Risk-Prone and Risk-Averse Behavior in Natural Emergencies: An Appraisal Theory Approach

Abstract

Individuals who shared actionable information during Hurricane Sandy were significantly more likely to exhibit risk-prone behavior, as measured by a novel Risk Behavior Quotient (RBQ). Using a dataset of 36595 geo-located tweets from 774 users in the New York area, we found that a higher proportion of actional tweets predicted increased exposure to physical even if overall users ultimately moved toward lower-risk zones. This counterintuitive finding suggests that proactivity, manifested in sharing crisis relevant content, correlates with greater exposure to risk, possibly due to increased mobility or engagement in hazardous areas. In contrast, a greater number of social media peers was associated with reduced risk exposure. This study builds on appraisal theory, which frames risk-related decisions as outcomes of cognitively mediated emotional and rational evaluations. We extend this theory to digital crisis behavior, distinguishing between emotional and actional appraisals expressed via social media. Tweets were categorized using sentiment analysis and semantic classification, enabling the isolation of affective and behavioral signals. Our methodology combines natural language processing with spatial vector analysis to estimate individual movement paths and risk exposure based on evacuation and flooding maps. The resulting RBQ captures both direction and intensity of risk behavior, allowing us to model how online communication reflects and predicts real-world risk engagement during natural disasters.