USF Libraries
USF Digital Collections

Before the storm :

MISSING IMAGE

Material Information

Title:
Before the storm : evacuation intention and audience segmentation
Physical Description:
Book
Language:
English
Creator:
Rice, Homer
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla
Publication Date:

Subjects

Subjects / Keywords:
Disaster
Planning
Hurricanes
Chi-Square
CHAID
Dissertations, Academic -- Dean's Office -- Masters -- USF   ( lcsh )
Genre:
non-fiction   ( marcgt )

Notes

Abstract:
ABSTRACT: The purpose of this study was to describe the predictors of evacuation intention among coastal residents in the State of Florida and to determine if there are meaningful segments of the population who intend to evacuate when told to do so by governmental officials because of a major hurricane. In the America's and the Caribbean, 75,000 deaths have been attributed to hurricanes in the 20th century. A well planned evacuation can reduce injury and death, yet many people do not have an evacuation plan and do not intend to evacuate when told to do so. The study used secondary data from the Harvard School of Public Health, Hurricane in High Risk Areas study, a random sample of 5,046 non-institutionalized persons age 18 and older in coastal counties of Texas, Louisiana, Mississippi, Alabama, Georgia, North Carolina, South Carolina and Florida. Surveys for the State of Florida were segregated and used in this analysis, resulting in a study sample of 1,006 surveys from 42 counties. When asked if they would evacuate in the future if told to by government officials, 59.1% of Floridians surveyed said they would leave, 35.2% said they would not leave and 5.6% said it would depend. In Florida, 65.7% of the population had been threatened or hit by a major hurricane in the last three years and 26.6% of those had left their homes because of the hurricane. Of those whose communities were threatened by a hurricane, 83.3% of the communities were damaged and 33.8% experienced major flooding associated with the hurricane. Bivariate statistics and logistic regression were used to explore the interactions of predictors and evacuation intention. The best predictor of evacuation intention was prior evacuation from a hurricane (chi-square= 45.48, p < .01, Cramer's V = 0.266). Significant relationships were also demonstrated between evacuation intention and worry a future hurricane would hit the community (chi-square = 22.75, p < .01, Cramer's V = 0.11), the presence of pets (chi-square = 6.57, p < .01, Cramer's V = 0.084), concern the home would be damaged (chi-square = 19.41, p < .01, Cramer's V = 0.10), belief the home would withstand a major hurricane (chi-square = 19.55, p < .01, Cramer's V = 0.10), length of time in the community (chi-square = 26.59, p < .01, Cramer's V = 0.12), having children in the household (chi-square = 11.13, p < .01, Cramer's V = 0.11), having a generator (chi-square = 17.12, p < .01, Cramer's V = 0.13), age (chi-square = 24, p < .01, Cramer's V = 0.16) and race (chi-square = 12.21, p = .02, Cramer's V = 0.12). Logistic regression of the predictors of evacuation intention resulted in significant relationships with previous evacuation experience (OR = 4.99, p < .001), age 30 to 49 compared to age over 65 (OR = 2.776, p < .01), the presence of a generator (OR = .447, p < .01), having a home not very likely to be damaged compared to a home very likely to be damaged (OR =.444, p = .018), and experiencing poor prior government and voluntary agency response to previous hurricanes compared to excellent response (OR = .386, p < .027). Chi-squared Automatic Interaction Detection (CHAID) was used to identify segments of the population most likely and least likely to evacuate when told to do so. Those most likely to evacuate had evacuated due to a previous hurricane. Those least likely to evacuate when told to do so had not evacuated in a previous storm, do not own a generator and are over the age of 65. Information from this study can be used in planning for evacuation response by governmental entities. Available demographic information can be used to determine numbers of persons likely to evacuate before a storm. The results of this study can be used to inform a marketing strategy by government officials to encourage evacuation among those who say they would not evacuate when told to do so. Further research is needed to determine additional characteristics of the populations who say they will and will not evacuate when told to do so.
Thesis:
Dissertation (PHD)--University of South Florida, 2010.
Bibliography:
Includes bibliographical references.
System Details:
Mode of access: World Wide Web.
System Details:
System requirements: World Wide Web browser and PDF reader.
Statement of Responsibility:
by Homer Rice.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains X pages.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
usfldc doi - E14-SFE0004706
usfldc handle - e14.4706
System ID:
SFS0028013:00001


This item is only available as the following downloads:


Full Text

PAGE 1

Before the Storm: Evacuation Intention and Audience Segmentation By Homer J Rice, RS, MPH A d issertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Community and Family Health Co llege of Public Health University of South Florida Major Professor: Carol A Bryant, Ph.D. Moya Alfonso, Ph.D. Elizabeth Pathak, Ph.D. Graham Tobin, Ph.D. Wayne Westhoff, Ph.D. Date of Approval November 19, 2010 Keywords: Disaster, P lanning, H urri canes, ChiSquare, CHAID, Copyright 2010. Homer J. Rice, MPH

PAGE 2

DEDICATION This work is dedicated to my wife and children, who stood by me and encouraged me while I continued on this quest. I couldnt and wouldnt have done it without you. I e sp ecially dedicate this dissertation to my wife, who has put up with late nights, long days and the ramblings of a mad man. I have loved you since we met when I was only 6, and I will love you forever. To my father, who never made it out of 3rd grade, but who instilled in me a life long love of learning. I made it Dad, I hope youre watching.

PAGE 3

ACKNOWLEDGMENTS I would like to acknowledge the work of the faculty who taught me and the committee that put up with me through this process. I learned so much from all of you and will never forget you. I truly enjoyed the learning experience and the interaction with all of you. I am grateful to Moya Alfonso, Ph.D. for her insight and assistance with CHAID I also wish to acknowledge and thank Wayne Westhoff, Ph.D., Elizabeth Pathak, Ph.D., and Graham Tobin, Ph.D., who served on my committee and worked around my schedule to my meetings, proposal and defense. I wish to acknowledge and thank Dr. Michael Reid who served as my outside chair for my proposal and my defense, but also served as a mentor in the Florida Public Health Leadership Institute, a major influence on my decision to pursue this degree. I am especially grateful to my major professor and chair, Dr. Carol Bryant for her patience and guidance. I am not a traditional student and I am 300 miles away, but somehow she got me through. Thank you all.

PAGE 4

i Table of Contents List of Tables --------------------------------------------------------------------------------------i v List of Figures -------------------------------------------------------------------------------------vi Abstract -------------------------------------------------------------------------------------vii Chapter One: Introduction ---------------------------------------------------------------------1 Research P ro blem ----------------------------------------------------------------------3 Methodology ------------------------------------------------------------------------------9 Research O bjectives ----------------------------------------------------------9 Secondary Data Source -----------------------------------------------------10 Significance ------------------------------------------------------------------------------1 2 Chapter Two: Literature Review -------------------------------------------------------------1 3 Introduction ------------------------------------------------------------------------------1 3 Evacuation -------------------------------------------------------------------------------1 3 Evacuation Intention -------------------------------------------------------------------20 Predictors of Evacuation Intentions and Beh avior -----------------------------2 2 Theoretical Frameworks --------------------------------------------------------------2 3 Emergent n orm theory -------------------------------------------------------2 4 Risk p erception ----------------------------------------------------------------2 7 Risk communication ----------------------------------------------------------2 7 Receiving and understanding -------------------------------------2 8 Believing ----------------------------------------------------------------2 8 Personalizing ----------------------------------------------------------2 9 Information seeking --------------------------------------------------2 9 Responding ------------------------------------------------------------30 The perception of risk and prior experience ----------------------------30 Other factors that affect risk perception ---------------------------------31 The Warning and Response Model ---------------------------------------3 2 Identifying Population Segments At Risk ----------------------------------------3 8 ChiSquared Automatic Interaction Detection (CHAID) --------------4 2 Segmentation and disaster preparedness ------------------------------4 8 Chapter Three: Method s ----------------------------------------------------------------------5 0 Research A pproach -------------------------------------------------------------------50 Secondary D ataset --------------------------------------------------------------------5 1

PAGE 5

ii Protection of Human Subjects ------------------------------------------------------5 5 Analysis Plan ----------------------------------------------------------------------------5 5 Creation of dataset ------------------------------------------------------------5 5 Dependent v ariable ---------------------------------------------------------5 6 Independent variables -------------------------------------------------------5 6 Relationships between evacuation intention and other variables -------------------------------------------------------------------------5 8 Identification of meaningful s egment s of people who intend to evacuate ---------------------------------------------------------------------60 Limitations -------------------------------------------------------------------------------6 2 Chapter Four: Results -------------------------------------------------------------------------6 4 Introduction ------------------------------------------------------------------------------6 4 Research purpose ------------------------------------------------------------6 4 Description of the P opulation --------------------------------------------------------6 5 D ependent variable frequencies -------------------------------------------6 7 Independent variable frequenc ies -----------------------------------------6 8 R elationships with E vacuation I ntention ------------------------------------------7 1 Logistic regression of variables in bivariate analysis -----------------7 8 CHAID A nalyses -----------------------------------------------------------------------81 CHAID analysis with all variables of interest ---------------------------8 2 CHAID analysis with logistic regression variables --------------------8 6 CHAID and evacuati on experience ---------------------------------------9 1 Regression A nalyses of CHAID V ariables ---------------------------------------9 5 Summary of F indings -----------------------------------------------------------------9 6 Chapter Five: Discussion -------------------------------------------------------------------101 Introduction ----------------------------------------------------------------------------101 Purpose of the R esearch -----------------------------------------------------------101 Overview of the S tudy M ethod ----------------------------------------------------102 Summary of Findings ---------------------------------------------------------------103 Hypotheses tested ----------------------------------------------------------107 Evacuation consequences ------------------------------------------------108 Contribution to T heory --------------------------------------------------------------1 10 Strengths and L imitations ----------------------------------------------------------112 Future R esearch ---------------------------------------------------------------------116 Dissemination -------------------------------------------------------------------------118 Utilization ----------------------------------------------------------------------118 References -----------------------------------------------------------------------------------1 20 Appendices -----------------------------------------------------------------------------------132 Appendix A : Hurricane readiness in high risk areas survey --------------133 Appendix B : Populatio n statistics for surveyed counties ------------------1 4 1 Appendix C : Correlation m atrix of significant variables in b ivariate a nalysis ---------------------------------------------------------------144

PAGE 6

iii Appendix D: Correlation matrix of preparedness variables ---------------149 About the Author ----------------------------------------------------------------------End Page

PAGE 7

iv List of Tables Table 1 Stages of change and preparedness level ------------------------------7 Table 2 Hurricane evacuation expectations and actual behavior in Hurricane Lili ------------------------------------------------------------------2 3 Table 3 Predictors of evacuation and evacuation intention. -------------------2 3 Table 4 Warning and response propositions. -------------------------------------3 7 Table 5 Example Income Categories. --------------------------------------------4 4 Table 6 Survey population demographic description ---------------------------5 3 Table 7 Measures of risk perception, prior experience and family accountability ------------------------------------------------------------------5 7 Table 8 Cohens effect size interpretation rules of thumb ----------------------5 9 Table 9 Florida survey population demographics compared to the general survey population. --------------------------------------------------6 6 Table 10 Risk perception variables frequencies -----------------------------------6 9 Table 11 Family accountability frequencies -----------------------------------------70 Table 12 Prior hurricane experience --------------------------------------------------70 Table 13 Risk perception associations with evacuation intention -------------7 3 Table 14 Risk perception associations with prior evacuation experience ----------------------------------------------------------------------7 4 Table 15 Demographic characteristics with evacuation intention --------------7 6 Table 16 Variables significant i n chi square and correlations ------------------7 7 Table 17 Variables in the regression model ----------------------------------------80

PAGE 8

v Table 18 Variables significant in the regression model --------------------------81 Table 19 Effect size values for segmentation of evacuation intention --------8 6 Table 20 Effect size values for segmentation of evacuation intent ion with significant regression variables --------------------------------------8 8 Table 21 CHAID variables entered into the regression model -----------------9 5 Table 2 2 Factors associated with evacuation intention --------------------------9 8 Table 23 Summary of logistic regression --------------------------------------------9 8 Table 24 Risk perception variables associated with prior evacuation experience ----------------------------------------------------------------------9 9 Table 25 Factors significant in segmentation -------------------------------------100

PAGE 9

vi List of Figures Figure 1 The Warning and Response Model ---------------------------------------3 4 Figure 2 Sample CHAID tree diagram -----------------------------------------------4 6 Figure 3 Reasons respondents would not evacuate when told ---------------5 4 Figure 4 Reasons Florida respondents would not evacuate when told ------6 8 Figure 5 Segmentation of evacuation intention with all regression variables ------------------------------------------------------------------------8 5 Figure 6 Segmentation of evacuation intention with significant regression variables ----------------------------------------------------------90 Figure 7 Segmentation of evacuation intention without evacuation experience variable -----------------------------------------------------------9 3 Figure 8 Segmentation of evacuation intention with only those cases threatened by a hurr icane in the last three years ---------------------9 4 Figure 9 Frequency of indicators of planning -----------------------------------112

PAGE 10

vii ABSTRACT The purpose of this study was to describe the predictors of evacuation intention among coastal residents in the State of Florida and to determine if there are meaningful segments of the population who intend to evacuate when told to do so by governmental officials because of a major hurricane. In the Americas and the Caribbean, 75,000 deaths have been attributed to hurricanes in the 20th century A well planned evacuation can reduce injury and death, yet many people do not have an evacuation plan and do not intend to evacuate when told to do so. The study used secondary data from the H arvard School of Public Health, Hurricane in High Risk Areas study a random sample of 5,046 noninstitutionalized persons age 18 and older in coastal counties of Texas, Louisiana, Mississippi, Alabama, Georgia, North Carolina, South Carolina and Florida. Surveys for the State of Florida were segregated and used in this analysis, resulting in a study sample of 1,006 surveys from 42 counties When asked if they would evacuate in the future if told to by government officials, 59.1% of Floridians surveyed said they would leave, 35.2% said they would not leave and 5.6% said it would depend. In Florida, 65.7% of the population had been threatened or hit by a major hurricane in the last three years and 26.6% of those had left their homes because of the hurricane. Of those whose

PAGE 11

viii communities were threatened by a hurricane, 83.3% of the communities were damaged and 33.8% experienced major flooding associated with the hurricane. Bivariate statistics and logistic regression were used to explore the interactions of predictors and evacuation intention. The best predictor of evacuation intention was prior evacuation from a hurricane ( chi square= 45.48, p < .01, Cramers V = 0.266). Significant relationships were also demonstrated between evacuation intention and worry a future hurricane would hit the community ( chi square = 22.75, p < .01, Cramers V = 0.11) the presence of pets ( chi square = 6.57, p < .01, Cramers V = 0.084 ), concern the home would be damaged ( chi square = 19.41, p < .01, Cramers V = 0. 10) belief the home would withstand a major hurricane ( chi square = 19.55, p < .01, Cramers V = 0.10), length of time in the community ( chi square = 26.59, p < .01, Cramers V = 0.12) having children in the household ( chi square = 11.13, p < .01, Cramers V = 0. 11) having a generator ( chi square = 17.12, p < .01, Cramers V = 0.13) age ( chi square = 24, p < .01, Cramers V = 0. 16) and race ( chi sq uare = 12.21, p = .0 2 Cramers V = 0. 12). Logistic regression of the predictors of evacuation intention resulted in significant relationships with previous evacuation experience (OR = 4.99, p < .001), age 30 to 49 compared to age over 65 (OR = 2.776, p < .01), the presence of a generator (OR = .4 4 7, p < .01), having a home not very likely to be damaged compared to a home very likely to be damaged (OR =.444, p = .018) and experiencing poor prior government and voluntary agency response to previous hurricanes compared to excellent response (OR = .386, p < .027) Chisquared Automatic Interaction Detection (CHAID) was used to identify segments of the

PAGE 12

ix population most likely and least likely to evacuate when told to do so. Those most likely to evacuate had evacuated due to a previous hurricane. Those least likely to evacuate when told to do so had not evacuated in a previous storm, do not own a generator and are over the age of 65. Information from this study can be used in planning for evacuation response by governmental entities. Available demographic information can be used to determine numbers of persons likely to evacuate before a storm. The results of this study can be used to inform a marketing strategy by government officials to encourage evacuatio n among those who say they would not evacuate when told to do so. Further research is needed to determine additional characteristics of the populations who say they will and will not evacuate when told to do so.

