Predictors of compliance with tornado warnings ... - Wiley Online Library

4 downloads 9382 Views 485KB Size Report
of Business, Science, and Nursing, Cloud County Community College, United ... Keywords: emergency warning compliance, Joplin tornado, tornado warnings.
doi:10.1111/disa.12087

Predictors of compliance with tornado warnings issued in Joplin, Missouri, in 2011 Bimal Kanti Paul Professor, Department of Geography, Kansas State University, United States, Mitchel Stimers Instructor of Geography and Geoscience, Division of Business, Science, and Nursing, Cloud County Community College, United States, and Marcellus Caldas Associate Professor, Department of Geography, Kansas State University, United States

Joplin, a city in the southwest corner of Missouri, United States, suffered an EF-5 tornado in the late afternoon of 22 May 2011. This event, which claimed the lives of 162 people, represents the deadliest single tornado to strike the US since modern record-keeping began in 1950. This study examines the factors associated with responses to tornado warnings. Based on a posttornado survey of survivors in Joplin, it reveals that tornado warnings were adequate and timely. Multivariate logistic regression identified four statistically significant determinants of compliance with tornado warnings: number of warning sources, whether respondents were at home when the tornado struck, past tornado experience, and gender. The findings suggest several recommendations, the implementation of which will further improve responses to tornado warnings. Keywords: emergency warning compliance, Joplin tornado, tornado warnings

Introduction Joplin, a city in the southwest corner of Missouri, United States, was struck by an EF-5 multiple-vortex tornado in the late afternoon of 22 May 2011. Extending to nearly a mile in width after touching down in the southwest corner of the city, the tornado generated at least a six-mile-long path across this densely populated community of 50,150 (see Figure 1). It destroyed more than 7,000 homes and damaged hundreds more (The Joplin Globe, 2011a). Virtually every house in the area near McClelland Boulevard and 26 th Street was flattened, with some swept completely off their foundations, and many trees sustained severe debarking. As the tornado tracked eastward, it crossed Main Street between 20 th and 26 th Street, and Range Line Road between (approximately) 13th and 32 nd Streets. Every business along that stretch of the Range Line Road suffered either heavy damage or was completely destroyed (see Figure 1).   The damage spanned some 1,800 acres, nearly one-fourth of the territory of Joplin. Some of the severely affected commercial buildings were a Walmart Supercenter, a Home Depot store, and myriad restaurants. One of the community’s two hospitals, St. John’s Regional Medical Center, experienced heavy damage. Several buildings belonging to the other hospital in Joplin, Freeman Hospital West, also sustained Disasters, 2014, 39(1): 108−124. © 2014 The Author(s). Disasters © Overseas Development Institute, 2014 Published by John Wiley & Sons Ltd, 9600 Garsington Road, Oxford, OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA

Predictors of compliance with tornado warnings issued in Joplin, Missouri, in 2011

Figure 1. Estimated tornado path within the city of Joplin, MO

Source: author, based on various sources.

damage. The tornado destroyed four of Joplin’s schools, including the high school and a number of churches, nursing homes, and warehouses, and damaged two large apartment buildings and two fire stations. Six other schools and several industrial buildings also were damaged (The Joplin Globe, 2011b). In addition, gas leaks in various areas caused injuries and overnight fires around the city. This twister, which tracked just south of downtown, narrowly missing the area, caused an estimated USD 3 billion in insured losses, excluding uninsured structures that were damaged (Paul and Stimers, 2011).   The tornado killed 162 people and injured more than 1,000, and was the first such event in the US to result in more than 100 fatalities since the Flint, Michigan, tornado of 8 June 1953, which claimed 116 lives (Mustain, 2011; NOAA, 2011). To put this number in perspective, only 45 tornado fatalities were recorded in the US during all of 2010, and just 21 in 2009. Average annual tornado deaths in the US over the past 30 years amount to around 55 (Simmons and Sutter, 2011); thus, the record number of deaths caused by the Joplin tornado was far higher than the average number of annual tornado deaths in the US in the past three decades.   Nearly a month before the Joplin tornado, on 27 April 2011, tornadoes killed 339 people, 248 of whom were in Alabama (Klockow, 2011; Senkbeil, Rockman, and Mason, 2012). On that day, a powerful storm system developed across the southeast

