A comparison of frequentist and bayesian structural equationmodels in examining the relationships between social andemotional skills

The development of social and emotional skills is necessary for students to adapt to rapidly changing world conditions, cope with increasingly uncertain life conditions, and sustain their professional and academic success. Moreover, it is considered very important to examine social and emotional development, which are important dimensions of student development, and reveal the relationship between these skills. Structural equation modeling methods are widely applied in researches due to their advantage of being able to test and explain the relationship between variables more holistically. In structural equation modeling applications, one can employ the Bayesian methods alongside the frequentist approaches, and there may exist situations where these methods exhibit relative advantages or disadvantages. This study compared frequentist and Bayesian structural equation modeling methods based on the statistics they produced in the analysis for the relationships of social and emotional skills. The comparison was conducted using data from the OECD social and emotional skills survey from Istanbul and Moscow. The Istanbul sample consisted of 3167 participants and the Moscow sample consisted of 3429 participants. Both frequentist and Bayesian CFA were applied to the dimensions (5 dimensions: engaging with others, open-mindedness, task performance, emotional regulation, collaboration) and sub-dimensions (15 sub-dimensions) of social and emotional skills in the Big Five Model, and the correlations, fit indices and standardized factor loadings generated by the two methods were compared. In addition, the relationships between social and emotional skills separately and holistically in the Big Five Model were examined with frequentist and Bayesian SEM methods and the results were compared with each other. According to the findings, BCFI and BTLI fit indices used in the Bayesian method were found to be more compatible than their frequentist counterparts, CFI and TLI. On the other hand, it was observed that RMSEA used in the frequentist method gave more compatible results than its Bayesian counterpart, BRMSEA. It was concluded that BRMSEA, BCFI and BTLI fit indices can be reported when it is desired to use Bayesian SEM, which provides advantages in cases where missing data is high and the sample is small, the data does not have a normal distribution and a priori information is desired to be used. In addition, it was found that the standardized factor loadings and correlation values obtained from the models established with both methods were generally different from each other, and this difference increased as the model grew larger and more complex. Finally, it was found that relationship between social and emotional skills were generally at positive moderate or positive high level.