PAGE 13

1 Chapter One: INTRODUCTION At the 2003 Governors Hurricane Conference, then Florida Governor Jeb Bush said, We know what we need to do, we just dont do it. This sums up the state of affairs in 2004 prior to the impact of four major hurricanes on Florida ( Bailey, Glover, and Huang, 2005) Even after the destructive power of Hurricane Katrina was watched by the nation, recent evidence indicates that people still fail to plan and prepare for natural disasters (Baker, 2006) and even most New Orleans residents do not know where their evacuation shelters are located ( Blendon, Buhr, Benson, Weldon, and Herrmann, 2007) Hurricanes are among the most dangerous storms on Earth as well as the most frequent natural disasters to occur in the United States (Malilay, 1997) During the period 1992 to 1997, 71% of the federally declared disasters in the United States were related to hurricanes In the Americas and the Caribbean, 75,000 deaths have been attributed to hurricanes in the 20th century (Bourque, Siegel, Kano and Wood, 2006) According to the Florida Department of Community Affairs 36% of twentieth century U S hurricanes hit Florida. During the period from 1851 through 2004, 110 hurricanes made landfall in Florida and 273 made landfall on the mainland of the United States ( Blake, Rappaport and Landsea, 2007 ) Hurricane Andrew struck Dade County, Florida, in August 1992 causing

PAGE 14

2 damages estimated at 26 to 34 billion dollars (Rappaport and Fernandez Partagas, 1995 ) I n 1999 Hurricane Floyd caused the largest peacetime evacuation in U.S hi story up to that time by displacing 2.5 million people along the East Coast of Florida. During the height of evacuation, more than 2 million people were asked to evacuate due to Hurricane Charl ey in 2004 (Tobin, Bell, Montz, Hughey, Whiteford, Everist, Kelsey and Miller, 2005) Previous studies e stimate d the average annual cost of hurricane damage in the U nited S tates at $4.8 billion annually i n 1995 dollars (Pielke and Landsea, 1998) but that amount rose significantly after the 2004 and 2005 storm seas ons (National Oceanic and Atmospheric Administration, 2006). Hurricane Katrina alone caused over $100 billion in damages and cost as much as $1 million per mile of evacuated coast line from lost wages, lost tourism, lost commerce and personal expenses In another study of the cost to evacuate the North Carolina coast Whitehead (2003) estimated costs between $1 million and $50 million per county, depending on the nature of the storm and emergency management policies This is less than Katrinas $1 milli on per mile, but still a significant cost Evacuation is clearly not cheap. In addition to economic loss es hurricanes have been responsible for numerous deaths in the United States The Galveston, Texas, hurricane of September 1900 caused more than 8,000 deaths (Rappaport and Fernandez Partagas, 1995) Six hundred deaths in the United States were attributed to hurricanes between 1970 and 1999. Hurricane Andrew left 34 people dead in its wake in August 1992. In 2004, an unprecedented four storms made l andfall in

PAGE 15

3 Florida, causing 124 deaths ( Bailey, Glover, and Huang, 2004) and in 2005 hurricane Katrina devastated the Gulf Coast, taking the lives of 1,833 people ( NOAA, 2006) In addition to the physical da nger posed by wind and rain hurricanes create s econdary hazards that can kill or injure, such as carbon monoxide poisoning from improper generator use and electrocution from damaged power lines From the mid 1960 s through the mid 1990 s, hurricane activity was light (Pielke and Landsea, 1998) D uri ng that time Florida experienced increased development along its 11,000 mile coastline. Recently researchers have been predicting a return to increased major hurricane activity in the Atlantic In conjunction with the major coastal growth in Florida, thi s leads to the assumption that even greater economic losses and loss of life may occur in the near future. Research Problem Natural disasters cannot be prevented but much can be done to prepare for them and thus avoid morbidity and mortality Both Feder al and State emergency management agencies recommend that families have a current personal disaster plan that includes an evacuation plan, a communication plan, a designated place to meet if separated, and plans for sheltering in place. Templates for the creation of such plans are available at the official state disaster website ( http://www.floridadisaster.org ) the American Red Cross, and Homeland Security web site s. A well thought out disaster plan can save time and lives in the event of an emergency a nd yet the 2009 Ci tizen Corps national survey indicated 55% of the respondents did not have a household disaster plan. Even

PAGE 16

4 those who said they were prepared for a disaster were lacking key elements of a disaster plan. In the Citizen Corps survey only 38% of those who said they have been prepared for the last six months had emergency supplies at home and nearly 60% did not know their local community evacuation routes (FEMA, 2009) An earlier study in 1996 by the Florida Department of Community Affairs (FDCA) indicated that 30.6% of coastal county residents and 47.8% of noncoastal residents of Florida did not know the hurricane evacuation route for their area. In addition, 40.3% of coastal county residents and 44.9% of noncoastal residents did not know the location of the local hurricane shelter. In some natural disasters injury or death may be avoided through evacuation. In the case of U. S. hurricanes warnings are provided well in advance of landfall and a well planned evacuation can reduce injury and death (Bourque, et al. 2006). Evacuation also enhances the operation of emergency and recovery crews by allowing them to concentrate on preventing further damage and repairing existing damage rather than rescue and body recovery operations (Perry, 1979). A review of previous research supported the link between behavioral intention and actual behavior set forth in Fishbein and Ajzens (1975) theory of reasoned action (Sheppard, Hartwick and Warshaw, 1988) T he intent to evacuate has also been linked with actual evacuation behavior (Horney, MacDonald, Van Willigen, Berke, and Kaufman, 2010). Much of the previous research has been focused on past events, determining who evacuated after the fact. Emergency planners need to know who is planning to evac uate in the event

PAGE 17

5 of a storm so they can prepare for the needs of the population prior to the event. Unfortunately, many people do not evacuate, even when warned or ordered to leave. When Hurricane Alicia struck the Texas coastline near Galveston as a cat egory three storm in August 1983, only 47% of those in the warning area evacuated (Baker, 1991). Hurricane Andrew struck south Florida in 1992 as the third most powerful storm on record. When evacuation orders were issued for Broward and Dade Counties pr ior to impact 3 0% of those in high risk areas did not leave, leading to 14 deaths directly related to Andrew (US Army Corps of Engineers, 1993). In August 2005, 90% of southeast Louisiana evacuated before Hurricane Katrina; still more than 100,000 people stayed in the city of New Orleans, many of whom had to swim for their lives, wade through contaminated waters, or remain trapped on rooftops and in attics ( Effects of Hurricane Katrina in New Orleans 2009) A study of transient populations in five disas ters found that only 75% of tourists and 58% of homeless people evacuated when warned. Various other studies have found that evacuation rates ranged from 32% to 98%, depending on the level of perceived threat and the warning level (Dow and Cutter, 1998). A story in the Tallahassee Democrat published June 14, 2006, ran with the heading, Many shrug off evacuation ahead of Alberto. Low numbers worry officials. Even after the devastating storms of 2004 and 2005, the story stated In Steinhatchee, about 20 miles south of where the storm landed, locals shrugged off mandatory evacuation orders and stayed to take pictures, swim,

PAGE 18

6 party and watch the storm blow over their heads (Deslatte, 2006). Experience could logically be considered a predictor of evacuati on, yet as insinuated in the quote above, experience has not consistently predicted evacuation. In multiple storm studies experience has been cited as both a major predictor of evacuation and a major predictor of non evacuati on ( Perry, 1985; Baker, 1991; Riad, Norris and Ruback, 1999; Burnside, Miller, and Rivera, 2007) It seems that even with previous storm experience, people dont leave. Homeland Security Presidential Directive 8 directed the Secretary of Homeland Security to develop national preparedness goals. In response to this directive the National Preparedness Guidelines and associated Target Capabilities List were prepared. One of the Target Capabilities in the area of Community Preparedness and Participation recommends that 80% of the popul ation be prepared to evacuate in an emergency. That goal has not yet been met. In their analysis of surveys of personal and business preparedness since 2001, t he Community Preparedness Division of the F ederal E mergency M anagement A gency (FEMA) recommended additional research be performed to investigate contextual characteristics related to disaster preparedness and to explore motivational factors and barriers to preparedness to reach this goal (FEMA, 2009). In 2009, FEMA commissioned the Citizen Corps Na tional S urvey to gauge the state of disaster preparedness among citizens in the United States (FEMA, 2009) The survey included information on the individuals stage of disaster

PAGE 19

7 preparedness. Participants were asked to rate their preparedness level by ch oosing one of the statements : I am not planning to do anything about preparing; I have not yet prepared but I intend to in the next 6 months ; I have not yet prepared but I intend to in the next month; I just recently began preparing; and I have been prepar ed for at least the last 6 months. These preparedness levels were intended to place people in the five stages of change or readiness to change proposed by J.O. Prochaska and C.C. DiClemente in their Transtheoretical Model of Behavior Change (1983) The fi ve stages are precontemplation, contemplation, preparation, action, and maintenance. The stages of change are matched to the levels of preparedness above as shown in Table 1. Table 1. Stages of change and preparedness level Stage of Change Preparedness Level Precontemplation I am not planning to do anything about preparing. Contemplation I have not yet prepared but I intend to in the next 6 months. Preparation I have not yet prepared but I intend to in the next month. Action I just recently began preparing Maintenance I have been prepared for at least the last 6 months.

PAGE 20

8 In this study the largest group of respondents, 32%, had been prepared for at least the past 6 months (maintenance stage). The second largest group of respondents, 27%, fell in the precontemplation phase. The 2009 Citizen Corps National Survey recommended that efforts should focus on those in the contemplation and preparation stages of preparedness to move them into the preparation and action stages To reach people who do not cu rrently plan to evacuate, it would be important to know their demographic and attitudinal characteristics It also would be valuable to know the characteristics of those who already have plans to evacuate in face of danger. This information would allow o fficials to identify those in greatest need and design the most appropriate strategies for each audience segment. Many studies have looked at evacuation behavior in the face of disaster ; very few have examined evacuation intentions or readiness to change. The intent to evacuate is part of the process of planning for disaster and a pivotal point in the move from contemplation to action. Some studies have attempted to associate hurricane strike probabilities potential storm strength, or disaster warning s ources with intended evacuation or included evacuation intention as a predictor (Baker, 1979 ; Whitehead, Edwards, Willigen, Maiolo, Wilson and Smith, 2000; Burnside et al. 2007; Horn et al., 2010) A study of evacuation from Hurricane Isabel found intended evacuation from a future storm was associated with evacuation after controlling for home type. Horn concluded that evacuation intentions are an important factor in actual evacuation (Horn et al., 2010). However, no research could be identified that ex amined differences in personal

PAGE 21

9 characteristics beyond basic demographics in peoples readiness or intention to evacuate. Methodology Research o bjectives. The primary purpose of this research is to identify predictors of evacuation intention and describe t he personal and attitudinal characteristics of segment s within the population in terms of their intent to evacuate in the face of hurricane warnings. For purposes of this study evacuation is defined as an orderly vacating of the normal place of residence to seek shelter in another location. The location might be a governmental shelter, a friend or relatives, or temporary lodging at a hotel or motel. Questions to be investigated in this work include: What proportion of the coastal population intends to evacuate when recommended by public officials prior to an approaching hurricane ? What factors are associated with the intention to evacuate when recommended by public officials prior to an approaching hurricane? What factors are useful in identifying meani ngful segments of people who intend to evacuate when recommended by public officials prior to an approaching hurricane? Based upon Lindell and Perrys (1992) Warning and Response M odel I hypothesize that the intention to evacuate when recommended by publi c officials prior to an approaching hurricane will be influenced by the following factors :

PAGE 22

10 Prior experience evacuating during hurricane threats will increase the probability of evacuation intention; Prior experience evacuating during hurricane threats will be positively related to the level of risk perception; The higher the level of risk perception, the higher will be the probability of evacuation intention; Having family members together at the time of warning, or otherwise accounted for, will increase the probability of evacuation intention; Ethnic majority status will be inversely related to the probability of evacuation intent ; and Socioeconomic status will be inversely related to the probability of evacuation intent. Secondary data source. This study will rely on secondary analysis of data from the Hurricane Readiness in High Risk Areas study by the Harvard School of Public Health Project on Public Health and Biological Security. The Harvard School of Public Health Project on Public Health and Biologi cal Security is an ongoing program funded by the Centers for Disease Control and Prevention to conduct opinion surveys to assess public knowledge, attitudes, and behavior in response to public health threats. Through an agreement with the National Preparedness Leadership Initiative, the Project assists the CDC and other public officials by monitoring the publics response to public health threats (Harvard, 2008). Previous studies have included attitudes toward the use of quarantine, the public

PAGE 23

11 response t o severe acute respiratory syndrome, and the impact of the anthrax attacks on the American public. In 2007, the Project contracted with International Communications Research, an independent research company, to conduct interviews on hurricane readiness am ong a representative sample of 5,046 respondents age 18 and older in coastal counties of Texas, Louisiana, Mississippi, Alabama, and Florida. I nterviews were conducted by telephone from June 18 to July 10, 2007, in all counties within twenty miles of the coastline for each of these states (Blendon, Buhr, Benson, Weldon, and Herrmann, 2007). Interviewers asked respondents questions designed to assess perceived risk of danger from hurricanes prior experience evacuating during a hurricane threat prior experience with hurricanes, preparation activities and planning for hurricanes. Additional questions were used to determine family structure, residence and demographics of the population. The primary dependent variable to be used in this study is evacuati on intent, measured by the response to the question I f government officials said that you had to evacuate the area because there was going to be a major hurricane in the next few days, would you leave the area or would you stay? The Harvard study collec ted data on demographics, prior experience, risk perception, and current preparation. These variables were explored as potential predictors of evacuation intent (A copy of the Harvard study survey instrument can be found in Appendix A ). This data will be used to determine the proportion of the coastal populations who intend to evacuate when recommended by public officials prior to an approaching hurricane, factors associated with the intention to

PAGE 24

12 evacuate, the status of evacuation intention in the sample population, and correlates of evacuation intention. Significance Study results will assist public officials in their efforts to improve evacuation planning when the state is threatened by major hurricanes. Knowing in advance the number of residents who intend to evacuate would allow officials to take appropriate action to insure an orderly and timely evacuation and help reduce emergency response costs (Mozumder, Raheem, Talberth, and Berrens, 2008). The study will contribute to the literature on disast er planning and response. Although many studies have examined evacuation behavior post event very few studies have investigated evacuation intentions prior to an event. Of those studies, none have segmented the population based on their intention to evacuate. This study differs from previous research in two important ways. First, this study focuses on evacuation intention. Second, the analysis will identify segments with coastal populations that are most and least likely to evacuate. Segmentation may allow messaging to be designed and other actions taken that will encourage those who currently do not have any plans to evacuate to reevaluate their position and make preparations in case evacuation is necessary. Identification of the factors associated with the lack of evacuation intention will enable officials to design strategies for overcoming barriers and designing messages to motivate families to prepare for evacuation.

PAGE 25

13 Chapter Two: LITERATURE REVIEW Introduction This chapter begins with an overview of hurricane evacuation research and examines the major predictors of evacuation behavior previously found in the literature. Studies that have examined evacuation intention are then reviewed with the major influences for behavior examined and compared. Theories to explain evacuation behavior are discussed, with an emphasis on the emergent norm and risk perception leading to the warning and response model Finally, statistical techniques used to segment heterogeneous populations into smaller hom ogenous groups are examined. Evacuation Evacuation has been considered a valuable response to disaster since ancient times (Perry, 1985) As such, evacuation behavior has been studied extensively Studies of hurricane evacuation in the United States date back to at least 1953 when J F Rayner published a study of hurricane evacuation from Ocean City, Maryland (Baker, 1979) In 1979, Dr Earl Baker review ed data from four previous hurricane evacuation studies to summarize the state of knowledge at that time regarding evacuation response. Using data from studies of hurricanes Carla, Camille, and Eloise, he combined the 75 variables investigated in those studies into 13 categories These categories consisted of information

PAGE 26

14 source, evacuation advisement, storm watching, belief the storm would hit, expectation of damage, confidence in weather forecasting, recall of forecast information, knowledge about hurricanes, previous hurricane experience, length of residence, site characteristics, demographic charact eristics, and miscellaneous variables Of the four studies reviewed, none of the expected predictors of evacuation consistently turned out to significant ly predict evacuation. Baker maintains this is partially because of inadequate measures of evacuation (Baker, 1979) Baker (1991) reviewed an additional 12 storm studies published between 1963 through 1990. A gain he found inconsistencies in predictors of evacuation, sometimes even between studies of the same storm event Reviewing the storms overall, Baker (1991) found that the most consistent predictors of evacuation were: t he areas r isk level (low lying, flood prone) ; p ublic authorities actions ( evacuation orders) ; r isk level of h ousing (storm resistance, living in mobile homes) ; p erception of personal risk ; and storm characteristics (strength, landfall prediction) F amily structure and the ability of the family to communicate in a disaster are often cited as predictor s of evacuation. Families evacuated during World War II took extreme measures to stay together during evacuation (Bernert and Ikle, 1952). Research in the early 1950s indicated intense anxiety on the part of families th at were separated in evacuation (Bolin, 1976) Families resisted

PAGE 27

15 separation, even in emergency situations Familie s stayed together and sought shelter with relatives if possible when they evacuated ( Drabeck and Boggs 1968) More recent research also found families acted as units, either evacuating together or staying together Families with children were more likel y to evacuate, whereas families with senior citizens were less likely to evacuate. For f amilies without children, the presence of pets has been the most significant predictor of evacuation (Heath, Kass, Beck and Glickman, 2001) The source of evacuation information has been found to influence the evacuation decision. T he lack of an authoritative information source added to the confusion and rumors that proliferated after Hurricane Katrina (Atkins and Moy, 2005) Participants in a qualitative study involving people who did not evacuate from Hurricane Katrina cited confusing recommendations from different authorities as one of the reasons for not evacuating (Elder, Xirasagar, Miller, Bowen, Glover and Piper, 2007) Personal communication with family, fri ends and coworkers has been shown in several studies to have a stronger association with perceived risk than official warnings (Horney, et al. 2010, Eisenman, Cordasco, Asch, Golden and Glik, 2007). In a 1978 study assessing the impact of mass media and print materials on residents knowledge of hurricanes, exposure to television announcements had no impact on the individuals knowledge Written brochures did increase knowledge of the accuracy of statements regarding the definition of storm surge, the number of persons potentially killed by rising water, and the number of miles of coastline that could be damaged by hurricanes Residents who received a brochure were