109

110 

Bimal Kanti Paul, Mitchel Stimers, and Marcellus Caldas

US, producing 226 confirmed tornadoes, 70 of which were in Tennessee and 62 in Alabama. It produced more strong (EF-2–EF-3) to violent (EF-4–EF-5) tornadoes than the historic April 1974 tornado outbreak, when, in a single 24-hour period, 148 confirmed tornadoes occurred in 13 US states and the Canadian province of Ontario (Simmons and Sutter, 2012). Ultimately, 27 April 2011 was the deadliest tornado day in the US since the ‘tri-state’ tornado outbreak on 18 March 1925, which killed 695 people in Illinois, Indiana, and Missouri, (Grazulis, 1993). Note that Joplin was struck by a single tornado whereas a (super) outbreak occurred on 27 April.   Hazard researchers (see, for example, Paul and Stimers, 2011, 2012; Simmons and Sutter, 2012) have sought to understand the reasons for the unexpectedly high number of fatalities in Joplin. This study, however, presents the results of a posttornado survey of survivors in Joplin to determine factors associated with responses to tornado warnings.   For background, the paper reviews initially variables linked to behaviour governing shelter-seeking by an at-risk population. Next it describes the data collection procedures and statistical techniques used to fulfil the main objective of the research: to help policymakers and private and public emergency management agencies to improve tornado preparedness and responses not just in Joplin but also elsewhere in the US. The last two sections contain the results and a discussion and conclusion.

A review of factors associated with compliance with tornado warnings A vast array of studies conducted in the US have assessed the public’s responses to tornado warnings provided by the National Weather Service (NWS) via broadcast meteorologists and other relevant sources, including the activation of sirens (see, for example, Liu et al., 1996; Balluz et al., 2000; Paul et al., 2003; Schmidlin et al., 2009; Sherman-Morris, 2005, 2010; Simmons and Sutter, 2011, 2012). The findings provided the necessary background to carry out this research and acted as a guide in selecting appropriate determinants of compliance with tornado warnings. These studies claim that if tornado warnings are not issued in a timely manner (or not issued at all) with sufficient lead time, the population within the target area is less able to take cover, which may result in more deaths and injuries (Balluz et al., 2000; Simmons and Sutter, 2008, 2011).   Furthermore, knowing where to go during a tornado is just as important as receiving a warning (Senkbeil, Rockman, and Mason, 2012). Safe cover may not necessarily be a designated shelter located near a residence, workplace, or shopping facility. If a shelter is not available, people can move to the lowest level of the building after receiving a warning, placing as many walls between them and the outside as possible. In the absence of a basement in a residence or other building, interior rooms such as a bathroom can serve as cover. Moreover, experts recommend staying away from windows and covering one’s head, identifying a safe location in advance, and talking

Predictors of compliance with tornado warnings issued in Joplin, Missouri, in 2011

to family members and/or colleagues to ensure a quicker response time (Knapp, 2012; Senkbeil, Rockman, and Mason, 2012).   For those travelling by car, it is safer to stop and allow the tornado to pass than to try to drive through it (Knapp, 2012; Schmidlin et al., 2002). This is because not all tornadoes have a clear, visible funnel. Tornadoes can be wrapped completely in rain, making them difficult to detect with the naked eye. Moreover, severe storms, even without a tornado, are also a source of lightning, high winds, hail, and flash floods. Any of these hazards can be just as deadly as a tornado (Knapp, 2012). A highway overpass, however, is not a safe shelter in the event of a storm or tornado. The NWS recommends, therefore, that automobile occupants abandon their vehicles and lie in a ditch if a tornado approaches (Simmons and Sutter, 2011).   Some studies also claim that people frequently ignore tornado warnings and do not take cover for several reasons (see, for example, Sherman-Morris, 2005; Simmons and Sutter, 2011). For instance, people often disregard warnings if they have proven to be incorrect in the past. This is known in the hazard literature as the ‘cry wolf effect’ or the ‘false alarm effect’ (Barnes et al., 2007). Simmons and Sutter (2011, p. 118), referring to the false alarm ratio (FAR) in 2004, report that a tornado does not occur within the warned area ‘nearly three out of four times’. Although the FAR fluctuates over time, it does remain close to the 2004 figure. One reason for this high FAR is that tornado warnings for a single event are issued for a relatively large area, so the probability that a tornado will strike any one home in an area under warning is very low (Simmons and Sutter, 2011). Moreover, people are likely to ignore warnings if the same type of alert is issued for a less severe tornado. Barnes et al. (2007) calls this ‘over-warning’ in their conceptual model of a broader, more general depiction of warnings for possible events (see Figure 2). In contrast, a tornado could occur without any warning; this constitutes an un-warned tornado (Barnes et al., 2007).   Other reasons for not heeding warnings include failure to realise the danger, waiting to get additional credible confirmation of an impending tornado, not having enough time to seek shelter, not understanding the warning owing to a language barrier (particularly applicable to immigrants), not believing that a tornado is coming, or thinking that ‘God will protect us’ (Balluz et al., 2000; Paul et al., 2003; Shermann-Morris, 2005; Biddle, 2007). Responses to tornado warnings also depend on whether a warning conveys a sufficiently high level of risk. Some who do not Figure 2. Conceptual model of warning accuracy

