|
Abstract:
|
Investigating the effect of the alarm type on people’s level of fear and protective decision-making during a tornado event can result in a more effective warning system. There is an ongoing debate, however, on the appropriate method of analysis for the resulting ordinal and Likert type data. Collecting data about people’s responses to hypothetical tornado displays using the Probabilistic Hazard Information(PHI) tool, we study the effect of providing some uncertainty information associated with the threat-occurrence on people’s level of fear and chances of taking protective action. We use a Bayesian hierarchical multivariate model with non-informative priors to analyze this ordinal data. It is demonstrated that the common practice of treating ordinal data as metric can lead to false conclusions, and that the Bayesian approach is more powerful than nonparametric tests for handling this kind of ordinal data. Our method finds that providing uncertainty information about a tornado occurrence through PHI significantly increases people’s level of fear and chances of taking protective action,compared to the deterministic ‘WarnGen’ tool currently used by the National Weather Service (USA)
|
