As indicated at the beginning of this project, my aim is to understand how to use SmartPLS software for sophisticated data analysis like structural equation modeling (SEM). Structural Equation Modeling is commonly employed to explain various statistical relationships concurrently through both visualization and the validation of models. As my learning progressed, I came to understand that there are two distinct types of SEM, which are the PLS-SEM (partial least square structural equation modeling) and the covariance-based structural equation modeling (CB-SEM). However, the choice of which method to use is dependent on the goal of the study.
According to Dash and Paul (2021), PLS-SEM is used when a study is focused on making predictions and theory building, whereas CB-SEM is used when a study is focused on theory testing, confirmation of the hypothesis, and validating model fit. Since the beginning of my project, I have been exploring the PLS-SEM analysis features of the SmartPLS software. One of the instructional resources that made my learning easy was from Dr James Gaskin, a professor of information systems at Brigham Young University (BYU), USA. He also has a wiki page called Gaskination, which contains several contents that simplify the various abstract concepts in SEM. One of the reasons I hold his instructional resources in high regard is due to the numerous insightful concepts he presents and the comprehensive guidance he provides throughout a meticulous SEM procedure, along with numerous pieces of advice and considerations to bear in mind during each stage of the analysis.
When exploring the SmartPLS 4 user interface, I realized that the software has five different analysis models, which are the PLS-SEM, CB-SEM, GSCA (generalized structured component analysis), process, and regression. My previous post has been focused on the PLS-SEM analysis. I just started exploring the CB-SEM analysis, even though findings from what I have read online show that using AMOS or LISREL is more effective for conducting CB-SEM. The latest SmartPLS 4 software has updated features, which can also be used for CB-SEM, as illustrated in the video below.
Just like PLS-SEM approach explored in my previous learning reports, it is also important to always check the quality of my measurement model through the following:
Reliability Tests:
- Cronbach’s Alpha (α) > 0.7 (Internal consistency)
- Composite Reliability (CR) > 0.7 (overall construct reliability).
Validity Tests
- Convergent validity (how well items load onto their constructs)
- Average Variance Extracted (AVE) > 0.5
- Discriminant validity (how distinct one construct is from another)
- Fornell-Larcker Criterion: the square root of AVE for a construct should be greater than its correlations with other constructs.
Another important aspect of my learning is knowing that CB-SEM requires model fit indices to ensure the proposed model aligns with the data. These include the following :
Absolute Fit Indices
- Chi-square (χ²): Should be non-significant (p > 0.05).
- Root Mean Square Error of Approximation (RMSEA) < 0.08 (Good fit).
- Standardized Root Mean Square Residual (SRMR) < 0.08 (Good fit).
Incremental Fit Indices
- Comparative Fit Index (CFI) > 0.90
- Tucker-Lewis Index (TLI) > 0.90
Parsimony Fit Indices
- Adjusted Goodness-of-Fit Index (AGFI) > 0.80
If model fit is poor, one has to refine it by removing low-loading indicators (< 0.5), or by checking for high modification indices (indicating potential cross-loadings).