Observing the Unobservable: Connecting Persistent Emotional State to Behavior in Risky Contexts

Abstract

Economists rely on experiments and surveys to measure typically unobserved subjective attributes, e.g., preference, belief, and emotion, that are important for understanding human behaviors. Dependence on these data collection modes limits the ability to connect subjective traits with behavior in a large-scale natural setting. Here, we use posts on social media platforms and natural language processing to infer individuals’ persistent emotional state, and link this attribute to the behavior of 500,000 individuals during the COVID−19 pandemic. We find that fearfulness is a persistent trait associated with specific individuals overtime rather than associated solely with a single experience and that these high-fear individuals reduce out−of−home activities 4.2-6.9 percent more than low-fear individuals. The ability to observe subjective attributes at scale can enable better identified economic measures and better targeted economic policy.