Exploring Protective Factors Against Unhappiness Under the Global Crisis: A Machine Learning Approach

While a vast literature has explained the determinants of happiness, the Covid-19 pandemic has drastically changed the fundamental conditions for human flourishing. Although people have shown their resilience over the few years, evidence indicates that the unhappiness risk surged during the pandemic and has remained high ever since. This paper therefore explores protective factors against unhappiness under the global crisis using data from the General Social Survey (2018-2021, N=4,927). A series of supervised machine learning models with the binary outcome of feeling “not too happy” (against “very happy” and “pretty happy” in the three-point scale) consistently show that satisfactory social relationships/activities are key to preventing unhappiness both in 2018 and 2021. Importantly, its magnitude in terms of the SHapley Additive exPlanations (SHAP) value increased from 0.438 to 0.560 under the pandemic, whereas the contribution of income, occupational prestige, and physical health, among others, declined during the same period (0.277->0.107; 0.167->0.143; 0.184->0.130, respectively). Being married mitigates the unhappiness risk as the second most important factor in both years, although its SHAP value decreased from 0.436 to 0.322. In contrast, albeit modestly, the impact of attending religious services intensified from 0.066 to 0.098. These results suggest, under the crisis like the Covid-19 pandemic, connections with other people operate as an essential buffer against unhappiness. Examining their future trends, including effect heterogeneity across individuals and societies, is a crucial agenda to realize human flourishing.