Before interpreting relationships between variables in a PLS-SEM model, researchers need to establish whether their measurement model is reliable and valid. Measurement model assessment in SmartPLS helps determine whether the indicators adequately represent their intended constructs and whether those constructs measure distinct concepts.
A model may produce interesting path coefficients, but those results should not be interpreted confidently if the underlying measurements are weak. This is why researchers should evaluate indicator reliability, internal consistency reliability, convergent validity, and discriminant validity before proceeding to structural model assessment.
However, the assessment criteria depend on whether the constructs are specified as reflective or formative. Applying the wrong evaluation criteria can lead to incorrect conclusions about measurement quality.
In this guide, we explain how to assess the measurement model in SmartPLS, interpret the main output values, understand commonly used thresholds, and address measurement issues without making unnecessary changes to the research instrument.
What Is Measurement Model Assessment in SmartPLS?
Measurement model assessment is the process of evaluating how well observed indicators measure the latent constructs in a research model.
For example, a study examining customer satisfaction may measure the construct using several questionnaire items related to product quality, expectations, service experience, and overall satisfaction. Measurement model assessment determines whether these items adequately represent the intended construct.
In Partial Least Squares Structural Equation Modeling (PLS-SEM), researchers generally evaluate two types of measurement models:
Reflective measurement model: The construct is understood to influence or give rise to its indicators. For example, higher customer satisfaction may lead respondents to agree with several satisfaction-related statements.
Formative measurement model: The indicators collectively form or contribute to the construct. For example, different aspects of service quality may combine to form an overall service index.
The assessment criteria differ between these two models. Therefore, researchers should identify the correct measurement specification before interpreting SmartPLS results.
Key Criteria for Assessing a Reflective Measurement Model
For reflective constructs, measurement model assessment generally involves four main areas.
| Assessment criterion | What it evaluates | Common guideline |
|---|---|---|
| Indicator reliability | How strongly each indicator represents its construct | Outer loading around 0.708 or higher |
| Internal consistency reliability | Whether indicators consistently measure the same construct | Cronbach’s alpha, rho_A, and composite reliability generally ≥ 0.70 |
| Convergent validity | Whether indicators of a construct share sufficient variance | AVE ≥ 0.50 |
| Discriminant validity | Whether constructs are empirically distinct | HTMT commonly below 0.85 or 0.90, depending on the selected criterion |
These are commonly used guidelines rather than automatic pass-or-fail rules. Researchers should interpret them alongside the theoretical meaning of each construct, research design, measurement quality, and relevant methodological guidance.
How to Assess the Measurement Model in SmartPLS: Step-by-Step
Follow these steps to evaluate a reflective measurement model using SmartPLS.
Step 1: Run the PLS-SEM Algorithm
Begin by opening your SmartPLS project and checking that the measurement model has been specified correctly.
Before running the analysis, verify:
All latent constructs are correctly defined.
Indicators are assigned to their intended constructs.
Reflective and formative measurement directions are specified correctly.
Indicator labels match the questionnaire and research instrument.
The dataset has been imported correctly, with no unintended missing or incorrectly coded values.
Once the model is ready:
Open your PLS-SEM model.
Select the Calculate option.
Choose PLS-SEM Algorithm.
Review the calculation settings.
Run the algorithm.
SmartPLS will generate results that include outer loadings, construct reliability and validity, and discriminant validity statistics. The exact organisation of the results interface may vary between software versions.
These outputs provide the starting point for measurement model evaluation. However, researchers should first confirm that the assessment criteria match the measurement specification of each construct.
Step 2: Assess Indicator Reliability Using Outer Loadings
Outer loadings indicate how strongly an observed indicator is associated with its assigned latent construct.
For reflective measurement models, an outer loading of approximately 0.708 or higher is generally desirable. At this level, the construct explains approximately 50% or more of the indicator’s variance.
Consider this illustrative example:
| Indicator | Outer loading | Interpretation |
|---|---|---|
| CS1 | 0.842 | Meets the recommended guideline |
| CS2 | 0.816 | Meets the recommended guideline |
| CS3 | 0.761 | Meets the recommended guideline |
| CS4 | 0.645 | Requires further examination |
CS1, CS2, and CS3 meet the commonly recommended loading guideline. CS4 has a lower loading and requires further examination.
Should researchers remove indicators below 0.70?