PAGE 28

16 also significantly more likely to have a planned evacuation route. Residents who heard r adio announcements were actually less likely to correctly define storm surge or correctly estimate how many people were likely to be killed by storm surge. The radio spots improved knowledge of how to locate a local Red Cross or Civil Defense shelter Television images were able to change peoples beliefs about the destructive power of wind and the potential for homes on barrier islands and along the coastline to be destroyed (Christensen and Ruch, 1978) Television images may act as visual cues to action when coupled with evacuation orders. In a survey of New Orleans residents the inclusion of visual images in warning messages increased the likelihood of evacuation ( Burnside, Miller and Rivera 2007) When public officials are aggressive in issuing evacuation notices and disseminate the messages effectively, over 90 percent of the residents of highrisk barrier islands and open coasts evacuate. People hearing or believing they hear official evacuation advisories or orders are more than twice as likely to leave in most locations (Baker, 1991) In some studies age and family structure have been associated with disaster response. T he 2009 Citizen Corps National Survey study found that individuals from 3554 years old were more likely to believe that a natural disaster was likely to occur in their community, but persons over 55 were more likely to know shelter locations and community evacuation routes (FEMA, 2009) In a study of Hurricane Andrew Gladwin and Peacock found that family size and having an elder or children in the family decreased the probability of evacuation, yet other studies found no difference in evacuation behavior among those with

PAGE 29

17 and without elders in the home (Whitehead, et al., 2000; Baker, 1991). There are several agerelated changes in perception, attention, memory, text comprehension, and decision making that may reduce the likelihood of disaster response among those over 65 (Mayhorn, 2005). After Hurricane Katrina, an analysis of the early data showed that vulnerable elderly persons were over represented among the dead (Bour q ue, et al., 2006) In some previous studies, elderly residents in retirement areas were more likely to evacuate than other age groups (Dow and Cutter, 1998). In Florida, a confounding factor for elderly ev acuation may be housing type. There are many mobile home parks in Florida that cater to residents over the age of 55. Residents of mobile homes are ordered to evacuate early in the path of an oncoming storm due to the vulnerability of mobile homes to winds ( Koutnik 2000) Baker (1991) found that mobile home residents are more likely to evacuate than other residents, which may influence the perception of older residents evacuation behavior. Perceived personal risk and subsequent evacuation behavior may also vary by ethnicity (Lindell and Perry, 1992) In the case of Hurricane Andrew African American and Hispanic households were less likely to evacuate than W hite Americans if perception of risk was excluded (Whitehead, et al 2000) When risk percepti on factors were included in the analysis there was no difference in evacuation. The 2009 National Citizen Corps Survey found that African Americans had a higher perception of risk than w hite respondents African Americans were more likely to believe that disaster would strike their community However, African Americans were less prepared for disaster than

PAGE 30

18 W hite Americans (FEMA, 2009) In a study of evacuation response to Hurricane Lili in Texas there was no difference in evacuation decisions by ethnic groups (Lindell, Lu, and Prater, 2005) The Texas study contained a diverse sample, including African Americans and Native Americans though the percentage of Hispanics was only 1% In the 2009 Citizen Corps survey (FEMA, 2009) respondents were asked about barriers to evacuation. The most commonly cited reason (30%) for not preparing was the belief that emergency rescue personnel would find and help them if needed. Other results included 25% who said they had not had time to prepare, and 23% who said they didnt know what to do. In previous surveys conducted in 2007 and in 2009, 17% of the respondents said they did not think preparation would make a difference. Gender income, and access to personal resources have also been identified as predictor s of evacuation. The decision to evacuate or stay in place is complex, taking into consideration multiple factors. There are well documented dangers in riding out a storm, but there are also dangers in staying put. In addition to the monetary cost of evacuation, there is the added danger of crowded roads and potentially hazardous driving conditions. In 2005 a bus carrying elderly residents away from Hurricane Rita caught fire while stuck in gridlocked Interstate 45 traffic south of Dallas. The fire was fed by the residents oxygen tanks, which exploded, killing 24 people (Regnier, 2008).

PAGE 31

19 The evacuation decision may not happen at a single point Rather, every potential evacuation time period prior to the actual hurricane landfall the household makes a choice either to evacuate or to wait one more time period for a revised hurricane forecast (Czajkowski, 2007) In some cases individuals may decide to shelter in place. Most shelter in place recommendations are for hazardous materials incidents or wher e venturing into the open may lead to exposure to potential radiological or biological threats (American Red Cross, 2003) The American Red Cross and the CDC have posted instructions and diagrams for sealing off rooms for short term sheltering during a chemical or radiological event. In circumstances where the event occurs with little or no warning, is of short duration and evacuation could lead to exposure to hazardous materials or severe weather, such as a tornado, sheltering in place has an obvious adv antage over evacuation. Sheltering in place minimizes exposure of the population in the affected zone, is faster to implement in densely populated areas, is easier to implement among the institutionalized public, and requires fewer resources to implement. Sheltering in place is less familiar to the public and termination of sheltering must be controlled to prevent premature exposure to the hazard (South Florida Regional Planning Council (2007) Evacuation s tudy findings have not been consistent. Despit e increased research on emergency preparedness, very few attempts have been made to replicate or build on previous findings (Tierney, Lindell, and Perry, 2001).

PAGE 32

20 Evacuation I ntent ion Relatively few studies have examined peoples intention to evacuat e in th e face of a hurricane. As part of his review of previous storm studies Dr. Earl Baker conducted an experimental pencil and paper exercise to determine how participants would respond in 16 hurricane risk scenarios for Pinellas County, Florida (Baker, 1995 ) In the Baker study the participants were divided into two groups. One group was provided information on hurricane strength and speed and given a hurricane tracking chart showing the location of the hurricane in the Gulf of Mexico relative to the surv ey site. The second group was given the same hurricane strength and speed information and given an identical tracking chart but they were also provided landfall probabilities for Pinellas County and other sites The group given landfall probabilities res ponded as risk managers hoped; as the probability of nearby landfall rose, respondents stated they would evacuate or take other precautions. Baker found that people were able to understand the probability estimates but that the most important factor influencing intent to evacuate was official proclamations or orders. In the review of actual hurricane response studies landfall probabilities had little impact on actual evacuation behavior. After Hurricane Bonnie struck North Carolina in 1998, residents were surveyed to determine their response to Bonnie and a hypothetical future hurricane given different categories of storm and types of evacuation orders Storm intensity was the best predictor of evacuation intention (Whitehead et al. 2000)

PAGE 33

21 Prior to H urricane Katrina in 2004, 1,207 residents of New Orleans were asked if they would evacuate if officials recommended it ( Burnside, Miller and Rivera 2007) P articipants intended to evacuate more often if officials ordered evacuation, if they had previousl y evacuated, felt at risk, or had an evacuation plan in place. In a study of the influence of perceived and actual flood risk on hurricane evacuation, Horney et al. included evacuation intention as a potential modifier of evacuation. The study found no significant association between actual or perceived flood risk and evacuation, but did find a significant association between stated evacuation intention for a future storm and prior evacuation from Hurricane Isabelle in 2003 (Horney et al., 2010). We cannot tell from this study whether the experience evacuating from Isabelle in 2003 influenced the stated response to the survey of intent to evacuate in 2008. Somewhat related to this study, Alrikatti, Lindell, Prater, and Zhang (2006) studied the correlati on between a respondents ability to accurately gauge their risk area on a map and their evacuation behavior. Risk area accuracy was uncorrelated with evacuation intention but was negatively correlated with previous hurricane exposure and evacuation experi ence. In 2001, a survey was conducted of Texas coastal residents that asked about hurricane information sources, evacuation intent, how long they thought it would take to prepare for evacuation, and destination if evacuating (Lindell, et al., 2001 ) The intenti on to evacuate was significantly correlated with proximity to the coast or inland waterways storm strength and previous evacuation experience.

PAGE 34

22 In a follow up study the actual response to Hurricane Lili, a category 4 storm which impacted Texas and Louisiana in October 2002, was compared to the expectations from the 2001 study (Kang, Lindell, and Prater, 2007). Overall there was a 6 5 % agreement between evacuation intention and actual behavior. There was a strong correlation between those who st ated they would not evacuate and actual behavior. There was strong agreement between intention and behavior for the number of vehicles used in evacuation; however the intended and actual destinations in evacuation were quite different. These results are summarized in Table 2 Predictors of Evacuation Intentions and Behavior Many post hurricane studies have examine d factors that explain peoples evacuation behavior after a hurricane, and a number of studies have examined peoples intention to evacuate when recommended by public officials prior to an approaching hurricane The most common factors investigated include demographics, risk level, prior experience and information sources. A comparison of the predictors of evacuation and evacuation intention is shown in T able 3 Logically, p redictors of evacuation intention very closely mimic predictors of actual evacuation behavior

PAGE 35

23 Table 2. Hurricane evacuation expectations and actual behavior in Hurricane Lili Would evacuate Would not evacuate Transporta tion type Destination Expected 40 10 25 28 Actual 26 8 22 15 % Agreement 65% 80% 88% 53.6% Note: The Expected row for Transportation type and Destination indicates the total number of respondents. The Actual row indicates the number who used the Expec ted Transportation type and evacuated to the previously stated Destination. Table 3 Predictors of evacuation and evacuation intention. Evacuation Evacuation Intention Storm factors Landfall probability Risk Level/Flood risk Storm intensity Evac uation orders Evacuation Orders Personal Experience Previous evacuation experience Warning media Information source Risk perception Risk perception Housing type/Live in mobile homes Plan in place Family size Presence of children Presence of older adults Income resources Race Sex Family in communication or together P resence of pets Theor etical Frameworks Three complementary theoretical frameworks have been used to understand evacuation behavior in this study : emergent norm theory, risk perception theory, and a synergistic framework called the warning and response model. This section begins with the development of emergent norm theory from

PAGE 36

24 the roots of social behavioral theory of the early 20th century. I then discuss risk perception theory and the influence of risk communication on risk perception. This is followed by a discussion of the development of the warning and response model using elements from risk perception and emergent norm theories. E mergent n orm theory Theories and models of disaster behavior, including evacuation planning, have a history rooted in theories of crowd behavior As early as the first part of the 20th century Gustave Le Bon proposed that crowds formed a collective mind with transitory but clearly defined characteristics Le Bon proposed that no matter what the individual makeup of the crowd, the transformation from individuals to the collective mind of the crowd makes them act very differently from how the individual would act alone. This transformation takes place through three processes First, the individual forms a sense of power from the mere inclusion in a large group. Second, an individual gives up their personal interest to the collective interest through an almost hypnotic contagion of sentiment Third, a form of suggestibility overtakes the individual and makes the person more susceptible to the collective sentiment Le Bon stated that by the mere fact of being in a crowd, a man descends several rungs in the ladder of civilizatio n. (Widener, 1979) Emergent norm theory was developed by Ralph Turner and Lewis Killian (1957) as another way to explain crowd behavior Turner and Killian viewed crowds as rational and norm governed. When a situation is unstructured or ambiguous and the crowd does not share preexisting expectations about how

PAGE 37

25 they should behave, a new idea of appropriate behavior will emerge. Turner and Killian maintain that crowds communicate mood, imagery and an idea of the kind of action that is appropriate through a process of symbolic interaction. A crowd does not create a condition where there is an absence of culture; rather there is a breakdown of the normal culture of the group. If a sense of urgency is also present new nontraditional behaviors will emer ge. This new behavior reflects the needs of the crowd, but is guided by a behavior pattern that emerges as the situation unfolds (Turner, and Killian, 1957) People are compliant with this new behavior to earn the approval of the group, and thus a new norm emerges The new norm must be specific to the event since there is no formal organization, no obvious leader, and no criteria for membership in the group other than physical presence. Aguirre, Wenger, and Vigo (1998) quote Ralph Turner (1964) in proposing that nontraditional, collective behavior emerges from the crucible of a normative crisis. In the event of a crisis, the normal modes of personal behavior are often replaced by a sense of uncertainty; there is an essentially normless condition. Thomas Drabeck (1968, in Perry, 198 5) discusses the emergent norm as an orientation of behavior Drabeck posits that as people interact in crisis they create a new, emergent norm During a disaster threatened people are still governed by norms; those nor ms are simply different emerging norms of activity rather than established ones Individuals must reexamine their behavior in light of a change in the environment, arriving at a new understanding of their situation (Perry, 1985) Drabeck, in his study on evacuation, proposes: Societies are

PAGE 38

26 composed of individuals interacting in accordance with an immense multitude of norms, i.e., ideas about how individuals ought to behave Our position is that activities of individuals are guided by a normative structure in disaster just as in any other situation In disaster, these actions are governed by emergent rather than established norms, but norms nevertheless. (Drabeck, 1968 in Perry, 198 5) In a somewhat refined definition by Aguirre et al (1998), the emergent norm is viewed as less of a set of rules and more as an emergent revised definition of the situation in crisis People come to feel their collective behavior is appropriate, feasible, timely, permissible, necessary, or duty bound behavior In a disas ter this means an individual will evacuate or take other protective measures such as shelter seeking in conformance with his perception of what is normal for his society This perception is influenced by the social environment surrounding the individual perceiver The individuals will then behave in the way that society expects, or in terms of their own socialization (Luhmann, 1993) Emerging norm s depend upon group communication in a crisis through the milling process Milling is described by Aguirre, Wenger and Vigo (1998) as a form of social interaction that occurs as a crowd interacts to define and adopt new appropriate norms for behavior and find a solution to their collective problem When a crowd coalesces there is no norm governing the behavior of the crowd. There is normally no leader or centralized control The attention of the crowd is drawn towards those that act in a distinctive manner This distinctive behavior is taken as the norm and slowly a new norm that governs behavior emerges As time passes the norm becomes entrenched and there is

PAGE 39

27 pressure against nonconformity Inaction on the part of the crowd is interpreted as being a sign of acceptance of the new norm In a natural disaster this communication of the new norm takes plac e as i ndividuals take cues from the behavior of the community surrounding them, such as observing other neighbors evacuating or boarding up windows Baker (1979) found in his research that the extent of neighborhood evacuation and who neighbors discussed the storm with influenced the decision to evacuate. With the increase in the number of social networking sites and the advent of texting and Twitter, people can share emerging ideas of appropriate behavior with larger numbers of people. Risk p erception. Wh ile the emergent norm approach emphasizes the social processes involved in which information is received, risk perception theory emphasizes the cognitive aspects of an individuals prior personal experience with the hazard, the perceived characteristics of the hazard, and the alternative protective actions of which the individual is aware (Lindell and Perry, 1992) In this view t he individual must answer the following questions : Does the threat really exist? Is protection needed? Is protection feasible? A nswers to these questions will then determine the protective response, if any Risk communication. An important ingredient in risk perception is the communication of risk. A general model of the risk communication process in disasters is thought to

PAGE 40

28 consist of six stages : receiving, understanding, believing, personalizing, information seeking and responding (Sorensen and Mileti, 1991). Receiving and understanding. Receiving a warning message is more than just hearing the information. In order for the message to be received there must be a credible source of information. Mass media is a major source of disaster information but that information is filtered and interpreted through the medium (Sorensen and Mileti, 1991; Perry, 1985). There is a tendenc y to perceive the media as sensationalist and to take media warnings less seriously than might be warranted. (Mileti and Peek, 2000). Believing. Assuming a warning message is received and understood, the individual s must believe the warning is intended for them. Whether or not the individual believes the warning will depend on a number of factors. In the case of hurricane warnings people will first confirm the warning that was received. Confirmation may involve contacting another person or family mem ber to verify the warning was actually intended for the individual. Perry suggests that people are more likely to seek confirmation of warnings when they are received from the media or peers than when received from authority figures or relatives (Perry, 1985). Baker (1979) found that the extent of neighborhood evacuation and with whom persons discussed the storm influenced the decision to evacuate. In a study of volcano and flood evacuation, the main reason cited for leaving was seeing actual evidence of the threat. In contrast, when asked the reason for not evacuating

PAGE 41

29 during the Three Mile Island nuclear accident 38% of the people said they did not see any real danger (Perry, 1985). For a storm or hurricane warning, the individual may look outside to see if the weather looks bad. Personalizing. P ersonalizing refers to an individuals tendency to evaluate personal risk or the risk to their families from the threat. Personalizing may occur by evaluating how close an individual is to the threatened area or how severe the consequences of impact might be. Information seeking. Information seeking occurs when people check for further information by watching news coverage or seeking information over the internet. Information seeking has changed in the digital world. Many people now have access to instant information through the internet and cable weather news. In a survey conducted for the Florida Association of Broadcasters Baker found that most people sought information by watching television progra ms about hurricanes. More than 70% of those surveyed had internet access and more than 50% had visited the National Hurricane Center web site for information. Less than 30% of those surveyed had visited local government websites for hurricane information, and just over 10% had sought information from www. floridadisaster.org, the Florida emergency management website (Baker, 2006). Some researchers have been concerned that because of the need for officials to call for evacuation well in advance of hurrica ne landfall unnecessary evacuation and official false alarms would impact future evacuation. A study of

PAGE 42

30 Hurricanes Bertha and Fran in the Carolinas found that official false alarms and near misses did not impact future plans of residents to evacuate. Instead, the authors found that media use for hurricane warning information, specifically The Weather Channel, influenced evacuation decision making more than official warnings as an evacuation prompt (Dow and Cutter, 1998). Responding. In the final step, if individuals believe the warning is appropriate for them, the threat is real and there is high personal risk, they should respond by taking protective action. The p erception of risk and prior experience. Risk perception is often thought to be shaped by prior experience with natural disaster If we assume a warning is confirmed, past experience would seem to play a part in the belief that danger is imminent, but experience is infrequently cited as a reason for evacuating. Indeed, a study of evacuation response to hurricanes along the Gulf Coast reported that many people who did not evacuate were long time residents of the area and presumably would have had prior experience with hurricanes (Perry, 1985) In studies of Hurricane Lili in Texas and other storms, personal experience with hurricanes had no consistent correlation with evacuation (Baker, 1991; Lindell et al. 2005) Direct experience with hurricanes may increase a persons perception of risk due to a storm, but in Florida there are many more near misses than direct hits by hurricanes. A person who experiences a near miss may think they have experienced a hurricane and thus decrease their perception of risk.