Source: modified version of information in Barnes et al. (2007).

111

112 

Bimal Kanti Paul, Mitchel Stimers, and Marcellus Caldas

respond to warnings may think that warnings are issued ‘all the time’ (Simmons and Sutter, 2011, p. 118).   Tornado warnings are most effective when they are often repeated, confirmed, and perceived as credible by the public (Shermann-Morris, 2005; Sorensen and Vogt, 2006; Paul, 2011). Sorensen (2000) maintains that when a hazard warning is first issued, the public usually does not act immediately. Instead, people attempt to find other sources of information before taking appropriate steps. Access to more than one warning source also is important because if one system is not operational, people can utilise another. Thus, having access to many sources of information bolsters an individual’s response to a tornado warning (Klockow, 2011; Simmons and Sutter, 2011).   Simmons and Sutter (2006, 2011) further claim that, among other factors, the cost of sheltering and the effectiveness of sheltering influence the response. They maintain that the cost of sheltering reduces the likelihood of responding to a tornado warning. This cost varies across the day, being higher for individuals at work. It also depends on the exact activities disrupted, the ease of rescheduling these activities around the warning, and the activities people can undertake while sheltering. For example, a person who was reading a newspaper when a warning was issued and has a safe room could take shelter and still continue to read during the warning, which generally is in effect for nearly an hour (Simmons and Sutter, 2011). Hence, the cost of sheltering for some residents could be quite low. According to Simmons and Sutter (2011), the effectiveness of sheltering, meanwhile, is reduced likelihood of injury and/or death. An increase in the effectiveness of sheltering would make residents more likely to respond to a tornado warning (Simmons and Sutter, 2011).   Several individual characteristics (such as age, gender, and level of education) and household characteristics (such as annual income, family size, and landownership) also influence how the public responds to a tornado warning (Balluz et al., 2000; Paul, 2011; Simmons and Sutter, 2011). These characteristics can interact with more than one of the factors above (including access to warning sources, greater lead time, existence of a shelter-seeking plan, and past experience of warnings) (Simmons and Sutter, 2011). For instance, a lack of or limited access to warning sources means that poor and less educated persons may not take protective action either because they do not receive a warning in the first place or because they do not receive confirmation of the warning (Paul, 2011). Economists have found that higher incomes might lead residents to purchase an NOAA (National Oceanic and Atmospheric Administration) weather radio or another emergency alert product (Simmons and Sutter, 2011). Weather radio offers access to a nationwide network of radio stations broadcasting continuous weather information from a nearby NWS Office. Some versions of these radios have additional features, such as bed-shakers and strobe lights, which can aid those with special needs.   In addition, ownership of a ‘safe room’ within a house influences the public response to tornado warnings. A safe room is a hardened structure specifically designed to meet Federal Emergency Management Agency (FEMA) criteria and provide ‘nearabsolute protection’ during extreme weather events, including hurricanes and tornadoes (FEMA, 2008). Since safe rooms cost a considerable amount, the poor again

