The Impact of Color-Coded Probabilistic Tornado Warnings on Risk Perceptions and Responses. Part I: Experiment

A tornado warning means that a tornado threat is imminent or occurring. Currently, however, tornado warnings issued by the U.S. National Weather Service (NWS) are binary, one is either in or out of the warning area, implying an even distribution of risk inside the warning. However, not every location in a warning shares an equivalent chance of being hit by a tornado. In order to convey more nuanced chances of exposure, the NWS is considering issuing probabilistic tornado warnings in the future. The rapid-onset nature of tornadoes leaves little room for confusion about these risk messages. Yet, with few exceptions (Ash et al., 2014; Klockow 2013; Ling et al., 2015) there has been little research to investigate the public’s ability to accurately interpret and use probabilistic warnings. The work presented here addresses these critical knowledge gaps through a mixed-methods research approach with an initial experimental study followed by interviews. Both methods evaluated different tornado warning visuals, including a deterministic polygon warning (as is currently issued) and multiple probabilistic warning formats, as well as different locations in each visual.

Both methods elicited people’s perceived likelihood of a tornado hit, perceived severity if a tornado hits, worry, protective response, and trust in the information.

In the experiment, four probabilistic formats, two with color and two with numeric probabilities alone, were compared to the conventional polygon format. In a computerized decision task, participants (M-turkers) decided whether or not to take shelter based on a tornado warning. Probabilistic formats led to better understanding of the underlying likelihood of a tornado and higher trust than did the conventional polygon. However, formats that included color led to over-estimation of the likelihood and users tended to conflate likelihood and severity compared to formats with numeric probabilities alone. Although there are some benefits to color-coded likelihood displays, they can give rise to some misunderstandings.