Testing Hypotheses about Language Learning Using Structural Equation Modeling

Abstract

In this chapter, I review a statistical method for hypothesis or theory testing called structural equation modeling (SEM). First, I describe what a model of second language acquisition (SLA) is. I do this so anyone, even those new to the field of applied linguistics, can understand the basic concepts underlying SEM; that is, SEM researchers first articulate a model of SLA, then get empirical data from the real world that operationalize the variables in the model. Researchers use an SEM program to test the model on the data (to see if the model fits the data; if the model is plausible in relation to the learning context of the people from whom the data were collected). After explaining the basics of SEM, I provide a review of 39 applied linguistics studies that have been published in the last five years (between 2008 and 2013) and that present at least one SEM analysis as part of the results. I discuss four problematic areas related to the use of SEM that I believe these 39 studies highlighted: (a) sample size, (b) model presentation, (c) reliability, and (d) the number of Likert-scale points. I conclude with possible solutions for the four problem areas and outline future directions.

The Relationship Between Vocabulary Learning Strategies and Breadth and Depth of Vocabulary Knowledge

“…SEM is a family of relatively complex statistical methods for exploring or testing relationships among variables (Byrne, ). It is beyond the scope of the current study to provide a comprehensive introduction to SEM; interested readers should consult Kline () for a detailed discussion of the model and Winke () for a review of recent SLA studies that employed SEM. We chose to use SEM instead of multiple regression because it takes into account “the measurement errors of the independent variables and dependent variables” (Raykov & Marcoulides, , p. 7), while the multiple regression method ignores measurement errors of all independent variables.…”

Best Practice Recommendations for Using Structural Equation Modelling in Psychological Research

“…Manifest variables are those that can be directly observed (measured); alternatively, latent variables are those that cannot be observed (measured) directly due to their abstract nature (Byrne, 2016;Xiong, Skitmore, & Xia, 2015 (Byrne, 2016;Winke, 2014).…”

The development of statistical literacy in applied linguistics graduate students

“…In recent years, there has been an increased concern regarding the quality of statistical knowledge and methodology in the field. For instance, several applied linguists (e.g., Lindstromberg, 2016;Norris, 2015;Norris, Plonsky, Ross & Schoonen, 2015;Plonsky, 2011Plonsky, , 2013Plonsky, , 2015aPlonsky & Gass, 2011;Plonsky & Gonulal, 2015;Plonsky & Oswald, 2016;Winke, 2014) have attempted to describe the current state of methodological practices as observed in published L2 research, and have called for a better statistical rigor, if not statistical sophistication. However, to date no research has investigated how applied linguists acquire their statistical knowledge.…”

“…In the last two decades, the field of SLA has also become "more sophisticated in its use of statistics" (Gass, 2009, p. 19). Indeed, novel and more sophisticated statistical techniques (e.g., factor analysis, structural equation modeling, mixed effects models, bootstrapping) are slowly finding their ways into L2 research (Cunnings, 2012;Larson-Hall & Herrington, 2010;Loewen & Gonulal, 2015;Plonsky, Egbert & LaFlair, 2014;Plonsky & Gass, 2011;Winke, 2014).…”