Understanding the Psychological, Relational, Sociocultural, and Demographic Predictors of Loneliness Using Explainable Machine Learning

Abstract

Loneliness—an important indicator of social health—is increasingly recognized to derive from factors operating at multiple levels. However, simultaneously examining the role of factors at multiple levels implies using large samples and testing multiple factors at the same time, which traditional statistical methods cannot accommodate. We used machine learning techniques to address this problem. We identify the most important out of 32 correlates of loneliness frequency in a large sample of people ages 16+ years, residing all over the world, who took part in the British Broadcasting Corporation Loneliness Experiment. Factors spanned individual, relational, sociocultural, and demographical areas. The most statistically important associate of loneliness was daily experiences with prejudice (or stigma), followed by couple satisfaction, neuroticism (emotional stability), personal self-esteem, average hours spent alone daily, extraversion, social capital, and relational mobility. Interaction effects were also evident, showing that experiences with prejudice were most negatively associated with loneliness when individuals spent a lot of time alone and the least when individuals were emotionally stable, had high personal self-esteem, or had high levels of couple satisfaction. This research highlights what factors need to be considered when developing effective interventions to mitigate loneliness.