Not automatically. Indicators with loadings between 0.40 and 0.70 should be considered for removal only after evaluating their effect on composite reliability and AVE, along with their theoretical importance and contribution to content validity.
Removing an indicator simply to improve a statistical value may weaken the construct’s conceptual coverage. Researchers should also examine whether a low loading reflects a measurement problem, unclear questionnaire wording, or a theoretically important aspect of the construct.
Indicators with very low loadings may require closer attention, but removal should always be supported by a methodological and theoretical justification.
Step 3: Evaluate Internal Consistency Reliability
Internal consistency reliability determines whether the indicators of a reflective construct consistently measure the same underlying concept.
SmartPLS provides three commonly reported reliability measures.
1. Cronbach’s Alpha
Cronbach’s alpha estimates the internal consistency of a set of indicators. A value of 0.70 or higher is generally considered acceptable in many research settings.
However, Cronbach’s alpha assumes certain measurement conditions and can underestimate reliability when indicators have different loadings. It should therefore be interpreted alongside other reliability measures.
2. Composite Reliability (rho_C)
Composite reliability considers the individual indicator loadings rather than assuming that all indicators contribute equally.
A value between 0.70 and 0.95 is generally desirable. Values above 0.95 may indicate that some indicators are excessively similar or redundant.
3. rho_A
Dijkstra-Henseler’s rho_A is another reliability measure used in PLS-SEM. It provides additional evidence of internal consistency and is commonly reported alongside Cronbach’s alpha and composite reliability.
A value of 0.70 or higher is generally used as a guideline for acceptable reliability.
The following table summarises the commonly used interpretation:
| Reliability measure | General guideline |
|---|---|
| Cronbach’s alpha | ≥ 0.70 |
| Composite reliability (rho_C) | ≥ 0.70 and preferably below 0.95 |
| rho_A | ≥ 0.70 |
Where can we find these values in SmartPLS?
Navigate to the PLS-SEM Algorithm results and open the Quality Criteria section. Select Construct Reliability and Validity to review the available reliability measures for each construct.
Do not rely on a single reliability measure when the other available measures provide additional information. Evaluate the results together and consider the conceptual meaning of the indicators.
Step 4: Assess Convergent Validity Using AVE
Convergent validity determines whether indicators intended to measure the same construct share a sufficient proportion of variance.
In SmartPLS, researchers commonly evaluate convergent validity using Average Variance Extracted (AVE).
The commonly recommended guideline is:
AVE ≥ 0.50
An AVE of 0.50 indicates that a construct explains at least half of the variance in its indicators, on average.
For example:
| Construct | AVE | Interpretation |
|---|---|---|
| Customer Satisfaction | 0.642 | Meets the recommended guideline |
| Perceived Value | 0.571 | Meets the recommended guideline |
| Customer Loyalty | 0.483 | Requires further examination |
Customer Satisfaction and Perceived Value meet the recommended guideline. Customer Loyalty has an AVE below 0.50, suggesting that its indicators may not explain enough shared variance.
When AVE falls below the recommended threshold, review the indicator loadings, reliability statistics, and theoretical relevance of the measurement items.
Do not remove indicators without examining the impact on the overall construct. A statistically improved AVE does not necessarily mean that the revised measurement model represents the original concept more accurately.
Step 5: Assess Discriminant Validity Using HTMT
Discriminant validity determines whether each construct measures a concept that is empirically distinct from other constructs in the model.
For example, a research model may include Customer Satisfaction, Customer Loyalty, and Perceived Value. Although these constructs may be related, they should not measure essentially the same concept.
The Heterotrait-Monotrait Ratio (HTMT) is a widely recommended method for assessing discriminant validity in reflective measurement models.
Commonly applied thresholds include:
HTMT below 0.85: A stricter criterion often used when constructs are conceptually distinct.
HTMT below 0.90: A more lenient criterion that may be suitable for conceptually related constructs.
The appropriate threshold should be selected based on the conceptual relationship between the constructs and the methodological justification for the study.
Consider this illustrative HTMT matrix:
| Construct relationship | HTMT value | Interpretation |
|---|---|---|
| Satisfaction – Loyalty | 0.782 | Meets the 0.85 guideline |
| Satisfaction – Perceived Value | 0.841 | Meets the 0.90 guideline, but not 0.85 |
| Loyalty – Perceived Value | 0.927 | Exceeds the 0.90 guideline and requires review |
The relationship between Loyalty and Perceived Value has an HTMT value of 0.927, exceeding the commonly used 0.90 threshold. This indicates a potential discriminant validity concern.
