CBSEM using #SmartPLS4 | 11 | Simple Structural Model Analysis
Based on Research With Fawad's video on YouTube. If you like this content, support the original creators by watching, liking and subscribing to their content.
Build a full structural model after measurement model validation, adding direct paths between latent constructs while retaining indicators and error terms to account for measurement error.
Briefing
Structural model analysis in SmartPLS4 is used to test whether hypothesized relationships between latent constructs hold—while explicitly accounting for measurement error through the model’s indicators, error terms, and latent variables. After a measurement model has already established reliability and validity, the next step is to build a full structural model that links constructs (e.g., direct paths from predictors to an outcome) and then evaluate both model fit and the statistical significance of the paths.
In this session, the structural model targets the impact of organizational learning (OC) and collaborative culture (CC) on organizational performance (OP). The workflow starts by creating the model with CC, OC, and OP as latent variables, then attaching their measurement indicators on the left side and drawing the direct paths among constructs. Because the approach is covariance-based structural equation modeling, the exogenous constructs (CC and OC) are correlated in the model.
Model fit is checked first using the “Calculate” function with the basic PLS-SEM algorithm. The results are described as “poor” at first glance but still considered acceptable because key fit guidance used here is that the relevant fit statistic is less than 5. The video then moves to hypothesis testing: significance of the CC→OP and OC→OP relationships is assessed via bootstrapping. A one-tailed test is selected because the hypotheses specify positive effects. Although the usual recommendation is 5,000–10,000 bootstrap samples, the session uses 500 for speed, applying bias-corrected and accelerated (BCa) bootstrapping to stabilize estimates.
The bootstrapping output is used to judge significance through T values and P values. With a one-tailed threshold of 1.645 for the T statistic, both paths clear the cutoff: CC has a standardized path coefficient (beta) of 0.273 with T greater than 1.645 and P less than 0.05, indicating a significant positive effect on OP. OC also shows a significant positive effect on OP, with a stronger impact than CC based on the relative size of the standardized coefficients.
Finally, the session interprets explanatory power using R square (R²) for the dependent construct. For organizational performance, R² is reported as 0.469 (about 46.9% of the variance in OP explained by CC and OC). The reporting guidance emphasizes presenting standardized regression weights (standardized betas), T values, P values, model fit statistics, and R² for the dependent variable—while avoiding re-reporting measurement indicators and loadings, since those belong in the measurement model results.
A practical reporting template is provided: accept a “good fitting model” when the relevant criterion (CMIN) is less than 5, and use additional fit indices such as TLI and CFI (with values above 0.90) plus SRMR and RMSEA thresholds (noted as part of the fit-index set). The session concludes with a concise checklist for how to populate a structural model results table: fit indices, R², and the hypothesis test results (beta, T, P) for each hypothesized path.
Cornell Notes
The structural model step in SmartPLS4 tests whether hypothesized relationships between latent constructs are statistically significant, after the measurement model has already confirmed reliability and validity. A full structural model includes latent variables, their indicators, and error terms, allowing relationships between constructs to account for measurement error. In the example, collaborative culture (CC) and organizational learning (OC) are linked to organizational performance (OP) with direct paths, and the exogenous constructs are correlated (covariance-based SEM). Bootstrapping (BCa, one-tailed) is used to evaluate significance using T values (threshold 1.645) and P values (<0.05). The model’s explanatory power is summarized with R² for OP (reported as 0.469), and results are reported using standardized betas, T values, P values, fit indices, and R²—without repeating measurement indicator loadings.
What changes when moving from a measurement model to a full structural model in SmartPLS4?
Why does the model correlate exogenous constructs like CC and OC in this workflow?
How are hypotheses tested for significance in the structural model results?
What does the R² value for organizational performance mean here?
What should be reported for structural model analysis, and what should be omitted?
Review Questions
- In a covariance-based SEM setup, how should exogenous constructs be treated in the model diagram (correlated vs. causally linked)?
- What decision rule is used for path significance with a one-tailed test in this workflow (T and P thresholds)?
- Which statistics are prioritized when reporting structural model results, and why are measurement indicator loadings not repeated?
Key Points
- 1
Build a full structural model after measurement model validation, adding direct paths between latent constructs while retaining indicators and error terms to account for measurement error.
- 2
For covariance-based SEM in SmartPLS4, correlate exogenous constructs (e.g., CC and OC) rather than drawing causal paths between them.
- 3
Check model fit first using the basic PLS-SEM algorithm, then proceed to bootstrapping for hypothesis testing.
- 4
Use bootstrapping (BCa) to test path significance; with a one-tailed test, treat T > 1.645 and P < 0.05 as significant.
- 5
Report standardized betas, T values, and P values for each hypothesized relationship (CC→OP and OC→OP).
- 6
Interpret explanatory power using R² for the dependent construct; here, OP has R² = 0.469 (46.9% variance explained).
- 7
When writing results, include fit indices, R², and path test statistics, but avoid re-reporting measurement indicator loadings already covered in the measurement model.