Developing and validating the Second Language Buoyancy Scale (L2BS): Evidence from psychometric and machine learning analyses

Given that existing academic buoyancy measures do not capture learners’ everyday capacity to cope with setbacks in the L2 learning, an L2-specific scale is needed to assess second language (L2) buoyancy. This study aimed to develop and validate the Second Language Buoyancy Scale (L2BS). Using convenience sampling, data were collected from 554 university students at two mainland Chinese institutions and randomly split into two equal subsets (n = 277 per subset). Content validity was established via qualitative item generation (17 interviews) and expert review (ICC = .83). For structural validity, EFA on Subset 1 (KMO = .826; Bartlett’s χ²(6) = 616.99, p < .001) supported a single-factor, four-item solution with loadings > .65; CFA on Subset 2 showed good fit (RMSEA = .096, CFI = .992, TLI = .977). Internal consistency was strong (Cronbach’s α = .898; McDonald’s ω = .898). Construct validity was supported by AVE = .689 and small-to-moderate correlations with academic buoyancy, growth mindset, grit, and conscientiousness. Criterion-related validity was evidenced by hierarchical regressions (incremental variance: ΔR² = .173 for L2 engagement; ΔR² = .160 for L2 enjoyment) and machine-learning models (Random Forest/XGBoost/LightGBM), in which L2BS consistently outperformed academic buoyancy (best accuracies: 73.21% for engagement; 64.29% for enjoyment). Overall, L2BS provides a brief, reliable, and valid measure of L2 buoyancy with clear utility for predicting key L2 outcomes such as L2 engagement and L2 enjoyment.