Vol. 8:2 (2022) ► pp.237–260
On learner characteristics and why we should model them as latent variables
Learner corpus research has a strong tradition of collecting metadata. However, while we tend to collect rich descriptive information about learners on directly measurable variables such as age, year of study, and time spent abroad, we frequently do not know much about learner characteristics that cannot be measured directly (and that thus need to be measured through questionnaires and tests) such as language aptitude, working memory, and motivation, which have been identified as important variables in neighboring fields such as Second Language Acquisition. In this position piece, we (i) join the proponents of increased focus on learner characteristics in LCR in arguing in favor of collecting information about such variables and (ii) introduce an analytical framework that can be used to model these variables. Specifically, the primary focus of this paper is to discuss the concept of latent variables as it relates to LCR and show how their standard form can be used to model learner characteristics within the structural equation modeling analytical framework.
Article outline
 1.Introduction
 2.Latent variables: A conceptual overview
 3.Structural equation modeling: A framework for modeling learner characteristics as latent variables
 4.Modeling latent variables
 5.A worked example
 5.1Latent variable path analysis
 5.2Latent variable path analysis vs. multiple regression
 6.Conclusion
 Notes

References
https://doi.org/10.1075/ijlcr.21007.lar