Predictors of compliance with tornado warnings issued in Joplin, Missouri, in 2011

are at a disadvantage (Simmons and Sutter, 2011). The poor also are inclined to be more fatalistic about fatalities and injuries, so households with higher incomes tend to place a greater value on their avoidance, making a response more likely (Simmons and Sutter, 2011).   Studies have found that the elderly, the infirm, or the disabled tend to dismiss warnings in a cognitive process framed by situational factors such as compromised mobility and hearing difficulty (see, for example, Gruntfest, 1987; Daley et al., 2005). These people typically require additional time and/or assistance to get to the best available shelter. Moreover, Sherman-Morris (2005) claims that females are more likely than males to take cover against an impending tornado. Mileti and Sorensen (1990) claim that people directly affected by disasters are significantly more responsive to future hazard predictions and warnings than those who have never been affected by a disaster. However, if the event occurred in the distant past, people may fail to comply with hazard warnings because memories of such events tend to fade over time (Simmons and Sutter, 2011).   If moving to a safer location means leaving home, people at risk often defy warnings of an impending disaster (Simmons and Sutter, 2011). This is because, for instance, leaving home translates into a higher cost of responding. In addition, a lack of transportation, a belief that ‘it won’t happen to us’, or concerns about being unable to return home after a disaster because of curfews, job responsibilities, and the wellbeing of pets and other household members may prompt individuals to stay at home (Dow and Cutter, 1998; Health et al., 2001).   Most people generally take shelter at the last minute, so traveling even a short distance to a safer location in increasingly adverse weather conditions becomes difficult (Klockow, 2011). Klockow (2011), assessing the tornado outbreak across northern Alabama on 27 April 2011, reported that virtually all respondents headed to cover only in the two or less minutes before the event. The delay in sheltering resulted from the disbelief that a tornado was coming. Apparently, people ultimately sought shelter only after observing strong environmental evidence and/or physically seeing the advancing tornado (Klockow, 2011). Clearly, many factors influence compliance with tornado warnings, and these factors are interrelated. Thus, the decision to take cover in the face of an advancing tornado is complex.

Methods The primary data set utilised in this study has its origins in a population-based, crosssectional questionnaire survey of residents of Joplin, Missouri, that was done between 24 June 2011 and 16 February 2012. Given that 88 per cent of the families displaced by the tornado were living outside of Joplin (Tang, 2011), not many survivors were available at the time of the survey. For this reason, we used a combination of convenience and snowball sampling to identify residents of Joplin for interview. In addition, we selected participants from both tornado-affected and non-tornadoaffected parts of the city, and we implemented a pre-structured interview schedule

113

114 

Bimal Kanti Paul, Mitchel Stimers, and Marcellus Caldas

to collect relevant information from them. The questionnaire requested two types of information: (i) knowledge of tornado warnings and compliance with those warnings; and (ii) information on age, annual household income, educational attainment, employment status, gender, marital status, and relevant demographic features, procured via socioeconomic questions. The survey instrument also included questions about property damage, injuries sustained, number of deaths in the household, past experience of tornadoes, and characteristics of residential structure.   Collected data contained no identifiers. Resident participation in the interview process was voluntary and non-coercive, and participant confidentiality was respected throughout. We sought and subsequently received approval to use the interview schedule from the Human Subject Review Committee at Kansas State University. We obtained formal written consent from each and every respondent who agreed to take part in the interview after we explained the nature and the objectives of the study.   Six trained personnel, including two of the three authors of this paper, conducted face-to-face interviews with 133 respondents. Although all six personnel have carried out field surveys previously, all of them met and discussed the questionnaire before holding actual interviews, ensuring comparability and reliability of data from the field. At the outset, we held interviews in various locations in Joplin, including bars, libraries, gas stations, golf courses, residences, restaurants, retail outlets, shopping malls, temporary shelters, and the offices of emergency agencies. In addition, we completed eight interviews via telephone and the social media internet site, Facebook, garnering a total of 141 responses.   We also amassed relevant data through informal discussions with local emergency management personnel for Joplin and surrounding areas, such as Joplin city officials, fire and law-enforcement dispatchers, volunteers responsible for clearing debris, local professionals, and residents of Joplin and neighbouring areas. None of these individuals was selected for the questionnaire survey; instead, face-to-face conversations occurred with each to understand residents’ points of view on the process of warning reception to warning response and how decisions were made and to supplement the information gathered in respondent interviews. The Joplin Globe, the local daily newspaper, and other print media, such as The Kansas City Star, also provided valuable information for this study. Ultimately, this research attempted to synthesise the most common responses received from all of the various sources.   Based on the literature review, we pinpointed eight determinants of safety-seeking behaviour (see Table 1): • • • • • • • •

number of warning sources; presence (or lack) of a basement under a home; whether respondents were at home when the tornado struck; past tornado experience; age; gender; education level; and income level.

