Statistics

Chromatography

Applied mathematics

Latent variable

Computer science

Structural equation modeling

Covariance

Natural language processing

Least-squares function approximation

Psychology

Mathematics

Estimator

Formative assessment

Partial least squares regression

Econometrics

Artificial intelligence

Chemistry

Sample (material)

Machine learning

Partial least squares structural equation modeling (PLS-SEM) is an alternative method to the historically more commonly used covariance-based SEM (CB-SEM) when analyzing the data using structural equation modeling (SEM). The article starts by introducing PLS-SEM to second language and education research, followed by a discussion of situations in which PLS-SEM should be the method of choice for structural equation modeling. It is argued that PLS-SEM is appropriate when complex models are analyzed, when prediction is the focus of the research – particularly out-of-sample prediction to support external validity, when data do not meet normal distribution assumptions, when formative constructs are included, and when higher-order constructs facilitate better understanding of theoretical models. The most up-to-date guidelines for applying PLS-SEM are provided, and step-by-step guidance is offered on how to apply the method using an R statistical package (i.e., SEMinR) that is available. An example is provided that shows how the results of PLS-SEM are interpreted and reported. We also make the data publicly available for readers to start learning PLS-SEM by replicating our findings. The paper concludes with important considerations for the utilization of SEM, especially PLS-SEM, in future L2 research.

Partial Least Squares Structural Equation Modeling (PLS-SEM) in second language and education research: Guidelines using an applied example