Which social contexts best predict adolescents’ social-emotional skills: Combining machine learning and traditional statistics

Social-emotional skills have been demonstrated to be crucial for adolescents’ success in the 21 st century. However, past studies have primarily focused on the consequences of these skills, with little attention devoted to understanding the factors that could facilitate these skills. This gap prevents educators from targeting the most important factors that could lead to improvements in adolescents’ social-emotional skills. To address this gap, this study aimed to examine the role of distinct family, school, peer, and community factors in predicting adolescents’ social-emotional skills. We utilized data from the Organization for Economic Co-operation and Development’s Survey on Social and Emotional Skills, including 29,798 15-year-olds from nine countries. An integrative modeling approach that combined machine learning (i.e., Extra Gradient Boost) and traditional statistics (i.e., hierarchical linear modeling) was used to rank the relative importance of factors in predicting adolescents’ social-emotional skills. Results revealed that school belonging was, by far, the most important predictor of all the different types of social-emotional skills. Other important predictors included social connectedness, school cooperation, and parents’ social-emotional skills. This study not only highlights the complexity of factors that might be associated with adolescents’ social-emotional skills but also provides empirical evidence to help educators identify the most effective targets for educational interventions.