"Hey @weather, I'm Really Getting Tired of Huddling My Little Girls in the Closet": Using Twitter to Examine Risk Messages, Risk Perceptions, and Responses during Tornadoes

Tornadoes are spatially-localized, rapid-onset hazards that pose tremendous risk of harm to people. Because the predictability of tornado hazards is limited, it is difficult to use surveys or interviews to study what risk information people obtain, their risk perceptions, and their responses during the timeframe when a tornado threatens and leading up to when it occurs. Thus, most risk-related research about tornadoes is conducted by asking people to recall the event retrospectively. Twitter data is a useful complement to traditional research methods and data sources, because it offers a quasi-real-time, longitudinal record of people’s risk assessments. The research presented here uses Twitter data to examine how people’s risk communication behaviors, risk perceptions, and responses evolve over a series of tornado threats in southern Georgia. The analysis begins with two deadly tornadoes that occurred in January 2017, and then it looks at three subsequent tornado threats that affected the same geographic area between February and May 2017. People on Twitter who were at-risk of these tornadoes were identified by using tornado-based and place-based keywords. Once the sample of Twitterers was determined, people’s full Twitter narratives from January through May were collected. A quantitative content analysis of these data was conducted to code for mentions of different types of risk information (e.g., weather forecasts, recommended preparedness actions, social cues), risk perceptions, and responses. This presentation will illustrate the ways that people attend to, interpret, and use tornado risk information across multiple threats that evolve in space and time. The implications of this knowledge for improving tornado forecast and warning communication and response also will be discussed.