PAGE 43

31 Other studies have reported prior evacuation experience as the single best predict or of evacuation during Hurricanes Hugo and Andrew (Riad, Norris and Ruback, 1999) and a significant predictor for evacuation in New Orleans (Burnside et al. 2007) Reports about evacuation may also influence future intentions After Hurricane Georges passed by the Florida Keys in 1998 there were long delays in reentry due to infrastructure damage and clean up concerns In a survey of residents of Dade and Monroe Counties after the Hurricane, Dash and Morrow (2001) found that those who heard about the long delays for reentry stated they would be less likely to evacuate in the face of a future storm than those who actually experienced the delays Other factors that affect risk perception. Risk perception is not just comprised of the elements of danger but also the emotional content of the risk Peter Bennett says risk perception is heavily influenced by fright factors, conditions that are generally perceived by persons as negative (Bennett, 1999) These conditions include events that are: i nvoluntar y; i nequitably distributed; i nescapable through personal precautions ; u nfamiliar ; m an made; t he cause is hidden and causing irreversible damage;

PAGE 44

32 p articularly dangerous to small children or pregnant women (i.e they endanger future generations) ; i nvol ve a form of death or illness that is particularly dreadful ; p oorly understood by science; and subject to contradictory statements from responsible sources Using these factors it is easy to understand how the risks associated with hurricanes can be underestimated. Hurricanes have become familiar to many people in affected areas the effects are immediate, there is no perceived risk to future generations, victims are considered statistical, most individuals do not consider themselves personally at risk and hurricanes are acts of nature. Each of these decreases public concern and thus the individual perception of risk With a lowered perception of risk people are less likely to respond appropriately when a warning is issued. The w arning and r esponse m odel Lindell and Perry (1992) proposed that evacuation behavior is based upon a combination of personal and situational factors found in risk perception and emergent norm theor ies They call this perspective the Warning and Response Model Fi gure 1 sum marizes this perspective. The warning and response model was used by Lindell and Perry to create the Protective Action Decision Model (PADM) (Lindell and Perry, 2004). This model further defines the third stage of decision making, risk reduction, in term s of protective action search, protective action assessment and protective action implementation. The model strengthens the recognition that at all stages of disaster threat response people act on the

PAGE 45

33 basis of available information. If persons recognize that they do not have adequate information they will begin an information search, beginning with an information needs assessment. They will ask What information do I need to answer my question? This will be followed by Where and how can I find this in formation? and Do I need the information now? Even though the model has been updated, the original variables from the warning and response model remain valid and have been used in other research on evacuation (Horney, et al., 2010). Those variables set forth in the original warning and response model will be used to inform this research.

PAGE 46

34 Situational Factors Physical Cues Social Behavior Risk Communication Recipient Characteristics Prior Beliefs Experience Education Adaptive Plan Personality Traits Personal Resources Social Context Family context Kin relations Community Involvement Ethnicity Age Socioeconomic Status Risk Identification: Does the threat really exist? Risk Assessment: Is protection needed? Risk Reduction Is protection feasible? Protective Response Figure 1. The Warning and Response Model Adapted from Behavioral Foundations of Community Emergency Planning by M. Lindell and R. Per ry, 1992, p. 135. Emergent norm theory contributes the social factors of family, kin relations, ethnicity, age, and group membership included in the warning and response model The social interaction of groups and kin create the milling process that dist ributes information about warnings to a population. A person

PAGE 47

35 with a strong kin network or network of community ties is likely to receive additional warnings from kin and community and more likely to believe the warning (Perry, 1979). Communication with peers and hearing from friends who are evacuating during a disaster helps form the new, emergent norm in response to the disaster situation. Knowing that others in the community are taking action will influence the individual to take action as well. Risk perception theory contributes recipient characteristics and situational factors such as previous experience, locus of control, individual personality traits, and prior beliefs. T he three basic questions Does the threat really exist?, Is protection needed?, and Is protection feasible? are interdependent: the answer to one impacts the decision process for the others If there is no perception of threat the individual will not take protective action. Even if the threat exists the individual must be convinced that the threat applies to them personally. If a person is not convinced that impact is certain and the person is within the danger area, the person will not take protective action (Lindell and Perry, 1992) If a real threat exists and the person is in the danger area but has delayed evacuation until landfall is only hours away, the person may believe that evacuation is no longer effective and so will take no action rather than risk getting stuck on the road in the storm T he entire deci sion process is influenced by the individual social context, personal characteristics and situational factors outside of the individual The social context includes the family structure the network of kin relationships in the family, the level of community involvement of the family, and demographic

PAGE 48

36 variables of ethnicity, age, and socioeconomic status Persons with a close family may also refuse to evacuate before a storm unless they know that the family members are taken care of and safe. Individual c haracteristics of the person, including experience, education, and having a current adaptive plan influence the perception of the threat and the decision to take action. Previous experience will influence the perception of the threat and whether or not pr otective action is feasible or useful. The source of warning information, observation cues such as weather, or the observable behavior of others will determine whether a person believes the warning and takes protective action or not Based on this synth esis of emergent norm and risk perception theories Lindell and Perry (1992) proposed a warning and response model composed of 12 primary propositions with 17 subpropositions. The warning and response propositions apply to multiple hazard situations that might result in evacuation from man made conditions or natural disaster, including hurricanes The Harvard study dataset contains information that can be used to test some of these propositions. These pr opositions and sub propositions are shown in Table 4

PAGE 49

37 Table 4 Warning and response propositions. 1. The greater the individuals belief in the warning the higher the level of protective action motivation and thus the probability of protective response a.) Prior experience with similar disasters increases the likelihood of developing a warning belief b.) Receipt of a warning from a credible source increases the degree to which the threat is perceived as real 2. The higher the level of perceived personal risk, the higher the level of protective action motivation and thus the probability of protective response. a.) Prior experience with similar disasters is positively related to the level of perceived personal risk b.) Ethnic majority status is inversely related to a higher level of perceived personal risk c.) Socioeconomic status is inversely related to the level of perceived personal risk 3. The more specific an individuals adaptive plan, the higher the probability of protective response a.) Prior experience with similar disasters increases the chance that an individual will develop an adaptive plan 4. If family members are together at the time of warning or otherwise accounted for, the less likely they are to perceive the existence of barriers to implementation and the greater is the probability of evacuation. 5. Individuals characterized by an external locus of control are less likely to engage in any type of protective action a.) Membership in an ethnic minority group increases the chance that an individual will have an external locus of control 6. The greater the frequency of contacts with kin the greater the number of warnings an individual will receive. 7. The greater the frequency of contacts with kin, the more likely one is to receive additional waning information through these contacts. 8. Membership in an ethnic minority group is positively related to the nature and frequency of contacts with kin. 9. The greater the level of community involvement, the greater the number of warnings an individual is likely to receive. a.) Membership in an ethnic minority group is positively related to level of commu nity involvement. b.) The lower the individual's socioeconomic status, the lower the level of community involvement. 10. The greater the level of community involvement, the more likely one is to receive additional warning information from these contacts. 11. Members hip in an ethnic minority group is positively correlated with lower perceived credibility of authorities. 12. Membership in an ethnic minority group is associated with lower socioeconomic status Adapted from Behavioral foundations of community emergency planning. M. K. Lindell & R.W. Perry, p. 136, Hemisphere Publishing, 1992.

PAGE 50

38 Although Lindell and Perry proposed their model to address evacuation behavior specifically, I propose the same personal and situational factors will influence evacuation planning and i ntent in the face of a hurricane threat Based upon the warning and response propositions I hypothesize that the intention to evacuate when recommended by public officials prior to an approaching hurricane will be influenced by the following factors : per ceived personal risk, prior experience with similar disasters, prior evacuation experience, ethnicity, socioeconomic status, and having a family contact plan. The Harvard data set contains information that can be used to test these hypothes e s. Identifying Population Segments at Risk Part of the confusion in evaluating evacuation behavior results from officials use of a single strategy to address the needs of a large homogeneous sample. This approach fails to recognize that people who intend to evacuate d iffer along many lines from those who do not. Segmentation of a single heterogeneous population into smaller more homogenous segments makes it possible to develop strategies that better meet each segment s needs and concerns. To some extent segmentation occurs in the State of Florida through disparate treatment of certain populations. Mobile home residents are segmented through early evacuation orders, household pet owners are referred to pet friendly shelters, and those medically dependent who are not acutely ill, are sent to Special Needs Shelters. These are voluntary segmentations that are difficult to quantify. Even special needs client lists mandated under Florida Statute 252.355, are often inaccurate. For example, Leon County has an

PAGE 51

39 estimated 2,100 persons dependent on oxygen and only 500 persons on the Special Needs Client list ( Florida Department of Health, 2010) Segmentation has been used in commercial marketing for many years and has been called the foundation for the success or failure of public communication efforts (Slater, in Maibach and Parrot, 1995) Market segmentation as a marketing strategy was introduced by Wendell Smith in 1956 (Wedel and Kamakura, 2000) Smith (1956) stated: Market segmentation involves viewing a heterogeneous market as a number of smaller homogeneous markets, in response to differing preferences, attributable to the desires of consumers for more precise satisfaction of their varying wants. Audience segmentation can improve both ef ficiency and effectiveness in strategic planning (Andreasen, 1995; Kotler and Andreasen, 1991 ) Mardburg (1996) studied audience segment ation in a sample from a 1993 Norwegian study of perception, motivation, coping, knowledge and belief in information from nuclear accidents and other radiation sources He concluded that t here is need for further research in the area of segmentation of high risk catastrophe situations Segmentation is optimal when it identifies subgroups in a population that are mutually exclusive, exhaustive, measurable, accessible, substantial, and differentially responsive. A segment base is mutually exclusive when each segment is separated from every other segment conceptually It is exhaustive when every member of the group is included in some segment M easurability refers to the degree to which a segment can be measured. Accessibility is the

PAGE 52

40 degree that to which a segment can be reached and communicated with. Substantiality refers to the size of the segment It must be large enough to be worth pursuing Finally, segments in the segmentation scheme must respond differently ( Kotler and Andreasen, 1991) A common approach to segmentation is to use demographic characteristics of the audience to identify subgroups within a population. This approach is only valid when demographics are correlated with the behavior in question (Slater, 1995) Other methods include segmentation based on audience perceptions, motivations, purchasing habits, geography, etcetera (Weinstein, 1994) Various secondary data sourc es for these perceptions, motivations and habits are available for use in market segmentation. These include multiple trade journals, directories, governmental and private computer data bases and statistical sources, such as the U.S Census There are al so commercial data sources available, for a fee, which specialize in segmentation. These include PRIZM and ClusterPlus from the Nielsen Company (Nielsen Company, 2010) Depending on the marketing need these sources can be a part of the overall marketing plan. Many methods have been used to perform segmentation, but they can be classified into two basic categories : a priori and post hoc (Wedel and Kamakura, 2000) Segmentation is called apriori when the number and type of segments are determined in advance Post hoc segmentation occurs when the type and number of segments are unknown and are determined by the results of the analysis

PAGE 53

41 Cluster analysis is a post hoc segmentation tool that merges populations or objects together to maximize the groups sim ilarities Cluster analysis requires the designation of the number of clusters to be formed, and can be used to find structures in data, but cannot provide explanations for the structures. In his Norwegian segmentation study, Mardberg (1996) used latent profile analysis as a statistical clustering method. Segmentation is also classified by whether descriptive or predictive statistical methods are used. Descriptive methods are used across a single set of segmentation bases with no distinction between dependent or independent variables Descriptive methods for apriori variables include contingency tables and loglinear models. Predictive methods are used to analyze the association between two sets of variables, with dependent variables to be explained by a set of independent variables Predicting membership in a group can be pursued through discriminant analysis when there are two or more mutually exclusive groups. However discriminant analysis requires that you know group membership for some cases i n order to derive the rules for classifying the remaining cases. When there are only two groups of cases multiple regression is closely related to discriminant analysis H owever multiple regression categorizes the linear relationship of independent var iables to the dependent variable; it does not produce clusters of the population based upon the interaction of the independent variables. (Norusis, 2008)

PAGE 54

42 Predictive post hoc methods include automatic interaction detection types and latent class regressi on. These include the use of chi square automatic interaction detection for categorical dependent variables Chi Squared Automatic Interaction Detection (CHAID) Automatic Interaction Detection (AID) was developed in 1963 at the University of Michigan as an alternative to the traditional regression approach to criterion based modeling (Magidson, 1994) AID is an ad hoc method, also called binary tree analysis , which splits a population based on the best predictor variable. It continues to split each of the groups until no predictor can be found that meets the selection condition. AID was widely used in marketing research until it was found that the method capitalized on chance occurrences to the extent that segments identified in AID did not validate against other samples Gordon Kass developed the Ch i square A utomatic I nteraction D etection (CHAID) algorithm in 1978 to address this issue in AID (Neville, 1999) CHAID is an exploratory approach that can be used to study the relationship between a dependent variable and a potentially large number of independent variables. CHAID modeling selects a set of predictors and their interactions that optimally predict the dependent measure. CHAID modeling produces a tree diagram that shows how major segments f ormed from the interaction of the independent variables predict the dependent variable. CHAID follows three stages : merging, splitting, and stopping. CHAID merges or splits variables based on the chi square statistic Variables that are significantly di fferent in their ability to predict the dependent variable area split; those that are

PAGE 55

43 not significantly different are merged. The exhaustive CHAID method performs a more thorough merging and testing of the independent variables; the merging process continues to merge any similar pairs until only a single pair remains. The Exhaustive CHAID function also allows the resplitting of merged categories to obtain a better fit. When all of the subgroups have been analyzed or have too few observations, the procedure will stop CHAID also adjusts for the use of multiple tests by using the Bonferroni multiplier The Bonferroni multiplier adjusts the pvalue of the series of comparisons to control the probability of false positives CHAID is most powerful when th e dependent variable is dichotomous The base CHAID method can be used for polytomous dependent variables, but there is a loss of power T he method was modified by Jay Magidson in 1992 to extend to ordinal polytomous variables without the loss of power ( Magidson, 1994) Predictor categories on CHAID are merged in accordance with specific predictor types; free, monotonic, and float Monotonic variables are ordinal in nature and merged by CHAID if they are next to each other For example, if one of the predictor variables is income and categories of income variables are divided as shown in T able 5 below the categories next to one another could be merged in the analysis Categor ies 1 and 2 could be merged together or categor ies 3 and 4 could be merged together, but categor ies 1 and 3 or 2 and 4 w ould not

PAGE 56

44 Table 5 Example Income Categories Category Income 1 $10,000 $14,999 2 $15,000 $19,999 3 $20,000 $24,999 4 $25,000 $29,999 Free variables may be combined whether they are next to each other or n ot These variables contain no natural order Descriptive variables such as occupation or race are treated as free variables. Floating variables are treated like monotonic variables except for the last category If T able 5 contained a category for refused or unknown, it would be treated as a floating variable. The final category is combined with any other variable that it is most similar to in terms of the relationship to the dependent variable. CHAID is similar to cluster analysis in that it divides the population into subgroups CHAID, however makes use of a specific dependent variable to form the subgroups Cluster analysis may or may not be predictive of a dependent variable, but CHAID is designed for prediction. CHAID results in a tree diagram and a gains table. The tree diagram can be thought of as a trunk with smaller and smaller branches The initial tree trunk represents all the participants in the study and the branches represent the smaller groups of participants based on the predictor variables relationship to the dependent variable. The trunk of the tree is known as the root node. The branches are known as subnodes, which may act as parent nodes of further

PAGE 57

45 child nodes When a branch can no longer be divided, the final segments are called terminal or end nodes In the analysis process, the investigator can set parameters to determine the minimum size of the parent and child nodes The gains table consists of these end node segments ranked from high to low and summarized relative to their response rate. A sample tree diagram from Magidson (1994) is illustrated in Figure 2. This diagram illustrates the split among magazine subscription responders and nonresponders. Since the dependent variable is dichotomous only one percentage is shown in the diagram, the percentage of responders. The root node here breaks into four subnodes based upon household size. The household size variable in Magidson (1994) consisted of six categories depending on the number of persons in each household; one, two, three, four, five or more, and unknown. Further variables include the occupation of the head of household and gender of the head of household. Occupation is defined as White Collar, Blue Collar, Other and Unknown. The categories for 2 and 3 person households and 4 and 5 or more person households were merged together in the analysis to create four categories. In the second row occupation is merged for Blue Collar, Other and Unknown. The sum of the values in the child nodes will equal the value of the parent node. Thus the sum of the values in the second row of Household Size will equal the total, 81,040, and the sum of the values in the third row of the Occupation nodes will equal the value of the parent node 23 Household size category (1,758 + 14,374 = 16,132).

