SEM vs PLS-SEM

As aforeposted, we have discussed about what Structural Equation Modelling (SEM) is, as a research method.

But, there`s alternative called PLS-SEM (Partial Least Square - Structural Equation Modelling).

Let us explain the difference between the two.

Structural Equation Modeling (SEM) is a general term that encompasses a variety of statistical methods designed to assess and quantify unobservable (latent) constructs from observable measures and to examine relationships among these latent constructs. PLS-SEM (Partial Least Squares Structural Equation Modeling) is a specific approach to SEM. To better understand the differences, let's compare PLS-SEM with the more traditional, covariance-based approach to SEM (CB-SEM):

1. Foundational Objective:

• CB-SEM: This method primarily focuses on confirming theoretical models, emphasizing the goodness-of-fit between the observed data and the hypothesized model. The aim is often theory testing.

• PLS-SEM: While PLS-SEM can also be used for theory testing, it's especially appropriate for exploratory research and theory development. Its main objective is maximizing the explained variance of dependent constructs.

2. Distributional Assumptions:

• CB-SEM: It usually requires multivariate normality assumptions for valid results.

• PLS-SEM: PLS-SEM is distribution-free, meaning it does not have strict assumptions about data distributions, making it suitable for non-normally distributed data.

3. Model Specification:

• CB-SEM: Models usually should not have overly complex structures, and constructs are typically modeled as reflective.

• PLS-SEM: This approach can handle more complex models and both reflective and formative measurement models.

4. Sample Size:

• CB-SEM: Typically requires a larger sample size and is sensitive to small sample sizes.

• PLS-SEM: Can work effectively with smaller sample sizes, offering an advantage in studies where obtaining a large sample is challenging.

5. Model Evaluation:

• CB-SEM: Provides global goodness-of-fit indices (like chi-square, RMSEA, CFI) to assess how well the model fits the data.

• PLS-SEM: Does not provide an overall goodness-of-fit statistic. Instead, the focus is on individual path coefficients, R-squared values, and predictive relevance.

6. Parameter Estimation:

• CB-SEM: Uses a maximum likelihood estimation or other similar methods to find parameter estimates.

• PLS-SEM: Uses a component-based approach, leveraging iterative algorithms to find parameter estimates.

7. Flexibility & Complexity:

• CB-SEM: Less flexible in handling complex models with many indicators per construct or higher-order constructs.

• PLS-SEM: More flexible in handling complex models and better suited for situations with many constructs and indicators.

In summary, while both PLS-SEM and CB-SEM fall under the broader umbrella of SEM, they differ in their objectives, assumptions, strengths, and weaknesses. The choice between them should be dictated by the research objectives, nature of the data, and the specific questions being addressed.