Evaluating the dimensionality of L2 grit: A confirmatory bifactor model

Abstract

This study applied bifactor models to evaluate the dimensionality of two L2 grit models: the 2-factor model (Teimouri et al., 2022) and the 3-factor model we proposed based on Datu et al. (2017). We tested the bifactor models in conjunction with several feasible models including the unidimensional models, the correlated-factors models, and the second-order factor models. Our objective was to investigate whether the grit models effectively measure a unidimensional general construct while also assessing the viability of the domain-specific subconstructs. We utilized a range of bifactor statistical indices to evaluate the unidimensionality of the grit instrument, including omega coefficients (ω), omega subscale coefficients (ωs), omega hierarchical coefficients (ωH), omega hierarchical subscale coefficients (ωHs), explained common variance (ECV), Percent of Uncontaminated Correlations (PUC), item ECV (I-ECV), and construct replicability H-index. The major findings include: (1) the general L2 grit construct comprising 3 domain-specific subconstructs can be treated as essentially unidimensional; (2) the sum scores of the 3-factor grit scale can reliably measure a general L2 grit construct; (3) an alternative version consists of 8 items fits a unidimensional model well. These findings offer valuable implications regarding how to effectively use the sum scores and which items best represent a unidimensional construct.