PAGE 58

46 The tree diagram suggests that larger households have a better response rate. Overall the diagram indicates those households with 2 or 3 persons where the head of the household is a white collar worker have the best response rate, 2.39%. The lowest response rate, 0.81%, comes from households of unknown size with a male head of household. Figure 2. Sample CHAID tree diagram Total Sample Y Responder 1.15 % N= 81 ,040 1 1.09% 25,384 23 1 .52% 16,132 ? 0.87% 33,326 W 2. 39% 1,758 B O ? 1 .42% 14 ,374 M 0.81% 25,531 F 1.08 % 7,795 45 1.92% 6,198 Household Size Occupation Gender

PAGE 59

47 Unlike cluster analysis, CHAID explores the data without a predetermined number of segments or clusters to be formed. CHAID also has the advantage of allowing the independent variables to be measured at different levels (nominal, ordinal, or interval). Several software packages offer decision tree algorithms, including SAS and SPSS. SPSS incorpo rated, an IBM company, established Predictive Analytics Software (PASW) as a market segment in 2003. The 2010 version of PASW from SPSS includes CHAID, Exhaustive CHAID, QUEST, and C & RT classification and regression tree modeling (SPSS, 2010). The Exha ustive CHAID algorithm is an extension of the original CHAID algorithm that uses an exhaustive search procedure to merge any similar category pairs until only a single pair remains. Based on the selected variable types and the prior use of CHAID in market ing segmentation, this study will us e CHAID to segment the data. CHAID disadvantages include its use of a stepforward model fitting method when not in automatic mode. As in other stepforward regression fitting models results depend on the order in whic h the variables are entered into the model. The automatic function of CHAID addresses this by simultaneously adding all of the variables at once. CHAID also allows a forced option, where a variable can be forced into different stages for consideration in the analysis. Since CHAID uses the chi squared statistic it is assumed to follow the chi squared distribution. This assumption requires a large sample size to ensure the validity of the test though some authors have used samples of 500 cases with satis factory results (van Diepen and Franses, 2006 ) An important concern is the

PAGE 60

48 danger of over fitting the data. The training sample may contain random variations not in other samples that cause variations in the tree when new data are supplied to the algori thm (Neville, 1999) Over fitting is detected by applying the tree to new data, the test sample, and comparing the outcome. When over fitting occurs it can be addressed by pruning the tree and removing extraneous variables that may cause interference i n the model. Segmentation and disaster preparedness D isaster research has looked for predictor variables in behavior, but only one study could be located that used segmentation for the purposes of disaster communication. This exception wa s the use of the Stages of Change M odel ( Prochaska and DiClemente, 1983) to describe groups of respondents in the 2007 and 2009 Citizen Corps National surveys (FEMA, 2009) Various demographic characteristics of the population were examined and individuals were classifie d by their place in the Stages of Change continuum For example, men were found to be more likely than women to have been prepared for the last 6 months ( 39% compared to 31% ). Recommendations were made in the study for further research into the needs of people in each stage or audience segment and to develop strategies to reach those who do not prepare for disasters. Following the above recommendation, the primary purposes of this research are to identify predictors of evacuation intention and to segment the audience in terms of their intent to evacuate in the face of hurricane warnings. The research answer s the following questions:

PAGE 61

49 What proportion of the coastal population intends to evacuate when recommended by public officials prior to an approaching hurricane? What factors are associated with the intention to evacuate when recommended by public officials prior to an approaching hurricane? What factors are useful in identifying meaningful segments of people who intend to evacuate when recommended by public officials prior to an approaching hurricane? This research will test the primary conditions of the warning and response model as they relate to evacuation intention: T he higher the level of perceived personal risk, the higher the probability of evacuation intention; Prior experience evacuating during disasters is positively related to the level of perceived personal risk and evacuation intention; Ethnic majority status is inversely related to the level of perceived personal risk ; Socioeconomic stat us is inversely related to the level of perceived personal risk ; and Having family members together at the time of warning, or otherwise accounted for increases the probability of evacuation.

PAGE 62

50 Chapter Three: M ETHODS This chapter describes the research approach used in the study. It includes a restatement of the study purpose and research questions to be addressed, and an overview of the study design, the research population, the dataset and the analysis plan. This chapter ends with a discussion of the study limitations. Research A pproach The primary purpose of the research is to identify predictors of evacuation intention and to define segments of the population that differ with respect to their intent to evacuate. The research will attempt to answer the questions: What proportions of coastal populations intend to evacuate when recommended by public officials prior to an approaching hurricane? What factors are associated with the intention to evacuate when recommended by public officials prior to an approaching hurricane? What factors are useful in identifying meaningful segments of people who intend to evacuate when recommended by public officials prior to an approaching hurricane? This research will also test the following hypotheses based on the war ning and response model:

PAGE 63

51 Prior experience evacuating during hurricane threats will increase the probability of evacuation intention; Prior experience evacuating during hurricane threats will be positively related to the level of risk perception; The higher the level of risk perception, the higher will be the probability of evacuation intention; Having family members together at the time of warning, or otherwise accounted for, will increase the probability of evacuation intention; Ethnic majority status will be inversely related to the probability of evacuation intent ; and Socioeconomic status will be inversely related to the probability of evacuation intent. Secondary D ataset This study uses secondary data from a survey by the Harvard School of Public Heal th, Project on the Public and Biological Security Hurricane in High Risk Areas study The study was designed and initially analyzed by researchers at the Harvard School of Public Health. Harvard contracted with International Communications Research, an independent research company, to conduct interviews on hurricane readiness among a representative sample of 5,046 noninstitutionalized respondents age 18 and older in coastal counties of Texas, Louisiana, Mississippi, Alabama, Georgia, North Carolina, South Carolina and Florida. Interview s w ere conducted by telephone from June 18 to July 10, 2007, in all counties within twenty miles of the coastline for each of these states The

PAGE 64

52 survey instrument consist ed of 109 questions plus combined measures of race and income. The survey included questions about risk perception, current preparation, preparation knowledge, previous experience, and evacuation intention. In addition, questions regarding age, sex, family makeup, and housing were included (see Appendix A) To compensate for nonresponse bias inherent in telephone surveys the sample data were weighted to represent the total adult population in the region as a whole based on the most recent U.S. Census available. In addition, random digit dialing was us ed as well as call backs staggered over various times of day and days of the week Although eight states were included in the survey, this research concentrates on the Florida data. In Florida, the survey included 42 counties. Data w ere entered into SPS S for analysis by the Harvard School of Public Health. The Harvard School of Public Health was contacted directly for access to the data and they supplied the SPSS data file. An initial review of the data indicates that there are no missing data elements or out of range data. The basic demographics of the survey population are given in Table 6

PAGE 65

53 Table 6 Survey population demographic description Total respondents 5046 Age % 18 29 19 30 49 35 50 64 22 65+ 18 Race White 57 Black 15 Hispanic 19 Asian 1 Other 3 Gender Male 49 Female 51 Income Less than $40,000 34 $40,000 $75,000 22 $75,000 $100,000 10 $100,000 + 13 A nalysis by the Harvard School of Public Health indicated that 46% of the surveyed population lived in a community damaged by hurricanes during the past three years and 22% of the population left their homes due to hurricanes in the last three years When asked if they would evacuate in the future if told to do so, 31% responded they would not leave. The reasons given for not intending to evacuate when told are shown in Figure 3

PAGE 66

54 Reasons would not evacuate 75% 56% 36% 33% 27% 0% 10% 20% 30% 40% 50% 60% 70% 80% Home is well built Roads will be too crowded Evacuating would be dangerous Posessions might be stolen Would not leave pet Figure 3 Reasons respondents would not evacuate when told Counties in the data set are designated by Federal Information Processing Standards (FIPS) county code numbers. The FIPS county code is a five digit number used to identify counties within the United States and U.S. possessions. The first two numbers of the code identify the state and the last three identify the county. The numbers, for the most part, follow the alphabetic listing of counties by state. The last three digits are usually odd numbers so that additional counties can be added alphabetically without disrupting the sequence. Using the FIPS code and U.S. census data, the surveyed counties were identified by name. Census data were also used to determine the demographics of the 42 Florida counties in the survey, including race, age category, education, income, population, and population density ( see A ppendix B ) The survey population was compared with the general population of the surveyed counties to explore sample bias

PAGE 67

55 Protection of Human Subjects The data set was received from the Harvard School of Public Health without any individual identifiable information. Individuals are identified by case number only with no li nk between the case number and any identifiable information. The study protocol received an expedited review and was approved by the University of South Florida Institutional Review Board, Social and Behavioral Sciences Division. Analysis Plan Creation of dataset To obtain the Florida sample, data were sorted by State and a separate working file was constructed containing only the responses obtained from people living in Florida. The Harvard data were weighted, resulting in a reported number of 3,045 Flo rida cases in SPSS, however when weighting was removed there we re 1,006 unweighted Florida interviews. The analysis was performed on the unweighted data. A cross check was conducted to insure only Florida data were included by sorting the file by FIPS number and verifying only Florida FIPS numbers are included. All variables were coded consistently to represent the characteristic measured. Yes and No answers were coded 0 = No and 1 = Yes while ordinal categorical values were coded 1 = Not at All, 2 = N ot Too, 3 = Somewhat and 4 = Very Missing data issues were explored using the SPSS Missing Values function. Only one variable, Income Summary had significant numbers of missing values. In reviewing the pattern of missing values there was no discernable predictor From previous income level questions in the dataset

PAGE 68

56 the majority of the missing values were in categories over $40,000, therefore the missing data were replaced with the existing value 19 = $40,000 or more, (unspecified). Dependent variable. For purposes of this study, the dependent variable was evacuation intention. E vacuation was defined as an orderly vacating of the normal place of residence to seek shelter in another location. Evacuation intention was measured through the question: 1. If government officials said that you had to evacuate the area because there was going to be a major hurricane in the next few days, would you leave the area or would you stay? a. I would leave the area b. I would stay c. Depends d. Dont know e. Refused Independent v ar iable s Independent v ariables were selected for analyses to assist in answering the proposed research questions including demographic characteristics, measures of risk perception, prior hurricane experience, prior evacuation experience, and family disaster planning. Ethnicity and socioeconomic status were measured through individual questions of ethnicity, education, and income. These measures are used to test the warning and response hypotheses Measures of risk perception, prior experience with hurric ane threats, prior experience with evacuation, and family accountability are shown in Table 7

PAGE 69

57 Table 7. Measures of risk perception, prior experience and family accountability. Risk perception How worried are you that a major hurricane will hit your community during the next 6 months? Overall, how prepared are you if a major hurricane were to strike your community during the next 6 months? If a major hurricane were to hit your community and for whatever reason you did not leave your home, how confid ent are you that you would be rescued if you needed to be? Thinking about where your home is located, how likely is your home to be flooded or damaged due to wind in a major hurricane? Do you think your home would withstand a major hurricane of Category 3 or higher without significant damage? How long have you lived in your community? Prior experience with evacuation Thinking back over the past three years was your community threatened or hit by a major hurricane, or not? o Because of this hurricane, did you leave your home where you lived, or did you stay in your home? Prior experience with hurricanes Was your community damaged by this hurricane, or not? Was there major flooding associated with this hurricane in your community or not? During and immediately following this hurricane, was there a time when you had any of the following problems? o You didnt have enough fresh water to drink o You didnt have enough food to eat o You didnt have the prescription drugs or medicines that you needed o You were threatened by violence o You needed medical care and couldnt get it o You had problems getting gas to evacuate o You had other problems evacuating o You didnt have enough money o You had problems because you were disabled or chronically ill o You had problems caring for a dis abled, chronically ill or elderly member of your household o You suffered from heat exhaustion due to power failure o You were injured as a result of the storm How would you rate the response of government and voluntary agencies to the problems created by this storm?

PAGE 70

58 Family accountability Has your family agreed on a phone number outside the region that all members of your immediate family could call in the event of a hurricane if you are unable to communicate, or havent you done that? Has your family agreed on a place you could meet after a hurricane is over if you got separated and could not go back home, or havent you done that? Relationships between evacuation intention and other variables. Bivariate relationships between evacuation intent and the possible influential variables in Table 7 were calculated through chi square tests and correlation coefficients as appropriate. Bivariate analysis and logistic regression w ere used to address the question, What factors are associated with the intention to evac uate when recommended by public officials prior to an approaching hurricane? In addition, logistic regression w as used to address the hypotheses: Prior experience evacuating during hurricane threats will increase the probability of evacuation intention; Prior experience evacuating during hurricane threats will be positively related to the level of risk perception; The higher the level of risk perception, the higher will be the probability of evacuation intention; Having family members together at the time of warning, or otherwise accounted for, will increase the probability of evacuation intention; Ethnic majority status will be inversely related to the probability of evacuation intent; and

PAGE 71

59 Socioeconomic status will be inversely related to the probability of evacuation intent. The independent variables are categorical, so Spearmans and Kendalls correlations were used. Cramers V and Phi were calculated to test the strength of association of the calculations. Cohens rule of thumb, as shown in Table 8, w as used to categorize effect size ( Alfonso, 2007) Categorical variables were dummy coded into dichotomous variables for analysis. Logistic regression was run through SPSS statistical software using a forward stepwise analysis with an entry value of p =.0 5 and removal value of p =.10. Table 8. Cohens Effect Size Interpretation Rules of thumb Cohens d Correlation Coefficient Odds Ratio Cramers V Small .20 .10 1.50 df = 1; .10 < V < .30 df = 2; .07 < V < .21 df = 3; .06 < V < .17 Medium .50 .25 2. 50 df = 1; .30 < V < .50 df = 2; .21 < V < .35 df = 3; .17 < V < .29 Large .80 .40 4.30 df = 1; V > .50 df = 2; V > .35 df = 3; V > .29 Note: The guideline for chi square tests of independence with 3 degrees of freedom was used for tests with greater t han three degrees of freedom. Socioeconomic status (SES) is primarily gauged on family income, education level, occupation and status in the community. In this survey only income and education level are available, so a proxy for socioeconomic status was constructed based on those variables. Andrew Beveridge and Susan Weber of Queens College constructed an interactive table used to determine SES based

PAGE 72

60 on income and education levels (New York Times, 2007) This table is used in the analysis to create an S ES level and that SES level is then compared with evacuation intention Identification of meaningful segment s of people who intend to evacuate. Segmentation analysis was conducted using the SPSS Statistics software CHAID function to answer the final questi on: What factors are useful in identifying meaningful segments of people who intend to evacuate when recommended by public officials prior to an approaching hurricane? The significance levels for splitting and merging categories can be set by the invest igator, the default level is .05. By default CHAID uses a Bonferroni multiplier to adjust significance values. Cramers V was calculated at each tree node to determine effect size. Where effect size did not at least meet Cramers rule of thumb for smal l affect size the node was not included in the segmentation tree (Alfonso, 2007) CHAID was used to determine subpopulations based upon interactions with the dependent variable of intent to evacuate and significant independent variables from previous analys is. The dependent variable, intent to evacuate, is dichotomous and the independent variables are nominal or ordinal. T he analysis was run using only those variables that were significant in previous analysis The default parent and child node size setti ng for CHAID is 100 cases for the parent node and 50 for a child node, but because the Florida dataset only contains 1 ,000 records, the minimum parent node size was set at 50 and the minimum

PAGE 73

61 child node size was set to 2 5 (The Measurement Group, 1999 2005). Segmentation analysis was conducted using the automatic growth function in CHAID. The software provides the risk estimate and classification accuracy of the model. Classification accuracy is determined through a cross tabulation of the actual categories and the predicted categories of the cases. The data tree and gains tables produced were reviewed in the context of the w arning and r esponse model by Lindell and Perry (1992) to determine variables useful in segmenting the population in terms of evacuation intention. A description of each segment was developed from the analysis. CHAID allows for two forms of validation; cross validation and split sample validation. Split sample validation is often used to determine segmentation reliability (Magidson, 1994) In split sample validation, the data set is randomly split in two, based on a specified percentage of cases. The analysis is first performed on a training sample to create a model. The model is applied to the second test sample and the results are compared across both samples. A good model will produce a test tree that closely follows the training sample tree. The trees accuracy is judged by the performance on the test data (Rodeghier, 2007). Caution must be used if the sample size is small. Small samples may yield poor models since there may not be enough cases in some categories to grow valid trees. In contrast cross validation divides the sample into a number of subsamples, or folds. The user can specify the number of folds up to 25. The analysis is then run on the main sample, excluding the data from each subsample in turn. Multiple tree diagrams are produced; the first tree excludes

PAGE 74

62 the cases in the first sample fold, the second tree is based on all of the cases except those in the s econd sample fold, and so on. For each tree misclassification risk is estimated by applying the tree to the subsample excluded in generating it. Considering the size of the dataset a 50/50 split sample validation was used. The produced segments were co mpared with the warning and response model conditions for evacuation to determine if the variables in Table 7 were validated as predictors of evacuation intent. Variables that were found statistically significant in CHAID analysis were used in multivariate analysis. At each branch of the tree Cramers V statistic was calculated to determine the strength of association of the variables used. Logistic regression was then used to determine those variables most influential in evacuation intention. Categori cal variables were dummy coded for this purpose. The regression results were then compared to the warning and response model. Limitations The Harvard project used random digit dialing to obtain a representative sampling of the study area. Participation in tele phone surveys may lead to coverage bias due to the growing number of cell phone only and cell phone primary use adults. Typical random digit dialing does not include cell phone exchanges. Cell phone only and primarily cell phone users appear different than other telephone usage groups even controlling for demographic differences (Lee, Brick, Brown, and Grant, 2010). Unfortunately, no information was available about nonrespondents in this sample. This nonresponse bias and exclusion of individual s relying exclusively on cell phones prevents the study from having the

PAGE 75

63 power of a truly random sample, making it unwise to generalize to the overall population. In addition, there are potential question wording and ordering effects that may have influenc e d responses (Harvard, 2007). For example, t he individual answer s may be influenced by the use of leading wording such as, Was that because----. This study used secondary data for analysis that did not include all of the possible variables in the warni ng and response model. Issues of locus of control, kin relationships and warning receipt are not included as variables and thus their influence was not considered. Household size was not included as a variable and so household income could not be used to create a true representation of socioeconomic status. All of these issues affect the ability of the results to be generalized to the nonrespondent population.

