Exploratory Structural Equation Modeling in Second Language Research

Abstract

This study offers methodological synergy in the examination of factorial structure in second language (L2) research. It illustrates the effectiveness and flexibility of the recently developed exploratory structural equation modeling (ESEM) method, which integrates the advantages of exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) into one complete measurement model. Two sets of data were collected using the L2 Passion Scale, which measures a dualistic model of passion. Study 1 participants were 220 L2 students. A comparison was made between the CFA and the ESEM models. The results demonstrated the superiority of the ESEM method relative to CFA in terms of better goodness-of-fit indices and realistic correlated factors. These results were replicated in another sample of 272 L2 students, providing support for the predictive validity using a structural ESEM model. Guidelines are provided and Mplus syntax files (codes) are included to help analysts apply the methods. We also make the data available publicly. Overall, this research demonstrated the usefulness of ESEM for examining the construct, discriminant, and convergent validity of L2 scales over CFA.

How Accurate Is Your Correlation? Different Methods Derive Different Results and Different Interpretations

“…Alternatively, research shows that certain levels of true scores can be relevant for other conceptually related constructs ( Guay et al, 2015 ; Morin et al, 2016 , 2020 ). This observation was confirmed by Alamer and Marsh (2022) as they found CFA fit indices were not acceptable for the L2-Passion scale. The researchers noted that CFA was rather a restrictive structure in that it ignores cross-loadings among the items of two conceptually related constructs, which, in turn, resulted in inflated factor correlation.…”

“…Beyond bivariate correlation, researchers have started to endorse second-generation methods ( Hair et al, 2022 ) that are built on the property of structural equation modeling (SEM) to assess the associations between variables. Among these methods are confirmatory factor analysis (CFA) and exploratory structural equation modeling (ESEM) [see Alamer and Marsh (2022) and Alamer (2022b) for details and applied examples about ESEM]. CFA is a method that is used to understand the underlying factor structure of the constructs ( Marsh et al, 2009 ; Morin et al, 2016 ).…”

“…Early literature in EFA and CFA ( Jöreskog, 1973 ) made the assumption that factors should be anchored in distinctive clusters of observed variables to constitute the latent variable. Nevertheless, this restrictive system of the measurement model has been challenged in the last decade ( Marsh et al, 2009 ; Guay et al, 2015 ; Morin et al, 2016 , 2020 ; Alamer, 2021a , 2022b ; Alamer and Marsh, 2022 ). This is because conceptual constructs have certain levels of similarities; they can overlap especially when they are conceptually related.…”

“…This is because conceptual constructs have certain levels of similarities; they can overlap especially when they are conceptually related. Consider, for example, the measurement model reported in Alamer and Marsh (2022) study where two constructs, harmonious passion and obsessive passion , were involved in the analysis. To provide context to the example, harmonious passion reflects the strong desire to freely engage in language activity whereas obsessive passion reflects the controlled pressure combined with an uncontrollable urge to partake in the language activity.…”

“…To provide context to the example, harmonious passion reflects the strong desire to freely engage in language activity whereas obsessive passion reflects the controlled pressure combined with an uncontrollable urge to partake in the language activity. In the L2-Passion scale ( Alamer and Marsh, 2022 ), an item in harmonious passion reads “the new things that I discover in English allow me to appreciate it even more” while an item in obsessive passion reads “learning English is the only thing that really turns me on.” One can see how these two items belong to two different types of passion, but also each item seems to present significant true scores on the other (untargeted) type of passion. If the item on harmonious passion has no role at all to play in contributing to the meaning of the other factor, then why factor correlation is relatively high?…”