Income, psychological security, and subjective well-being in urban China: a machine learning analysis with SHAP...

Area of Science:

• Social Sciences
• Psychology
• Economics
• Data Science

Background:

• Subjective well-being is a key metric for social progress, yet the Easterlin Paradox highlights a disconnect between economic growth and well-being.
• Traditional economic indicators like income have diminishing marginal effects on well-being.
• The multidimensional construct of psychological security and its influence on well-being remain underexplored, particularly its nonlinear relationships.

Purpose of the Study:

• To investigate the complex, nonlinear relationship between psychological security, income, and subjective well-being using advanced machine learning techniques.
• To identify the primary drivers of subjective well-being and elucidate the mechanisms underlying its formation across different socioeconomic groups.
• To overcome the limitations of traditional linear models in capturing intricate variable interactions influencing well-being.

Main Methods:

• Employed machine learning algorithms (LightGBM, XGBoost, Random Forest) on a nationwide survey of 1,369 Chinese urban residents.
• Utilized SHapley additive exPlanations (SHAP) for interpretability to analyze variable importance and impact.
• Implemented data preprocessing techniques including imputation, standardization, and SMOTE-Tomek sampling, with model optimization via cross-validation and grid search.

Main Results:

• The LightGBM model achieved the highest performance (AUC=0.854), outperforming traditional linear models.
• Psychological security emerged as the most crucial factor influencing subjective well-being across all income levels.
• Income demonstrated a threshold effect, with middle-income levels showing the strongest positive impact, significantly amplified by psychological security.

Conclusions:

• Psychological security is the most stable and core determinant of subjective well-being, with significant synergistic effects with income.
• Machine learning and SHAP analysis effectively capture complex nonlinearities and provide interpretable insights into well-being mechanisms.
• Findings support differentiated policy interventions, emphasizing psychological security and resource-matching strategies for diverse well-being groups.