PAGE 76

64 Chapter Four: RESULTS Introduction This chapter provides a review of the research question s and purpose of the research. A summary description of the population and the dependent and independent variable frequencies follows. The next section describes the relationship between the individual predictor variables and the dependent variable of ev acuation intent followed by an exploration of the interrelationship between the predictor variables. The next section presents CHAID analyses of the significant predictor variables. This was followed by multivariate analysis of the significant predictors and the dependent variable. The chapter concludes with a summary of the results of the analyses applied to the research questions and the tene ts of the Warning and Response model. Research p urpose. The primary purpose s of the research w ere to identify pr edictors of evacuation intention and to define segments of the population that differ with respect to their intent to evacuate. The research was guided by the following questions: What proportion of Florida coastal populations intend to evacuate when recommended by public officials prior to an approaching hurricane?

PAGE 77

65 What factors are associated with the intention to evacuate when recommended by public officials prior to an approaching hurricane? What factors are useful in identifying meaningful segments of people who intend to evacuate when recommended by public officials prior to an approaching hurricane? This research also tested the hypotheses that: Prior experience evacuating during hurricane threats will increase the probability of evacuation intenti on; Prior experience evacuating during hurricane threats will be positively related to the level of risk perception; The higher the level of risk perception, the higher will be the probability of evacuation intention; Having family members together at the time of warning, or otherwise accounted for, will increase the probability of evacuation intention; Ethnic majority status will be inversely related to the probability of evacuation intent; and Socioeconomic status will be inversely related to the probabil ity of evacuation intent. Description of the P opulation The Florida population sample varies from the general population in several ways. A comparison of the general survey population demographics as opposed to the Florida population sample is shown in Table 9

PAGE 78

66 The Florida population sample contains a higher percentage of persons over 65 than in other states which is consistent with t he US Census Bureau findings that a higher percentage of Floridians are over 65 years old as compared to the US population i n general (US Census Bureau, 2010) M inority populations, Blacks and Hispanics, are under represented in the Florida survey. The 2009 US Census estimates the Florida population at 16% Black and 21.5% Hispanic, whereas only 8% Black and 10% Hispanic were actually interviewed. Table 9 Florida survey population demographics compared to the tota l survey population demographics R espondents Florida Survey Total Survey 1006 5046 Age % % 18 29 9 19 30 49 29 35 50 64 29 22 65+ 33 18 Race White 74 57 Black 8 15 Hispanic 10 19 Asian 1 1 Other 3 3 Gender Male 53 49 Female 47 51 Income Less than $40,000 33 34 $40,000 $75,000 27 22 $75,000 $100,000 13 10 $100,000 + 20 13

PAGE 79

67 The majority of the Florida population (67.6%) interview ed had completed some level of post high school education. Of those responding, 38.7% of those interviewed had completed college and/or some level of post graduate training. Dependent variable frequencies. When asked if they would evacuate in the future i f told by government officials to do so, 59.1% of Floridians surveyed said they would leave, 35.2% said they would not leave and 5.6% said it would depend. This compares to the 31% of the total surveyed population who said they would not leave if told to do so, not surprising given Floridians exposure to four hurricane landfalls during the 2004 hurricane season. Although Floridi a ns are more likely to evacuate than those in other states, the reasons given for not intending to evacuate when told are the same as for the general survey population. The top reasons given by those interviewed in Florida are shown in Figure 4.

PAGE 80

68 Figure 4. Reasons Florida respondents would not evacuate when told Independent variable frequencies. Among Floridians surveyed, 66.9% lived in a community that was threatened by a hurricane in the last three years and 83.3% of those survey participants lived in a community that was damaged by a hurricane. Among those who had been threatened by a hurricane, 26. 6% had left their homes. Risk perception, prior experience with hurricane threats, prior experience with evacuation, and family accountability are all concepts that relate to the proposed research questions. Questions related to these concepts are shown in Table 7. The frequencies for the Florida population of questions in the survey that address these concepts are shown in Tables 10, 11 and 12. 29% 34% 34% 55 % 79% Would not leave pet Possessions might be stolen Evacuating would be dangerous Roads will be too crowded Home is well built

PAGE 81

69 Table 10. Risk perception variables frequencies Very % Somewhat % Not too % Not at all % Don't Know % How worried are you that a major hurricane will hit your community during the next 6 months? 11.1 34.3 31.3 22.7 .4 Overall, how prepared are you if a major hurricane were to strike your community during the next 6 months? 38.3 46.9 7.6 6.1 1.1 If a major hu rricane were to hit your community and for whatever reason you did not leave your home, how confident are you that you would be rescued if you needed to be? 34.7 37.0 15.0 9.8 3.0 Thinking about where your home is located, how likely is your home to be fl ooded or damaged due to wind in a major hurricane? 18.8 31.3 31.5 16.1 2.0 <1 year 1 5 years 6 10 years 1120 years >20 years How long have you lived in your community? 5.6 % 25.7 % 19.4 % 20.4 % 28.8 % Yes % No % Don't Know Do you think your home would withstand a major hurricane of Category 3 or higher without significant damage? 59.9 27.2 12.7 Family accountability was measured through two questions asking if the family has determined a place to meet and a phone number outside of the region i n case they are separated. The frequencies for these questions are shown in Table 11.

PAGE 82

70 Table 11. Family accountability frequencies Yes % No % Don't Know % Has your family agreed on a phone number outside the region that all members of your immediate fam ily could call in the event of a hurricane if you are unable to communicate, or haven't you done that? 49.7 49.6 0.7 Has your family agreed on a place you could meet after a hurricane is over if you got separated and could not go back home, or haven't you done that? 32.5 65.6 1.9 Prior experience with hurricanes was measured through a series of questions answered by those whose community had been threatened or damaged by a storm. These variables are summarized in Table 12. Table 1 2 Prior hurricane exper ience Yes % No % Don't Know % Thinking back over the past three years was your community threatened or hit by a major hurricane? 65. 7 32. 6 1.7 Because of this hurricane did you leave your home where you lived? 26.6 71.8 1. 5 Was your community damaged by this hurricane?* 83.3 16.2 0 .5 Was there major flooding associated with this hurricane in your community? 33.8 65.5 0.7 During and immediately following this hurricane, was there a time when y ou didn't have enough fresh water to drink? 10.2 89.6 0.2 During and immediately following this hurricane, was there a time when y ou didn't have enough food to eat? 7.8 92.0 0 .2 D uring and immediately following this hurricane, was there a time when y ou didn't have the prescription drugs or medicines that you needed? 8.2 91.6 0 .2 During and immediately following this hurricane, was there a time when y ou were threatened by violence? 2.4 97.3 0 .4 (continued)

PAGE 83

71 During and immediately following this hurricane, was there a time when y ou needed medical care and couldn't get it? 3.5 96.2 0.4 During and immediately following this hurricane, was there a time when y ou had problems getting gas to evacuate? 31.3 68.5 0.2 During and immediately following this hurricane, was there a time when y ou had other problems evacuati ng? 12.0 87.6 0 .4 During and immediately following this hurricane, was there a time when y ou didn't have enough money? 13.3 86.2 0 .5 During and immediately following this hurricane, was there a time when y ou had problems because you were disabled or chronically ill? 5.1 94.7 0 .2 During and immediately following this hurricane, was there a time when You had problems caring for a disabled, chronically ill or elderly member of your household ? 6.5 93.3 0.2 During and immediately following this hurri cane, was there a time when You suffered from heat exhaustion due to power failure? 17.1 82.5 0 .4 During and immediately following this hurricane, was there a time when You were injured as a result of the storm? 4.2 95.6 0.2 Excellent Good Fair Poor How would you rate the response of government and voluntary agencies to the problems created by this storm? 24.5 39.8 22.4 13.2 Frequencies are for those who responded Yes to the question Thinking back over the past three years was your community t hreatened or hit by a major hurricane, or not? R elationships W ith E vacuation I ntention Relationships between the dependent variable of interest and the independent variables were calculated in SPSS using the Crosstabs function and binary logistic regressi on. The Crosstabs function computes a chi square test for each variable pair and produces measures of association. For analysis purposes, variables were considered significant at the p < .05 level. Spearman correlation coefficients were calculated for t hose variables with ordinal responses. Bivariate analysis was used to address the question, What factors

PAGE 84

72 are associated with the intention to evacuate when recommended by public officials prior to an approaching hurricane? In addition, logistic regress ion was used to test the hypotheses: Prior experience evacuating during hurricane threats will increase the probability of evacuation intention; Prior experience evacuating during hurricane threats will be positively related to the level of risk perception; The higher the level of risk perception, the higher will be the probability of evacuation intention; Having family members together at the time of warning, or otherwise accounted for, will increase the probability of evacuation intention; Ethnic majority status will be inversely related to the probability of evacuation intent; and Socioeconomic status will be inversely related to the probability of evacuation intent. Prior experience with evacuation was significantly associated with future evacuation intent Whereas 84% of those who had previously evacuated said they would leave if told to do so, only 54% of those who had not previously evacuated said they would leave, 2 (N= 657, 2) = 45.48, p < .01, Cramers V = .266. There was no statistically signific ant association between evacuation intent and the remaining variables in Table 12. Risk perception variables were inconsistent in their relationship with intended evacuation. Being worried that a hurricane would strike in the next six

PAGE 85

73 months a nd having a home likely to be flooded or damaged in a hurricane were positively correlated with evacuation intent at the small effect level. Having a generator, length of residence in the community, and believing the home would withstand a category 3 or higher storm were correlated with staying in place when ordered to leave, also at the small effect level (see Table 13) Living in an evacuation zone, self declared level of preparedness, and confidence in being rescued were not significantly associated with evacuatio n intention. Table 13. Risk perception associations with evacuation intention Correlation df p Cramers V How worried are you that a major hurricane will hit your community during the next 6 months? .114 a 22.75 6 < .01 0.11 Overall, how prepared are you if a major hurricane were to strike your community during the next 6 months? .066 b 7.7 2 6 0.26 0.06 Do you have a generator? .133 a 17.12 2 <.01 0.13 If a major hurricane were to hit your community and for whatever reason you did not leave your home, how confident are you that you would be rescued if you needed to be? .093 b 11.44 6 0.08 0.08 How long have you lived in your community? .142 a 26.59 10 <.01 0.12 Thinking about where your home is located, how likely is your home to be flooded or damaged due to wind in a major hurricane? .132 a 19.41 6 < .01 0.10 Is your home located in an e vacuation zone or not, or don't you know if it is in an evacuation zone? .004 b 6.21 4 0.184 0.06 Do you think your home would withstand a major hurricane of Category 3 or higher without significant damage? .125 a 19.55 4 < .01 0.10 a Correlations signifi cant at p < .01 b Correlations not significant ? Risk perception was also inconsistent with prior evacuation experience. B elieving a home was likely to be flooded or damaged in a hurricane or having a

PAGE 86

74 home located in an evacuation zone was positively correlated with previous evacuation. Believing the home would withstand a category 3 or higher storm was significantly associated with not having evacuated previously All correlations were significant at p < .01. The significant relationships with prior evacuation experience are shown in Table 14. Table 14. Risk perception associations with prior evacuation experience df p Cramers V Thinking about where your home is located, how likely is your home to be flooded or damaged due to wind in a major hurricane? 14.89 3 < .01 0.15 Is your home located in an evacuation zone or not, or don't you know if it is in an evacua tion zone? 27.44 2 < .01 0.21 Do you think your home would withstand a major hurricane of Category 3 or higher without significant damage? 9.54 1 < .01 0.13 All correlations were significant at p < .01 Family accountability was measured through two quest ions ; Has your family agreed on a phone number outside the region that all members of your immediate family could call in the event of a hurricane if you are unable to communicate, or haven't you done that? and Has your family agreed on a place you could meet after a hurricane is over if you got separated and could not go back home, or haven't you done that? Neither of these two variables was significantly associated with evacuation intention at the p < .0 5 level

PAGE 87

75 Previous research has found an assoc iation between evacuation behavior and the presence of pets in the home (Drabeck, 2001; Heath, et al. 2001), so this relationship was tested as well. The presence of pets in the home was significantly associated with evacuation intention ( 2 (N= 940, 1 ) = 6.57, p = .0 1, Cramers V = .084) though the size of Cramers V failed to meet the minimum criteria for a small effect size set out in Cohens rule of thumb (see Table 8) The relationships between evacuation intent and a variety of demographic measures w ere examined. The demographic variables of age, race, income, and education were transformed into categories for analysis. The presence of children was negatively associated with evacuation intention; when children were in the home respondents were more likely to say they would remain in their homes when ordered to evacuate. The relationship between evacuation intention and age was also negative. The intent to evacuate decreased as age increased. There was an association with race, but the correlation coefficient did not meet the criteria for a small effect level ( r = .09) There was no significant association with education, income, gender or SES. The categories and their relationships with evacuation intent are shown in Table 15.

PAGE 88

76 Table 15. Demogr aphic characteristics with evacuation intention. Variable Correlation df p Cramers V Age 1829; 3049 5064 65+ .143 a 24 3 < .01 0.1 6 Race White Black or African American Asian American Hispanic Some other race .088 a 12.21 4 .02 0. 12 Income Less t han $15,000 $15,00 but less than $20,000 $20,000 but less than $25,000 $25,000 but less than $30,000 $30,000 but less than $40,000 Under $40,000 (unspecified) Over $40,000 (unspecified) $40,000 but less than $50,000 $50,000 but less than $75,000 $75,000 but less than $100,000 $100,000 or more .007 b 7.33 10 0.69 0. 10 Education Less than 8 th Less than high school High School Technical or vocational after high school Some college College graduate Post graduate training .017 b 11.37 6 0.08 0. 11 SES Top 20% Upp er middle Middle Lower Middle Bottom 20% .006 b 3.51 4 .48 0. 07 Children Yes No .110 a 11.13 1 < .01 0.1 1 Gender Male Female .049 b 2.27 1 0. 13 0.0 5 a Correlations significant at p < .01 b Correlations not significant

PAGE 89

77 The variables that were found to sig nificantly interact with evacuation intention are summarized in Table 16 Table 16 Variables significant in chi square and correlations Question How worried are you that a major hurricane will hit your community during the next 6 months? Do you have a g enerator? Do you or any other household members have any pets in your home, such as dogs, cats, birds and the like? Because of this hurricane did you leave your home or did you stay in your home? Thinking about where your home is located, how likely is your home to be flooded or damaged due to wind in a major hurricane? Do you think your home would withstand a major hurricane of Category 3 or higher without significant damage? How long have you lived in your community? Race Are there any children und er the age of 18 living in your household? What is your age? (18 29; 30 49; 50 64; 65+) These variables were used to construct a correlation matrix (see Appendix C) to examine any internal relationships that might impact evacuation intention. Where ques tions are interdependent correlation coefficients are not calculated For example, Question 21, If you had to evacuate because of a hurricane, do you have a place you can go where you can take your pet, or not? was dependent on a positive answer to ques tion 20, Do you or any other household members have any pets in your home, such as dogs, cats, birds and the like? Both of these questions had a significant association with evacuation intention, question 20, 2 (N=9 40, 1 ) = 6.5 7 p < = .0 1 Cramers V= 0.084 and question 21, 2 (N= 502, 1 ) = 31. 05. p < .01, Cramers V= 0. .249 but because question 21 has is only answered when there is a yes response to question 20, correlations between the two variables are not computed.