Researchers should also examine the bootstrapped HTMT confidence interval. If the confidence interval includes 1, this provides evidence that discriminant validity may not be established.
How to check HTMT in SmartPLS:
Run the PLS-SEM Algorithm.
Open the results.
Navigate to Quality Criteria.
Select Discriminant Validity.
Review the HTMT values for each construct pair.
Run bootstrapping if confidence interval assessment is required.
Although SmartPLS also provides the Fornell-Larcker criterion and cross-loadings, HTMT is generally recommended as a primary method for assessing discriminant validity because traditional approaches may fail to identify certain discriminant validity problems.
Step 6: Review the Measurement Model Before Proceeding
After evaluating the individual criteria, review the measurement model as a whole.
A useful assessment checklist includes:
Are the indicator loadings acceptable or theoretically justified?
Do the reliability measures provide adequate evidence of internal consistency?
Does each reflective construct meet the recommended AVE guideline?
Are the HTMT values within the selected thresholds?
Have any measurement concerns been examined and documented?
Are the measurement specifications consistent with the theoretical framework?
This final review helps ensure that researchers do not interpret individual statistical values in isolation.
What If Your Measurement Model Does Not Meet the Recommended Thresholds?
A measurement model may not satisfy every recommended criterion on the first attempt. This does not mean that researchers should immediately delete indicators or modify the model.
Instead, review the specific issue and identify its possible cause.
| Issue | What to examine | Recommended action |
|---|---|---|
| Low outer loading | Indicator wording and theoretical relevance | Review the item before considering removal |
| Low composite reliability | Consistency and quality of indicators | Examine weak or inconsistent indicators |
| AVE below 0.50 | Shared variance among indicators | Review outer loadings and construct specification |
| HTMT above 0.90 | Conceptual overlap between constructs | Revisit construct definitions and item wording |
| Very high composite reliability | Potentially repetitive indicators | Examine item redundancy and content coverage |
Any indicator removal or model modification should have a methodological justification. Researchers should avoid repeatedly changing the model simply to obtain acceptable statistical values.
If two constructs show high HTMT values, examine whether their definitions or questionnaire items overlap. The issue may originate in the conceptual framework or research instrument rather than the SmartPLS calculation itself.
At Research10X, we emphasise that measurement model modifications should be supported by statistical evidence and theoretical reasoning. Our SmartPLS SEM consulting and analysis services address measurement model assessment and interpretation so that researchers can understand the implications of their analytical decisions.
How to Assess a Formative Measurement Model in SmartPLS
The criteria discussed above primarily apply to reflective measurement models. Formative constructs require a different assessment procedure because their indicators contribute to forming the construct rather than reflecting an underlying latent variable.
For formative measurement models, researchers should generally examine three important areas.
1. Convergent Validity Through Redundancy Analysis
Redundancy analysis evaluates whether the formative construct is sufficiently related to an appropriate global measure of the same concept.
Researchers should use a theoretically suitable global measure to examine whether the formative construct demonstrates the expected relationship with that measure.
2. Indicator Collinearity Using VIF
Variance Inflation Factor (VIF) helps identify excessive collinearity among formative indicators.
Values below 3 are often preferred, while values below 5 are commonly treated as an acceptable upper guideline in some methodological contexts. Researchers should justify the threshold they apply rather than treating one value as universally mandatory.
3. Significance and Relevance of Outer Weights
Bootstrapping can be used to evaluate the statistical significance of formative indicator weights.
Researchers should also consider the absolute contribution and theoretical relevance of indicators, including cases where an outer weight is not statistically significant.
A non-significant outer weight does not automatically justify removing an indicator. Its absolute contribution, theoretical importance, and content validity should also be examined.
Important: Cronbach’s alpha, composite reliability, AVE, and HTMT should not be applied as routine assessment criteria to formative constructs. Their assessment requires criteria appropriate to formative measurement specifications.
What Comes After Measurement Model Assessment?
Once the measurement model has demonstrated adequate reliability and validity, researchers can proceed to structural model assessment.
The structural model evaluates the hypothesised relationships between constructs. Depending on the research objectives, this may involve:
Collinearity assessment using VIF: Examines whether predictor constructs have problematic collinearity.
Path coefficient evaluation: Examines the direction and strength of relationships between constructs.