Predictors of compliance with tornado warnings issued in Joplin, Missouri, in 2011

Table 1. Predictors of compliance with tornado warnings derived from bivariate analysis (N=126) Independent variable

Compliance with tornado warnings

Total Number (%)

Yes Number (%)

No Number (%)

One

39 (40.21)

18 (62.07)

57 (45.24)

More than one

58 (59.79)

11 (37.93)

69 (54.76)

Number of warning sources

Chi-square value= 4.31 (d.f.=1; p=0.038) Presence (or lack) of a basement under a home Yes

18 (18.56)

5 (17.24)

23 (18.25)

No

79 (81.44)

24 (82.76)

103 (81.75)

Chi-square value=0.03 (d.f.=1; p=0.863) Whether respondents were at home when the tornado struck Yes

63 (6495)

12 (41.38)

75 (59.52)

No

34 (35.05)

17 (58.62)

51 (40.48)

Chi-square value=5.15 (d.f.=2; p=0.023) Past tornado experience Yes

52 (53.61)

23 (79.31)

75 (59.52)

No

45 (46.39)

6 (20.69)

51 (40.48)

Chi-square value=6.121 (d.f.=1; p=0.013) Gender Male

53 (54.64)

25 (86.21)

78 (61.90)

Female

44 (45.36)

4 (13.79)

48 (38.10)

Chi-square value=9.43 (d.f.=1; p=0.002) Education level Up to high school

58 (59.79)

20 (68.97)

78 (61.90)

Above high school

39 (40.21)

9 (31.03)

48 (38.10)

Chi-square value=0.08 (d.f.=1; p=0.371) Income level $59,000

24 (24.74)

7 (24.14)

31 (24.60)

Chi-square value=1.29 (d.f.=3; p=0.731) Note: the table does not show age because it was treated as a continuous variable. Furthermore, the application of a t-test confirmed that this variable had no significant statistical association with taking shelter in response to tornado warnings.

115

116 

Bimal Kanti Paul, Mitchel Stimers, and Marcellus Caldas

  The first point is significant as earlier studies (see, for example, Klockow, 2011) have reported that access to many sources of information on an incoming tornado bolsters an individual’s response to such an event.   Second, according to the Jasper County Assessor’s Office, approximately 78 per cent of houses across the county lack basements because of the area’s rocky ground and high water table (The Joplin Globe, 2011b). In fact, Joplin has an even lower percentage of basements than the county as a whole. Since a basement is important in responding to tornado warnings (Bulluz et al., 2000), we included this variable in the study.   Third, people are more likely to receive tornado warnings if they are at home, instead of in a car, movie theatre, or church (Simmons and Sutter, 2011). For this reason, we included the locations of respondents (whether or not they were at home) prior to the touchdown of a tornado as a determinant for taking safer shelter or complying with warnings. Available literature (see, for example, Balluz et al., 2000; Paul and Stimers 2011) also suggests that myriad individual and household characteristics (such as past tornado experience, gender, and education and income level) are associated with who receives as well as who complies with tornado warnings. We dichotomise the dependent variable as compliance (yes) or non-compliance (no) with tornado warnings. Places considered to be safer shelters include the interior rooms (without windows) of residential and non-residential buildings, basements, bathrooms, commercial cooler, crawl spaces, safe rooms, and public tornado shelters.   Since the dependent variable and many of the independent variables are categorical, we used a multivariate logistic regression (LR) technique to determine the factors affecting a move to shelter and calculated odds ratios (ORs) as measures of the strength of the association among variables. Before applying the LR statistical technique, we tested differences in the frequency distribution of the dependent and each of the independent variables using the chi-square test. Specifically, independent variables included in the LR model were all statistically associated (p