PAGE 90

78 The only other large correlation, Spearmans .488, p <.01, was between age and the presence of children under the age of 18 in the home. Though considered large in Cohens effect size interpretation rule of thumb (see Table 8) the value is not large enough to create confounding. All other correlations were in the small range in accordance with Cohens rules of thumb. A separate analysis was conducted to examine the influence of previous hurricane experience on evacuation intention. Living in a community that had been threatened by a hurri cane was not significantly associated with evacuation intention ( (N=924) = .008, p = .928, Cramers V = .003) and surprisingly, neither was living in a community damaged by the hurricane ( (N=645) = .386, p = .534, Cramers V = .024) Personal experience variables such as problems getting fresh water, food, medicines, medical care, gas or cash were not significantly assoc iated with evacuation intention. Having been threatened by violence, problems because of a disability, injury as a result of the st orm and problems caring for a disabled household member were not significantly assoc iated with evacuation intention. P revious evacuation experience was significantly associated with evacuation intention, ( N=640) = 45.49, p < .01 Cramers V = .274. Logistic regression of variables in bivariate analysis. Regression analysis was used to explore the relationship between evacuation intention and variables in bivariate analysis In order to conduct logistic r egression evacuation intention was converted to a binary variable by recoding the Depends response as missing To verify that the missing values

PAGE 91

79 would not significantly impact the regression analysis, a separate logistic regression analysis was conducted where the response Depends values were recoded into I would stay and I would leave There was no significant difference in the results; however the Hosmer and Lemeshow Chi square goodness of fit statistic was better for the final model with the fi rst analysis. Therefore results are reported using the variable with Depends recoded as missing. The variables included in the initial logistic regression analyses were all of the variables of interest from Tables 10, 11, and 12. Categorical variables were dummy coded into dichotomous variables for analysis. Logistic regression was run through SPSS statistical software using a forward stepwise analysis with an entry value of p =.05 and removal value of p =.10. Variables were entered into the model if the significance value was less than .05 and removed if the significance value was greater than .10. Significance levels for the Beta values are calculated as well as the Exp (B), which in logistic regression is equivalent to the odds ratio (OR) of the event, the change in the dependent variable for a one unit change in the independent variable (Tabachnik, 2001). In this regression I was interested in those who said they would evacuate if told to do so, therefore Stay was coded 0 and Leave was coded 1. An odds ratio that is greater than one indicates an increase in the odds of evacuation intention, while an odds ratio less than one shows a decrease in the odds of evacuation intention for each unit increase in the independent variable. Confidence interv als for the odds ratio

PAGE 92

80 values are also calculated in the analysis. The model that best fit the regression equation included the variables in Table 17 Table 17 Variables in the regression model Question Overall, how prepared are you if a major hurrica ne were to strike your community during the next 6 months? Do you have a generator? Because of this hurricane did you leave your home or did you stay in your home? How would you rate the response of government and voluntary agencies to the problems created by this storm? How likely is your home to be flooded or damaged due to wind in a major hurricane? Do you think your home would withstand a major hurricane of Category 3 or higher without significant damage? Income Summary What is your age? (18 29; 30 49; 50 64; 65+) The results of the regression analysis are shown in Table 18 The variables that interacted significantly with evacuation int ention in the regression are evacuation experience (qn32), government and agency response (qn35), likelihood o f flood or damage (qn37), age, and presence of a generator (qn3h). Having a generator reduced the odds of evacuation intention by half ( OR = .447, p < .01). Government and agency response was only significant when the response was rated poor ( p < .05) Likelihood of flood or damage was only significant when damage was not very likely (qn372) or at the reference category, very likely ( p < .05) The largest odds ratio occurs with prior evacuation experience ( OR = 4.99 ) Someone who had previously evacuated was five times as likely to evacuate if told to do so by authorities. The reference category for age was 65+. It appears that younger people are more likely to say they would evacuate if told to, with those 30 to 49 (AgeCat 2) almost three times as l ikely to

PAGE 93

81 evacuate when told as those over 65 ( OR = 2.78, p < .01) The variable for those ages 50 to 65 did not reach the level of significance set in the analysis. Table 18 Variables significant in the regression model. B S.E. Wald df Sig. Odds Ratio 95 % C.I. for OR Lower Upper Generator .806 .251 10.271 1 .001 .447 .273 .731 Previous evacuation 1.607 .351 20.929 1 .000 4.988 2.506 9.931 Government response a Poor .952 .431 4.885 1 .027 .386 .166 .898 Fair .230 .358 .411 1 .521 .795 .394 1.604 Good .272 .318 .731 1 .393 1.313 .703 2.451 Home likely to be damaged b Not at all .517 .459 1.266 1 .261 .596 .242 1.468 Not very .812 .343 5.594 1 .018 .444 .226 .870 Somewhat .004 .349 .000 1 .990 1.004 .507 1.991 Age c 18 29 .880 .501 3.084 1 .079 2.410 .903 6.434 30 49 1.021 .341 8.946 1 .003 2.776 1.422 5.419 50 64 .350 .329 1.133 1 .287 1.419 .745 2.704 a Reference category is Excellent b Reference category is Very c Reference category is 65+ CHAID A nalyse s CHAID analysis was conducted to explore the interactions between evacuation intention and measures of risk perception, hurricane experience, family accountability and demographics found in Tables 10, 11, and 12. CHAID was also used to explore interactions between evacuation intention and variables in Table 18 that were significant ly associated with evacuation intention. At each node of the tree Cramers V was calculated to gauge the effect size of the interaction. Split sample validation was used to test the accuracy of the model. The data set was randomly split into two approximately equal samples; one

PAGE 94

82 sample was used for training to develop the model, while the other was used for testing the classification accuracy of the model The proportion of cases correctly classified was calculated for the training sample and the test sample. Using split sample validation reduced the size of the data set, so the s ettings for parent and child nodes were set at n = 5 0 and n = 25 respectively (The measurement group, 19992005) CHAID uses a recursive partitioning process using chi square values. Data were successively split into parent and child nodes based on a predetermined chi square significance value. When accuracy can not be improved, partitioning stops (Rodeghier, 2007). The default value of < .05 was maintained for the analysis in order to explore all potential relationships The analysis was conducted using SPSS software, which uses Bonferroni adjustment to correct for the use of multiple tests and avoid alpha inflation CHAID analysis with all variables of interest The general model with all predictors from Table 17 is shown in F igure 5. The analysis began with a training sample of 480 cases. In the test sample, 18% of the population had left their homes due to a previous hurricane, 46% had stayed in their homes, and 35% had not been threatened by a hurricane in the last 3 years. CHAID analysis identified interactions with prior evacuation experience, owning a generator, age, and owning pets. The best predictor of evacuation intention in this model was prior evacuation experience, with a medium effect level ( p <.001 Cra m ers V = .2 34: see Figure 5) Prior evacuation experience was divided into three distinct groups: 1) Stayed in my home; 2) L eft

PAGE 95

83 my home where I lived; and 3) representing those who had not been threatened by a major hurricane in the last three years Those who had stayed in their homes split on the presence of a generator (yes, no). Owning a generator creates a term inal node, but not owning a generator splits on age (<= 5064, >5064). Those who had not been threatened by a hurricane in the last three years (missing) split on the presence of a pet in the home (see Figure 5) In Figure 5 the segment most likely to evacuate was comprised of those who had previously evacuated; 81% of those who had previously evacuated said they would evacuate if told to do so. The segment least likely to evacuate when told to consisted of those who did not evacuate in a previous stor m, do not own a generator and are 65 or over (41.3%, n=46). Age influences evacuation intention; 68% of those under 65 (<=5064) who do not own a generator and have no previous evacuation experience sa id they would leave the area compared to 41% of those over 65 (>5064) who do not own a generator and have no previous evacuation experience who sa id they would leave the area. Among those without exposure to a hurricane in the last three years, the presence of pets influenced evacuation intent with 64% of t hose without pets intending evacuation compared to 57% of those with pets. Those who had stayed in their homes in a previous storm were more likely to say they would leave if they did not own a generator (58.3%, n=127). The segment most likely to leave t he area when told to do so, after those with previous evacuation experience, was comprised of those under 65 years old, who do not own a generator and stayed in their homes in a previous storm (67.9%, n=81). The

PAGE 96

84 effect level of these interactions is shown in Table 19. The overall classification accuracy of the model was 68% for the training sample and 65% for the test sample. This level of classification accuracy overall is low. One method of improving classification accuracy is to prune the tree of ext raneous variables in the analysis (Neville, 1999) In Figure 6 a CHAID tree is created using only variables significant in logistic regression. The overall classification accuracy of the model is improved using this reduced variable set.

PAGE 97

85 Figure 5 Segm entation of evacuation intention with all regression variables

PAGE 98

86 The CHAID analysis overall confirm ed the relationships found in the logistic regression in Table 18. Both demonstrated an association with prior evacuation experience, owning a generator, and age. The CHAID analysis show ed an additional association between evacuation intention and pet ownership ( p < .01, Cramers V = .23) not indicated in the logistic regression and dropped the association with the response of government and volunteer agenc ies Table 19. Effect size values for segmentation of evacuation intention with all regression variables Relationship Node 2 Cramer's V E vacuation e xperience with intention 0 0.2 3 Generator with experience (stayed) 1 6.874 0. 18 Age with Generato r 4 10. 923 0. 29 Pets with experience (missing) 3 8.266 0.2 3 CHAID analysis with logistic regression variables. A second CHAID analysis was conducted using t he variables from T able 1 7 that were used in the logistic regression analysis (see Figure 6). This CHAID analysis began with a training sample of 48 8 cases and a test sample of 4 52 cases responding to the dependent variable If government officials said that you had to evacuate the area because there was going to be a major hurricane in the next few days, would you leave the area or would you stay? In the test sample, 6 6 % said I would leave the area and 34% said I would stay. In the test

PAGE 99

87 sample 18.8% of those surveyed had left their homes due to a previous hurricane. The CHAID analyses identifi ed significant interactions between evacuation intention and previous evacuation experience, presence of a generator, and age. The best predictor of evacuation intention in this model remained prior evacuation experience, with a medium effect level ( p <.001, Cramers V = .2 17: see Figure 6). Prior evacuation experience was again divided into three distinct groups: 1) those that had stayed in their homes; 2) those that had left their homes; and 3) those who had not been threatened by a major hurricane in the last three years (). In this model, Left my home where I lived and < missing > create a terminal node, but "Stayed in my home split on the presence of a generator Owning a generator splits on age, as did not owning a generator. In Figure 6 the best predictor of evacuation intention was previous evacuation; 79.5% (n = 78) of those who had previously evacuated said they would evacuate if told to do so. The next segment most likely to evacuate when told to consisted of those did not evacuate in a previous storm, do not own a generator and are less than 65 years old (72%, n=93). The segment least likely to evacuate when told to consisted of those did not evacuate in a previous storm, do not own a generator and are age 65 or over ( 37.8%, n= 37). Age mediated the relationship between evacuation intention and the presence of a generator for those who previously stayed in their homes when threatened by a hurricane. For those who do not own a generator, 72% of those under 65 (<=5064) sa id they wou ld leave the area compared to 38% of those over 65 (>5064) For those

PAGE 100

88 who do own a generator 63.6% of those under 50 ( <= 30 49) would leave the area compared to 42% over 50 (> 3049) Older residents were more likely stay if told to evacuate because of a hurricane. T he relationship with prior evacuation experience and evacuation intention was in the medium range ( 2 = 21.294, p <.001, Cramers V = 217, df = 2). The relationship between evacuation intention and the presence of a generator was in the sm all range ( ( 2 = 10.882, p <.001, Cramers V = .2 29, df = 1) Only 18.8 % (n= 8 5 ) of the test population had previously evacuated. The effect level of these interactions is shown in Table 20. The overall classification accuracy for the model was 65.4% for the training sample and 69.9% for the test sample. Table 20. Effect size values for segmentation of evacuation intention with significant regression variables Relationship Node 2 Cramer's V Experience with intention 0 0.2 2 Generator with experi ence 2 10.882 0. 23 Age with Generator (No) 4 7.132 0. 2 3 Age with Generator ( Yes ) 5 6.203 0.2 8 In comparison, the CHAID analysis with all of the variables (Figure 5) was very similar to the model with the variables from the logistic regression analysis (Figure 6.) Both models initially split on previous evacuation experience. Evacuation experience split into Left my home where I lived Stayed in my home; and (see F igures 5 & 6 ). The value corresponds to those people who answered N o to the question Thinking back over the past three years was your community threatened or hit by a major hurricane, or not?

PAGE 101

89 (Table 12) Those who stayed in their homes further split on the presence of a generator. Those who did not have a generat or then split on age, with those over the age of 65 most likely to stay in their homes The best predictor of evacuation in both models was previous evacuation experience. The difference in the tree structure between the two models is that node 3, , split on pets in the model with all of the model variables whereas the model with only variables significant in the logistic regression did not have that branch, but split again on node 5, the presence of a generator, into an additional age branch. T he second model did not contain the variable for pets because that variable was not significant in the logistic regression analysis. Overall classification accuracy for the model with variables significant in logistic regression increased to 69.9% for the test sample.

PAGE 102

90 Figure 6. Segmentation of evacuation intention with significant regression variables

PAGE 103

91 CHAID and evacuation experience. The inclusion of extraneous variables in CHAID can cause the model to change. T he variable Because of this hurri cane did you leave your home or did you stay in your home? was missing 335 cases; those who were not threatened by a hurricane in the last three years. To determine if there was any interference by including previous evacuation experience a separate analysis was run with all of the regression variables with the same parameters as the overall model but without this variable included in the model. Once evacuation experience was removed from the model significant interactions only occurred with the presenc e of a generator and age (see Figure 7 ) All of the relationships were in the small effect size range. When previous evacuation experience was eliminated from the analysis, the presence of a generator and age less than 49 became the best predictor of evacuation intention. The overall classification accuracy for the model was slightly lower than the model with evacuation experience at 62% for the training model and 63% for the test model A separate analysis was conducted with only those cases that were not threatened by a hurricane in the last three years. With this limited data set, significant interactions occurred with the worry that a major hurricane would hit the community in the next 6 months and the presence of pets in the home. All of the relationships were in the medium effect range. The relationships with age and presence of a generator were no longer significant in the model. Classification accuracy was 67% for the training sample and 57% for the test sample. Such a large difference in ac curacy of the model may be an artifact of

PAGE 104

92 the smaller sample size. Using split sample validation resulted in a training sample of only 15 3 cases and a test sample of 169 cases. Due to the small sample size resulting from a split sample validation, cross validation was used in an effort to utiliz e the entire sample of 327 cases. Using cross validation the results had a higher classification accuracy of 60% but a tree structure segmented on different values than the split sample tree. With the larger cross validation sample, the tree split initially on concern about the ability of the home to withstand damage and length of residence in the community These results seem to support the concerns with CHAID and small samples. Conversely, the results of running the analysis with only those cases whose homes were threatened by a major hurricane (n = 631) bore a closer relationship to the results from the total sample. The classification tree split initially on evacuation experience, with those who had stayed in the home splitting again on age and then on the presence of a generator (see Figure 8), similar to the main sample. Classification accuracy was 69.5% for the training sample and 70.6% for the test sample. This model is shown in Figure 8.

PAGE 105

93 Figure 7 Segmentation of evacuation intention without evacuation experience variable.

PAGE 106

94 Figure 8 Segmentation of evacuation intention with only those cases threatened by a hurricane in the last three years.

