or theory may need revision
Instead of treating disagreement as a problem, the authors suggest it can be valuable diagnostic information.
8. Major Contributions of the Article
The article makes several important contributions to research methodology.
1. A new analytical framework: It provides a systematic method for combining different SEM approaches.
2. Integration of explanation and prediction: The framework helps researchers evaluate models for both theoretical insight and practical usefulness.
3. Improved reliability of findings: Using multiple methods helps ensure that results are not dependent on a single statistical technique.
4. Greater transparency in research: The approach encourages researchers to report how results change under different analytical assumptions.
9. Practical Implications for Researchers
The framework encourages researchers to:
- Avoid relying on a single statistical method
- Evaluate models using multiple analytical perspectives
- Focus on robust structural relationships rather than defending a specific method
- improve the credibility and usefulness of research findings
The authors also note that modern statistical software makes it increasingly easy to apply multiple SEM methods in the same study.
10. Overall Conclusion
The article argues that structural equation modeling should not rely on a single analytical tradition. Instead, researchers should combine factor-based and composite-based SEM approaches.
By applying both methods to the same model, researchers can:
- obtain stronger theoretical insights,
- evaluate predictive performance,
- and produce more reliable conclusions.
The proposed multimethod SEM framework therefore represents an important step toward more robust, transparent, and practically relevant research in the social sciences.