A Macchine Learning Approach to Well_Being in Late Childhood and Early Adolescence: The Children's Worlds data case

MÓNICA GONZALEZ CARRASCO; SILVANA ACIAR; FERRAN CASAS; RAMÓN FABREGAT

Social Indicators Research

Springer Netherlands

Año: 2023

0303-8300

Explaining what leads to a higher or lower subjective well-being (SWB) in childhood and adolescence is one of the cornerstones within this field of studies, insofar as it can help to develop more focused preventive and promotion actions. Many indicators of SWB have been identified. However, selecting one over the other to achieve a reasonably short list is not easy since the models are particularlysensitive to the indicators considered.This work aims to explore the possibilities of Machine Learning, compared to linear regression, to 77 indicators included in the 3 rd wave of the Children?s Worlds project. The second objective is to compare models for each of the 35 participating countries with that of the pooled sample. The answers of 93,349 children and adolescents belonging to the 10 and 12-year-olds age groups and collected through arepresentative sampling process, were analysed.The model calculated through Extreme Gradient Boosting outperforms both the Random Forest and the lineal regression model, while the lineal regression model does it better than the Random Forest model. Huge differences at the country level on the importance played by these 77 indicators when explaining the scores to the five-item-version of the CW-SWBS (Children?s Worlds Subjective Well-BeingScale), were found.The process followed highlights the greater capacity of some ML techniques to provide models with higher explanatory power and less error and to differentiate more clearly the contribution of the different indicators to explaining children?s and adolescents? SWB. This can help to design shorter but equally reliable questionnaires.