What is PLS-SEM?

Partial Least Squares Structural Equation Modeling (PLS-SEM) is a multivariate analysis technique designed to model intricate relationships between observed and latent variables. Unlike covariance-based SEM techniques such as AMOS or LISREL, PLS-SEM is component-based. Its relevance primarily emerges in exploratory research contexts where the primary objective revolves around prediction, especially when the research model is densely populated with constructs and indicators.

This technique finds extensive application across diverse fields, most notably in marketing, information systems, management, and other areas of social science. PLS-SEM stands out as especially beneficial when the data at hand doesn't meet normality assumptions, the sample size is limited, or the model encompasses formative indicators.

Diving deeper into the mechanics of PLS-SEM, it incorporates two central models. The first, the measurement model, relates the observed variables, known as indicators, to their underlying latent variables or constructs. This model plays a pivotal role in assessing the reliability and validity of the measures. The second, the structural model, outlines the relationships interlinking the latent variables, serving to validate the theoretical model or hypothesis.

To effectively evaluate the measurement model, researchers commonly check for indicator reliability, convergent validity (using methods like Average Variance Extracted – AVE and Composite Reliability – CR), and discriminant validity. For the structural model, evaluation criteria typically include R-squared values for endogenous latent variables, effect sizes (f^2), and path coefficients. The significance of these path coefficients can be ascertained through bootstrapping techniques.

Several software solutions are optimized for PLS-SEM analyses. Among the most renowned are SmartPLS, WarpPLS, and the 'plspm' package in R. When weighing the merits of PLS-SEM, one can note its capacity to manage complex models with a multitude of constructs and indicators. Moreover, its applicability to smaller sample sizes and its flexibility with data distribution assumptions are commendable. However, it's essential to understand its limitations as well. Given its predictive orientation, PLS-SEM might not always be the optimal choice for theories that demand confirmation. Moreover, it lacks a global goodness-of-fit measure, a metric available in CB-SEM methods like chi-square.

In recent times, the perception of PLS-SEM has witnessed a shift. While it was predominantly associated with exploratory research, advancements in the field have augmented its potential for theory testing and confirmation. Contemporary research in the area has introduced new criteria and guidelines for PLS-SEM results evaluation, amplifying both its rigor and relevance. However, as with any methodological approach, the key lies in its proper application to extract meaningful and reliable results.

To make it as an overview may be as follows:

PLS-SEM is a multivariate analysis technique used to model complex relationships between observed and latent variables. Here's an overview:

Basics:

• PLS-SEM is a component-based approach, in contrast to covariance-based SEM (like AMOS or LISREL).

• It is particularly suitable for exploratory research where the primary goal is prediction and when the research model has many constructs and indicators.

Applications:

• PLS-SEM is widely used in marketing, information systems, management, and other social sciences.

• It is especially used in cases where the normality assumptions are not met, the sample size is small, or the model has formative indicators.

Components:

• PLS-SEM involves two models: the measurement model and the structural model.

  • Measurement Model: Relates observed variables (indicators) to latent variables (constructs). It assesses the reliability and validity of the measures.

  • Structural Model: Describes the relationships between latent variables. It assesses the hypothesis or the theoretical model.

Evaluation:

• Measurement Model: Assessments typically include checking for indicator reliability, convergent validity (through Average Variance Extracted – AVE, and Composite Reliability – CR), and discriminant validity.

• Structural Model: R-squared values for endogenous latent variables, effect sizes (f^2), and path coefficients are evaluated. The significance of the path coefficients can be assessed using bootstrapping.

Software:

• Popular software for PLS-SEM includes SmartPLS, WarpPLS, and the 'plspm' package in R.

Advantages:

• Can handle complex models with many constructs and indicators.

• Suitable for smaller sample sizes.

• Does not have strict assumptions regarding data distribution.

Limitations:

• Its predictive orientation might not always be suitable for confirming theories.

• Doesn't provide a global goodness-of-fit measure, like chi-square in CB-SEM.

Recent Developments:

• While PLS-SEM was initially considered for exploratory research, recent developments have enhanced its potential for theory testing and confirmation.

• New criteria and guidelines have emerged for evaluating PLS-SEM results, enhancing its rigor and relevance.

When using PLS-SEM, researchers should carefully consider the nature of their data, the objectives of their study, and the advantages and limitations of the method. As with all methods, proper application is key to obtaining meaningful and reliable results.