South African translations of the Grit-S scale: factor structure, measurement invariance, and reliability across language groups

The short grit scale (Grit-S) and its numerous translations have been, and continue to be, ubiquitously cited as valid and reliable measures of psychological grit – a personality trait defined as passion and perseverance for long-term goals. However, despite this growing body of evidence attesting to the psychometric soundness of (translated versions of) the Grit-S scale, questions remain as to how grit manifests among cultures in the Global South, particularly in South Africa, where linguistic and cultural diversity is commonplace. Aiming to narrow this gap, this study examined the factor structure, measurement invariance, and reliability of English-Afrikaans, Nguni-Tsonga, and Sotho-Makua-Venda translations of the Grit-S scale among first-year university students ( n = 959). Confirmatory factor analysis and multigroup confirmatory factor analysis (configural, metric, and scalarinvariance) were conducted. Results indicated that the a priori two-factor model exhibited acceptable fit among the pooled sample (CFI = 0.97, RMSEA = 0.047, SRMR = 0.034) despite Q2’s (“ Setbacks don’t discourage me ”) unfavourable loading ( λ = 0.15). Configural and metric invariance were supported; however scalar invariance could not be established (ΔCFI = − 0.05). In terms of the scales’ reliability, Cronbach’s alpha and McDonald’s omega coefficients suggested that the two domains were not reliable, with the exception of the passion domain in the English-Afrikaans version, which demonstrated acceptable levels of internal consistency (> 0.70). The removal of Q2, however, improved the reliability of the perseverance domain in both the English-Afrikaans and Sotho-Makua-Venda translations ( ω = 0.70 − 0.73). Collectively, these results suggest that South African translations measure the same underlying same construct (i.e., psychological grit), however, latent variable means across the language groups are not comparable. Resultantly, grit levels should only be compared with latent variable models that allow for partial invariance and Q2 should be revised or removed in future iterations.