Social networks and depressive tendency: A longitudinal study from adolescence through adulthood

Abstract

This study aimed to predict adult depressive tendencies based on adolescent developmental trajectories and social networks using longitudinal data from the Taiwan Youth Project (TYP). Analyzing data from 1047 individuals across four waves (2000-2002 and 2017), we examined personal, family, and peer factors influencing depressive tendencies from adolescence to adulthood (age ∼30). Key findings indicate that high self-esteem, family cohesion, and peer support during adolescence are significant protective factors against adult depressive tendencies, while adverse family structures (e.g., divorced but cohabiting parents) increase risk. Graph Neural Network (GNN) models, particularly GraphSAGE (F1-score = 0.3678), outperformed traditional machine learning methods (e.g., logistic regression, F1-score = 0.3509) by leveraging social network structures to enhance prediction accuracy, although the absolute F1-scores remain modest (∼0.37), reflecting the inherent challenges in predicting long-term depressive tendencies from psychosocial data. These results underscore the long-term impact of adolescent psychosocial factors and the superior capability of GNNs in modeling complex social interactions. Findings inform early intervention policies, advocating for school-based programs to foster self-esteem and social connections, and the integration of GNN-driven tools to identify at-risk individuals for targeted mental health support in Taiwan and similar cultural contexts.

Keywords: Adolescent's developmental trajectory; Depressive tendency; Graph neural network; Machine learning; Social network.