Scholarlink

<p>This study examines key factors associated with subjective well-being (SWB) and their distributional patterns, with particular attention to threshold effects. We conducted a cross-sectional machine learning analysis using data from 10,712 adults in Taiwan drawn from a large-scale survey conducted in 2024. SHAP analysis was applied to interpret model outputs. Five central predictors—family relationships, health, interpersonal relationships, life goal clarity, and financial safety—were consistently associated with SWB. Family relationship satisfaction, interpersonal relationship satisfaction, and health functioned as protective factors and exhibited nonlinear associations with SWB, characterised by pronounced threshold effects: when scores exceeded approximately 5–6 on a 10-point scale, SWB increased sharply. In contrast, financial safety and life goal clarity showed more linear, motivational patterns of association. Among the machine learning models evaluated, Gradient Boosting demonstrated the strongest predictive performance. Overall, changes in SWB were characterised by substantial heterogeneity and the presence of threshold effects. Improvements were not evenly distributed across the population; instead, larger gains were concentrated among individuals with lower baseline SWB. These findings suggest that prioritising SWB-enhancing resources toward populations with lower SWB may yield greater marginal benefits. From a policy perspective, the results indicate that targeted intervention strategies focusing on low-SWB populations may be particularly effective. In practice, governments may develop directed intervention plans that prioritise social relationships, health, and financial safety as core dimensions. Such an approach may both support vulnerable groups and, under conditions of limited resources, contribute to maximising overall population-level SWB.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predictors of subjective well-being in Taiwan: a machine learning analysis with SHAP explanations

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月21日

查看全文
  • Melody Hsiao-San Yeh,
  • Chun-Tung Kuo,
  • Shu-Hui Hsieh,
  • Tzu-Pin Lu,
  • Vincent Cheng-Sheng Li,
  • John Tayu Lee

摘要

This study examines key factors associated with subjective well-being (SWB) and their distributional patterns, with particular attention to threshold effects. We conducted a cross-sectional machine learning analysis using data from 10,712 adults in Taiwan drawn from a large-scale survey conducted in 2024. SHAP analysis was applied to interpret model outputs. Five central predictors—family relationships, health, interpersonal relationships, life goal clarity, and financial safety—were consistently associated with SWB. Family relationship satisfaction, interpersonal relationship satisfaction, and health functioned as protective factors and exhibited nonlinear associations with SWB, characterised by pronounced threshold effects: when scores exceeded approximately 5–6 on a 10-point scale, SWB increased sharply. In contrast, financial safety and life goal clarity showed more linear, motivational patterns of association. Among the machine learning models evaluated, Gradient Boosting demonstrated the strongest predictive performance. Overall, changes in SWB were characterised by substantial heterogeneity and the presence of threshold effects. Improvements were not evenly distributed across the population; instead, larger gains were concentrated among individuals with lower baseline SWB. These findings suggest that prioritising SWB-enhancing resources toward populations with lower SWB may yield greater marginal benefits. From a policy perspective, the results indicate that targeted intervention strategies focusing on low-SWB populations may be particularly effective. In practice, governments may develop directed intervention plans that prioritise social relationships, health, and financial safety as core dimensions. Such an approach may both support vulnerable groups and, under conditions of limited resources, contribute to maximising overall population-level SWB.