Abstract / Summary
Abstract National life evaluation and suicide mortality both inform assessments of population well-being, yet their national patterns do not always align. We treat this divergence as country-level statistical discordance, not as evidence of an individual happiness-suicide paradox. The analysis links annual Life Ladder scores, World Health Organization (WHO) age-standardized suicide rates, and World Bank indicators, yielding 2,044 country-year observations for 161 countries during 2005–2021. We combine two-way fixed-effects estimation with a country-held-out additive spline model, eXtreme Gradient Boosting (XGBoost), Isolation Forest, SHapley Additive exPlanations (SHAP), and Gaussian-mixture country profiles. To extend the descriptive record beyond 2021, WHO Mortality Database rates for 2022–2023 are examined separately from the comparable Global Health Estimates series. Within-country increases in unemployment and alcohol consumption were associated with higher suicide mortality and greater residual discordance. In country-held-out tests, XGBoost predicted residual discordance more accurately than the spline GAM (R² = 0.243 versus 0.090); alcohol consumption, urbanization, and income contributed most to its predictions. Lithuania, the Russian Federation, Kazakhstan, and the Republic of Korea had the most persistent high-happiness/high-suicide profiles under the global benchmark. Korea's position depended on the comparison group: in 2021, happiness stood at the 63rd global percentile but the 19th OECD percentile, while suicide mortality stood at the 97th global percentile and was the highest among OECD members in the analytical sample. The results favor joint monitoring of subjective well-being and preventable mortality and caution against conflating predictive explanation, causal inference, and individual risk.