Bootstrapping: Evaluates statistical significance and confidence intervals for estimated relationships.
Coefficient of determination (R²): Evaluates the explanatory power of the model for endogenous constructs.
Effect size (f²): Examines the contribution of an exogenous construct to the R² of an endogenous construct.
Predictive assessment: Evaluates the model’s ability to predict relevant outcomes using appropriate procedures.
Measurement model assessment and structural model assessment serve different purposes. Establishing measurement quality does not automatically confirm the research hypotheses. Each stage requires its own evaluation and interpretation.
For a complete PLS-SEM analysis, researchers should ensure that measurement model results are adequately established before drawing conclusions about structural relationships and hypothesis testing.
How Research10X Supports SmartPLS Measurement Model Assessment
At Research10X, we recognise that interpreting SmartPLS results involves more than checking whether individual values meet recommended thresholds. Researchers also need to understand whether their measurement specifications, indicators, and statistical findings align with their research objectives and methodology.
Through our SmartPLS SEM consulting and analysis services, we provide support with:
Reflective and formative measurement model assessment.
Indicator loading evaluation.
Cronbach’s alpha, rho_A, and composite reliability.
Convergent validity using AVE.
Discriminant validity using HTMT.
Structural model evaluation and hypothesis testing.
Bootstrapping, mediation, and moderation analysis.
Statistical interpretation and academic reporting.
We also consider the relationship between measurement decisions and the wider research design. When measurement concerns originate from questionnaire design or construct specification, our research methodology consulting services are relevant to reviewing the methodological reasoning behind those decisions.
Our focus is on methodological clarity and meaningful interpretation so researchers can understand their SmartPLS outputs and present their findings appropriately.
Strengthen Your SmartPLS Measurement Model Before Interpreting Results
A reliable measurement model provides the foundation for meaningful PLS-SEM analysis. By evaluating indicator reliability, internal consistency reliability, convergent validity, and discriminant validity, researchers can identify measurement concerns before interpreting relationships between constructs.
Researchers should also distinguish between reflective and formative measurement specifications because the assessment criteria are not interchangeable.
At Research10X, we encourage a methodology-first approach to SmartPLS analysis. The objective is not simply to achieve acceptable statistical values, but to ensure that the measurement model is theoretically justified, methodologically appropriate, and clearly reported.
FAQ's - Measurement Model Assessment in SmartPLS
What is an acceptable outer loading in SmartPLS?
2. What is the acceptable AVE value in SmartPLS?
We generally consider an AVE value of 0.50 or higher acceptable for reflective constructs. This indicates that the construct explains at least half of the variance in its indicators, on average. When AVE falls below 0.50, we review the indicator loadings and theoretical relevance of the measurement items before suggesting any modification.
3. What is the difference between rho_A and composite reliability?
We use rho_A and composite reliability as complementary measures of internal consistency reliability. Composite reliability considers the individual indicator loadings, while rho_A provides another estimate of construct reliability based on the PLS-SEM measurement model. We recommend interpreting both alongside Cronbach’s alpha rather than relying on only one measure.
4. Which HTMT value should we use in SmartPLS?
We commonly consider 0.85 as a stricter HTMT threshold and 0.90 as a more lenient guideline for conceptually related constructs. We recommend selecting the appropriate threshold based on the theoretical relationship between the constructs and explaining that choice in the methodology. We also examine bootstrapped HTMT confidence intervals when conducting discriminant validity assessment.
5. Can we assess the structural model before the measurement model?
We recommend assessing the measurement model before interpreting the structural model in studies where construct measurement is involved. We first examine whether the indicators adequately measure their intended constructs. Once measurement quality has been established, we proceed to structural model evaluation, path coefficient interpretation, and hypothesis testing.
6. Should we remove indicators with outer loadings below 0.70?
We do not recommend removing indicators solely because their outer loadings are below 0.70. We first examine their effect on composite reliability and AVE, along with their theoretical importance and contribution to content validity. We recommend indicator removal only when statistical evidence and methodological reasoning support that decision.
7. Do formative constructs require Cronbach's alpha and AVE?
We do not use Cronbach’s alpha, composite reliability, or AVE as routine assessment criteria for formative constructs. Instead, we examine indicator collinearity using VIF, convergent validity through redundancy analysis, and the significance and relevance of outer weights. We also consider the theoretical importance of each indicator when evaluating the formative measurement model.



