International Choice Modelling Conference, International Choice Modelling Conference 2015

Font Size: 
Response to Hurricane Track Forecast Cones: an examination of map attributes that trigger risk perceptions and evacuation intentions
Ricardo A Daziano, Yutaka Motoaki, Clifford Scherer, Jonathon Schuldt, Gina Eosco, Laura Rickard

Last modified: 18 May 2015


Between October 28th and November 29th, 2012, Hurricane Sandy in New York, New Jersey, and nearby areas caused 117 fatalities. A number of these fatalities could have been prevented if residents had evacuated when mandated to; 45% of drowning deaths occurred in Evacuation Zone A, which had been identified as being at risk of flooding from any category of hurricane. This fact illustrates the key motivation behind studying evacuation behavior.

Existing research in the field of evacuation behavior has examined what different factors influence evacuation decisions and how they do so. Five of the most important factors include characteristics of the storm, risk perception, housing type, authorities’ actions, and the hazard level of the area. Socio-demographic factors related to evacuation behavior are gender, age, household size, income, race and ethnicity, and level of education. Physical disability, proximity to evacuation routes, previous experience with extreme weather events, the presence of pets, and media reports also affect the decision of whether or not to evacuate in the case of an extreme weather event. In the existing literature, however, little attention has been devoted to the problem of how storm information visualization impacts evacuation actions.  Some exceptions make use of audio-visual representations of the evolution of a storm for collecting evacuation intended actions. Nonetheless, previous work on understanding response to existing NOAA products to convey and visualize risks usually has been mostly qualitative. 

In this paper we analyze data collected in July of 2014 among 150 individuals living in the Metropolitan Area of New York City, using a unique instrument. In the 25-minute survey, cognitive, emotional, and behavioral responses to information about extreme weather events was collected. In fact, the instrument was designed to cover a thorough set of dimensions that we identified as critical to evacuation decisions, including pre-awareness, experience, behavioral influence, risk perceptions, affect, evacuation intentions, preventive actions, channel beliefs, source credibility, attribution of responsibility, and social networks. In particular, information about a hypothetical storm was conveyed using NOAA Track Forecast Cones. 3-day Track Forecast Cones represent the probable track of the center of a storm. Information on the maps also includes time to landfall and maximum sustained winds. Respondents were asked a set of questions for four maps randomly chosen from Track Forecast Cones for 84, 72, 57, 48, 36, 30, and 6 hours to landfall. We note that 48 hours is a standard threshold for beginning evacuation; 36 hours is considered optimal clearance time; 30 hours coincides with the first evacuation order for Sandy; and 6 hours is considered the last safe possible evacuation time. For each map respondents were asked to rate on a Likert scale the perceived harm and threat that the depicted forecast could cause to him or her, his or her family, and his or her local community. To take into account social norms, respondents were also asked about their beliefs regarding their family or friends wanting them to evacuate to safer area given the information provided in each map. Perceived capability of evacuating was also asked. Finally, given the information on each map we asked the respondents about their concrete plans to evacuate to a safer area. 

To explain the variability observed in the collected responses, we estimated ordered logit models for each question, but consolidating the different maps as choice situations. For example, for the question regarding making concrete plans to evacuate to a safer area, we considered the ordinal responses of the four maps seen by each respondent. We then constructed different attributes that summarized the information conveyed in each map and estimated the models to test parameter significance. The following map attributes resulted statistically significant in explaining propensity to make concrete evacuation plans: linear distance from the current position of the storm to New York City; indicator variable that is equal to 1 when New York City is within the bounds of the cone; maximum sustained winds; and time to landfall. Gender and whether the respondent actually evacuated for hurricane Sandy also appeared as significant variables. In terms of the point estimates, for example the odds ratio of the NYC indicator is 3.23, which means that if the cone covers the city of New York, respondents have a 223% higher likelihood of answering a higher propensity to make concrete evacuation plans. The odds ratio of Sandy actual evacuation experience is 2.85. Each additional mph in the maximum winds makes answering a higher propensity to make concrete evacuation plans 7.31% more likely. We also built 95% confidence intervals of the odds ratios. 

In terms of methodology, we analyzed not only ordered logit models with fixed parameters, but also with random parameters – using the maximum simulated likelihood estimator – to account for both unobserved heterogeneity and correlation among respondents (as explained above, each respondent answered to four maps, which were considered as choice situations). In addition, we analyzed estimates of an ordered logit model with endogenous, latent explanatory variables (measuring uncertainty, preparedness, and potential loss), in the form of a simultaneous system of ordered logit models (cf. hybrid choice modeling).

Conference registration is required in order to view papers.