A Comparative Analysis of Frequentist and Bayesian SEM Approaches in Modeling Social and Emotional Skills

The exploration of the development of social and emotional skills, as well as their interrelations, remains a significant emphasis in educational and psychological research. Structural equation modeling (SEM) is widely used for this purpose because it provides a comprehensive framework for examining complex relationships. Bayesian SEM has gained growing attention, offering unique strengths as well as certain limitations when compared to traditional frequentist approaches. In this study, we used data from the OECD's Social and Emotional Skills Survey (SESS) to compare results produced by frequentist and Bayesian SEM. Analyses focused on the five broad domains represented in the Big Five Model. Our findings revealed meaningful differences. Bayesian fit indices such as BCFI and BTLI demonstrated greater stability than their frequentist counterparts (CFI and TLI), while RMSEA showed stronger consistency within the frequentist approach compared to its Bayesian version (BRMSEA). These patterns suggest that BCFI, BTLI, and BRMSEA may be particularly informative and reliable in Bayesian SEM applications, especially when prior information is incorporated into the model. Additionally, we observed that discrepancies in standardized factor loadings and correlations became more evident as model complexity increased. Overall, the relationships among social and emotional skills were positive and ranged from moderate to strong.