Structural equation modeling (SEM) in L2 writing research: Simple tutorial and useful recommendations

Abstract

Research in second language (L2) writing has witnessed a surge in the endorsement of structural equation modeling (SEM) applications. This tutorial paper highlights the advantages of using SEM in the field through a showcase of basic as well as advanced SEM analyses. I begin by illustrating how SEM can reproduce basic analyses (i.e., first-generation methods) like correlation and t-test. More importantly, I show how SEM enhances these analyses by effectively handling missing data and deal with non-normality which leads to more valid and unbiased findings. Beyond enhancing basic analyses, SEM is typically used for advanced analyses such as mediation and moderation. Nonetheless, particular emphasis in this paper will be on justifiying the disticntion between two types of constrcuts: (1) latent variables (reflective/common factors) like ‘L2 intrinsic motivation’ where items are interchangeable and similar in meaning, and (2) emergent variables (informative/composites) like ‘L2 writing achievement’ that is formed by distinct, but relevant, elements such as spelling, writing sample, and sentence fluency. After that, I highlight new features of SEM that analysts should be aware of. Also, concise guidelines and recommendations for using and reporting SEM, such as sample size, model estimators, fit indices, and effect sizes of the paths are provided. To enhance the practicality of this article, a step-by-step tutorial using the free software Jamovi, along with a simulated dataset uploaded online, is presented to enable readers to gain hands-on experience and replicate the analyses. Given the increasing accessibility of user-friendly SEM applications, researchers should adopt this powerful methodology and follow the updated guidelines.