PAGE 107

95 Regression a nalys e s of CHAID variables Regression analysis was used to explore the relationship between evacuation intention and variables found significant in the CHAID analysis. The variables significant in CHAID analysis were determined by examining the terminal nodes on the classification trees. Va riables included in the logistic regression analyses were age, presence of a generator, and prior evacuation experience In this analysis all of the variables entered were significant (Table 2 1 ) As in the initial analysis, age does contribute to the model; however t he age category 50 to 64 versus 65+ does not contribute to the model after controlling for the other variables. Those 30 to 49 were more than twice as likely to evacuate as those over the age of 65. Prior evacuation experience remains the best predictor of evacuation intention. Table 2 1 CHAID va riables entered into the regression model. B S.E. Wald df Sig. Odds Ratio 95% C.I. for OR Lower Upper Presence of a generator .719 .185 15.088 1 .000 .487 .339 .700 Prior Evacuation 1.537 .241 40.507 1 .000 4.649 2.896 7.462 Age 18 29 .840 .351 5.729 1 .017 2.317 1.164 4.611 Age 30 49 .984 .236 17.334 1 .000 2.674 1.683 4.250 Age 50 64 .345 .227 2.304 1 .129 1.412 .905 2.203 *The reference category for Age is 65+ The variables that wer e significant in the logistic regression analysis we re prior evacuation, a ge, and presence of a generator The terminal nodes in the CHAID analysis were the same variables. In the regression and in the CHAID

PAGE 108

96 analysis in Figure 5 the model indicates younger residents and those without a generator are more likely to evacuate. Goodness of fit statistics indicate d that overall the model wa s good. When the Hosmer Lemeshow test value is not significant, this indicates the model adequately fits the data. The Hosmer Lemeshow test value for the model was 077. The classification table for the model in logistic regression gives a predicted number and percentage correct value. A model is considered a good fit if there is at least a 25% improvement in prediction over the chance prediction rate. In the model classification table the baseline model simply takes the most numerous category and predicts all of the values will be in that category. In this analysis the model therefore predicts 100% of those surveyed w ould leave. This gives an overall correct prediction of 62.1%. The actual percentages were 62.15% would leave and 37.85% would stay. The chance prediction rate in the model is found by taking the sum of the squared marginal percentages .62152 + 37852 = 386 + 143 = .5 29. A 25% increase over chance would be 66 19% and the full model prediction was 69.8%, therefore the observed hit rate indicates a good model Summary of F indings This study provides a description of the population and the predictors of evacuation intention for the coastal population of Florida. The study population was representative of the State of Florida as a whole except for an under representation in minority populations. In this study 59.1% of the population intended to evacuat e if told to do so by authorities when a major hurricane

PAGE 109

97 approaches Among Floridians surveyed, 66.9% lived in a community that was threatened by a hurricane in the last three years and 83.3% of those lived in a community that was damaged by a hurricane. Among those who had been threatened by a hurricane, 26.6% had left their homes. Most of the population surveyed felt they were somewhat or very prepared for a hurricane (85.2%) and thought their homes would withstand a category 3 or higher storm without major damage (59.9%). They were confident that if a hurricane did strike they would be rescued (71.7%). In addition, a majority were not too or not at all worried that a major hurricane would strike in the next 6 months (54%). Using bivariate analyses a number of factors interacted with evacuation intent ion The strongest association with evacuation intent was prior evacuation. Additionally, age, the presence of children in the home, race, length of residence in the community, concern about a future hurricane, presence of a generator, concern about flooding or wind damage, and belief the home would withstand a major hurricane were also associated with evacuation intent. These relationships are summarized in Table 22

PAGE 110

98 Table 2 2 Factors associated wi th evacuation intention 2 p Cramers V Prior evacuation 45.48 < .01 0.266 Age 24 < .01 0.16 Presence of children 11.13 < .01 0.11 Race 12.21 .0 2 0.12 Length of residence in the community 26.59 < .01 0.12 Concern about a future hurricane 22.75 < 01 0.11 Presence of a generator 17.12 < .01 0.13 Concern about home flooding or wind damage 19.41 < .01 0.1 Belief the home would withstand a major hurricane 19.55 < .01 0.1 Logistic regression analysis was conducted to explore the strength of interactions with evacuation intention and the variables from the bivariate analysis. In logistic regression, prior evacuation was the strongest predictor of evacuation intention (OR 4.99, p < .01). Other predictors included the presence of a generator, concer n about flooding or wind damage, age, and previous experience with government and voluntary agencies. Table 23 summarizes the logistic regression analysis. Table 23 Summary of logistic regression Sig. Odds Ratio 95% C.I. for OR Lower Upper Generator .001 .447 .273 .731 Prior evacuation .000 4.988 2.506 9.931 Prior governmental response ( poor ) a .027 .386 .166 .898 Home likely to be damaged ( Not very ) b .018 .444 .226 .870 AgeCat ( 30 49 ) c .003 2.776 1.422 5.419 a Reference category is Excellent b Reference category is Very c Reference category is 65+

PAGE 111

99 Due to the strength of the association between evacuation intention and prior evacuation, interactions with prior evacuation and significant variables from the bivariate analysis were explored. Prior evacuation experience was significantly associated with concern about damage from flooding or wind, living in an evacuation zone, the belief the home would withstand a major hurricane and length of residence in the community. These relationships are s hown in Table 24. Table 24 Risk perception variables associated w ith prior evacuation experience df p Cramers V Concern about flooding or wind damage 14.89 3 < .01 0.15 Home in an evacuation zone 27.44 2 < .01 0.21 Belief the home could withstand a major hurricane 9.54 1 < .01 0.13 Length of residence in the community 11.78 5 .0 4 0.14 CHAID was used to determine if there were meaningful segments of Florida coastal residents who intend to evacuate. The best predictor of evacuation was previous evacuation. The segment of the population most likely to evacuate when told to by authorities a re under 65, do not own a generator, and stayed in their homes in a previous hurricane. From CHAID analysis the variables in Table 2 5 were significant in segmentation.

PAGE 112

100 Table 2 5 Factors significant in segmentation. 2 p Cramers V Prior evacuation 21.29 < .01 0.2 2 Presence of a generator 10.88 < .01 0.2 3 Age (without a generator) 7.132 .0 2 0.23 Age (with generator) 6.20 .0 4 0.28 The variables significant in the CHAID analysis were used in a logistic regression model to examine the strength and di rection of association of the variables. The logistic regression model supports the CHAID model, with the terminal nodes in CHAID equivalent to the significant variables in the logistic regression; prior evacuation experience, the presence of a generator and age.

PAGE 113

101 Chapter 5: Discussion Introduction This chapter begins with a review of the purpose of the research and the research questions. An overview of the study methods and a summary of the research findings are provided. The results are discussed as they relate to the original study hypotheses as well as previous disaster research on evacuation behavior and evacuation intention. Limitations of the study are discussed along with implications for further research. Plans for disseminating the resul ts of the study are also presented. Purpose of the R esearch The primary purpose of this research was to identify predictors of evacuation intention and to define segments of the population that differ with respect to their intent to evacuate. The study was also intended to identify the proportion of Florida coastal populations that intends to evacuate when recommended by public officials prior to an approaching hurricane the factors associated with the intention to evacuate, and factors useful in identify ing meaningful segments of people who intend to evacuate when recommended by public officials prior to an approaching hurricane. The study focused on populations in Florida residing within 20 miles of the coast. The study looked at relationships between evacuation intent and previous evacuation behavior, prior

PAGE 114

102 hurricane experience, risk perception, residential factors, and demographic factors Segmentation analysis was used to identify factors associated with the intent to evacuate and logistic regression was used to test the strength and direction of association. Overview of the S tudy M ethod This study used secondary data from the Harvard School of Public Health, Hurricane Readiness in High Risk Areas study. The study consisted of telephone interviews with 5,046 noninstitutionalized persons age 18 and older in coastal counties of Texas, Louisiana, Mississippi, Alabama, Georgia, North Carolina, South Carolina and Florida. Interviews were conducted by telephone from June 18 to July 10, 2007, in all counties within twenty miles of the coastline for each of these states. Surveys for the State of Florida were segregated and used in this analysis, resulting in a total study sample of 1,006 surveys from 42 counties. Univariate statistics were used to desc ribe the Florida population and to determine frequencies of experience and evacuation intention. Bivariate statistics and logistic regression were used to explore associations with evacuation intention, previous evacuation behavior and multiple predictors A correlation matrix was constructed to determine significant interactions between the predictors from bivariate analysis. Barriers to evacuation intention in Florida were explored and compared to barriers for the entire Harvard survey population. Segm entation of the population based upon the significant predictors from bivariate analyses was conducted using SPSS softwares decision tree function

PAGE 115

103 to obtain a CHAID analysis. Split sample validation was used to create a training sample and a test sample. To explore potential confounding, a separate CHAID analyses was conducted with a reduced sample without the variable for previous evacuation experience. Goodness of fit and effect size statistics were calculated for the tree nodes to determine those var iables most influential in determining meaningful segments of people who intend to evacuate when told to. Variables that were found significant in CHAID analyses were used to construct logistic regression models to determine the strength and direction of a ssociation of the interactions. Summary of F indings The majority of the surveyed Florida population, 66.9%, lived in a community threatened by a hurricane and 26.6% of them had evacuated their homes in the last three years. When asked if they would evac uate in the future if told by government officials to do so, 59.1% of Florida residents surveyed said they intend to evacuate. In a similar survey of North Carolina residents, 52.7% of the population overall said they would evacuate if ordered to (Whitehead et al. 2000). This varies from actual evacuation behavior in Florida. During the 2004 hurricane season, when four storms made landfall in Florida, only 25% of the threatened population overall actually evacuated (Smith and McCarty, 2009). This gap b etween evacuation intention and actual evacuation in 2004 is larger than that in other studies. This may be due to the low hurricane activity rate in Florida for the ten years prior to the major storms of 2004. Although several tropical storms made landf all, the population did not experience a category 3 or

PAGE 116

104 higher hurricane and may have overestimated their ability to cope with a major storm. A study of Hurricane Lili evacuation in Texas and Louisiana found 65 % agreement between intention and actual evacuation behavior (Kang, Lindell, and Prater, 2007) The Homeland Security Community Preparedness and Participation Target Capability goal is that 80% of residents are prepared to evacuate when ordered to do so (FEMA, 2009). Most of the population surveyed was not too or not at all worried that a major hurricane would strike in the next 6 months (54%). In addition, a majority felt they were somewhat or very prepared for a hurricane (85.2%) and thought their homes would withstand a category 3 or higher stor m without major damage (59.9%). They were confident that if a hurricane did strike they would be rescued (71.7%). In the warning and response model by Lindell and Perry (1992), risk perception is based upon the determination that a threat really exist t hat protection is needed, and protection is feasible. Based on these criteria, t he responses in the survey indicate that most of the respondents d id not perceive hurricanes as a threat The strongest association with future evacuation intent was prior evacuation experience (Cramers V = .266) ; if they had evacuated before, they said they would evacuate again. In studies located that addressed evacuation intention or expectation, prior hurricane experience has not been measured consistently. Studies that have used experimental designs, such as Bakers 1983 pencil and paper experiment have shown weak or inconsistent relationships with previous experience and evacuation intention or planning (Baker, 1991). One

PAGE 117

105 study was located that addressed the impact of evacuation return delays on future evacuation intentions. The expectation of return delay was cited as a concern among those surveyed, but the inconvenience of return delay was not sufficient to prevent evacuation for most people (Dash, 2001). In previous studies of evacuation behavior, the best predictor of evacuation has been previous evacuation (Riad, et al., 1999; Baker, 1979; Baker, 1991). P rior experience, other than previous evacuation, has not been consistently associated with evacuation. Baker (1991) has asserted this was partially due to the difficulty in defining and measuring experience. Lindell and Perry (1992) stated that experience is only likely to produce an adaptive response when the experience is interpreted appropriately. E xperience with an event with minor effects may mistakenly be used to judge the impact of a more severe event and thus risk is underestimated. In this study experience was measured through a series of questions regarding threat, flooding, damage and problems that the individual had after the storm (see Table 12) These questions were more detailed than in any other study examined, yet none of the variables were significant in bivariate analysis In regression analysis in this study there was an association with evacuation intention and prior experience with government or volunteer agency response to hurricanes. Those who thought the government and volunteer agency response was poor were more likely to say they would stay home if told to evacuate (OR .596 (CI .166, .898), p = .027). Overall this research supports previous research that found no association with previous hurricane experience, other than evacuation, and evacuation intention.

PAGE 118

106 The expectation that the home would be flooded or damaged by wind and the expectation that the home could withstand a major hurricane were significantly associated with evacuation intention. Spearmans correlation coefficients ( r ) indicate that those who thought their homes would be flooded or damaged are likely to evacuate when told to do so ( r =.132, p < .01) and those who felt their homes would withstand a major hurricane were likely to stay when told to evacuate ( r = .125, p < .01) Logistic regression indicated that persons who were older, had children in the home, and owned generators were more likely to stay in their homes when ordered to evacuate. Other demographic variables that were significant in chi square analysis, such as race, and the presence of pets, fall below the p < .05 significance level in logistic regr ession analysis. CHAID analysis on all of the independent variables of concern identified interactions with previous evacuation, the presence of pets, age and the presence of a generator. When the person had not been threatened by a storm in the last three years, owning a pet influenced evacuation intention 2 (N = 163) = 8.27, p < .01). When pets were present in the home, persons w ere more likely to say they would stay in the home (43%) t han those without pets in the home (35%). In this analysis the segment most likely to stay in their homes when ordered to evacuate was comprised of people who had not evacuated in a previous storm, do not own a generator and are over the age of 65 (58.7%, N = 46). Overall 46.3% of the respondents had stayed in their hom es in a previous hurricane and 18.7% of them owned generators. A larger percentage of those

PAGE 119

107 who owned generators indicated they would stay (58.1%, N = 86) when told to leave than those who did not own a generator (41.7% N = 127). When CHAID analysis wa s conducted with only those variables significant in logistic regression, owning a generator split on age, <= 30 49 and > 3049. The segment most likely to stay in their homes when ordered to evacuate remained those people who had not evacuated in a previ ous storm, do not own a generator and are over the age of 65 (62.2%, N = 37). Hypotheses tested. Several hypotheses were tested during this research. The first was that prior experience evacuating during hurricane threats would be positively associated wi th evacuation intention. This hypothesis was supported. Throughout the analysis prior experience had a consistent positive association with evacuation intention ( 2 = 45.48, p < .01, Cramers V = 0.266). Prior experience was the first node in CHAID anal ysis as well as significant in regression and chi square. Another hypothesis tested was that prior experience evacuating during hurricane threats would be positively related to the level of risk perception. This hypothesis was supported to the extent that half of the risk perception related variables were positively associated with prior evacuation (see Table 14) Risk perception was also positively correlated with evacuation intention as hypothesized (see Table 13 ) Having family members together or ac counted for was hypothesized to be positively associated with evacuation intention. There was no significant

PAGE 120

108 association with evacuation intention and having a preselected place to meet or having an agreed on phone number for family members to call. However, with only two available variables, there was not adequate information to test this hypothesis. Ethnic majority status was hypothesized to be inversely related to evacuation intent. This was supported on the small effect level in chi square analysis but failed to reach the level of significance necessary in the logistic regression analysis. Race was also not significant in CHAID analysis. The racial and ethnic composition of this study population did not reflect the general population of Florida. The survey population was skewed toward white respondents. I am therefore unable to adequately test this hypothesis. The final hypothesis was that socioeconomic status would be inversely related to evacuation intent. Measures of income and education wer e not significant in either logistic regression or bivariate analysis. The computation of SES is a complex procedure utilizing information on income, education, profession, housing and wealth. The Harvard data set did not contain variables to adequately describe SES; I am therefore unable to adequately test this hypothesis. Evacuation c onsequences. The decision to evacuate entails a number of judgments on the part of the individual or family. Failure to evacuate can lead to injury and death; Czajkowski ( 2010) has state d that for every one category increase in storm strength, expected fatalities increase by a factor of 1.7 to 3.4. But at the same time,

PAGE 121

109 evacuation is not without its hazards and problems. In making the decision to evacuate the individual or family must decide whether they are at risk or not. When asked why they would not leave if told to do so by authorities, 34% of the respondents in this survey said they would not leave because they felt evacuating would be dangerous. Bad things do sometimes happen during evacuation. In 2005 a bus carrying elderly residents away from Hurricane Rita caught fire while stuck in gridlocked Interstate 45 traffic south of Dallas. The fire was fed by the residents oxygen tanks, which exploded, killing 24 people (Regnier, 2008). In New Orleans, many people said they did not evacuate because they did not own a car. Waiting for public transportation or trying to find a ride with friends is an impediment to leaving. When asked why they might not leave, 55% of the respondents in this survey said they thought the roads would be too crowded. Indeed, Hurricane Floyd left millions of people stranded on the highway for many hours, caught in gridlocked traffic. Evacuation can be perceived as leaving the home open to looting. Of the reasons given for not leaving when told to evacuate, 34% were afraid their possessions might be stolen if they left. A common scene in disaster movies is the crowd running rampant through the abandoned city stealing everything they can. In reality, looting of homes is rare (Perry, 1979) but the media portrayal of gangs of looters may well be confusing. Oftentimes the supposed looters are actually friends or relatives salvaging a disaster victims property (OLeary, 2004)

PAGE 122

110 When looti ng does occur it is usually carried out by outsiders, not local residents. However, the perception remains with some people. Evacuating requires not only finding a place to stay for the family, but also a place that will take the family pets. Barriers to pet evacuation include owning multiple pets, having outdoor dogs, or not having carriers for the animals. In this study, owning pets was significant in CHAID analysis ( 2 = 8.266, p < .01, Cramers V = .23), with pet owners more likely to stay in the hom e than non pet owners, but was not significant in logistic regression. In other studies the presence of pets has been associated with evacuation failure as well (Heath, et al., 2001; Smith and McCarty, 2009). In addition to the potential physical dangers and frustration of traffic, there is the economic cost of evacuation. Time away from a job and the cost of fuel and lodging were cited in this study as reasons persons might not evacuate when asked. Fifteen percent of those interviewed in this survey indicated that they could not afford to evacuate if told to leave. Contribution to Theory This study contributes to the limited body of knowledge regarding evacuation intention. As previously stated, there is a large body of research into evacuation behavior, but very little in the realm of evacuation intent. In many studies the presence of an adaptive plan is cited as a significant contributor to evacuation ( Burnside et al., 2007; Perry 1979, Lindell et al., 2001) but the barriers and incentives that lead to creating that adaptive plan are not explored. This study examines the actual correlates of evacuation intention. In this study,

PAGE 123

111 the variables that were significantly associat ed with prior evacuation were also significant ly associated with evacuation intention (see Tables 22 and 24). This study also contains a more detailed examination of prior experience and evacuation intention than previously found. This more detailed look at experience validates prior studies that have examined evacuation behavior One of the consistent significant variables in chi square analysis, logistic regression analysis and in CHAID was the presence of a generator in the home. The presence of a generator was just one of several questions in the survey used to determine preparedness. Other questions asked included whether the respondent had a battery operated radio, a flashlight, a first aid kit, a cell phone, cash and water purifying supplies. None of these other variables were significant individually in the analysis, but there is a significant correlation between the presence of a generator and these other variables (see Appendix D) There is also a significant correlation between having a generator and self reported preparedness for a hurricane ( r = .265, p < .01) and confidence the home would withstand a major hurricane ( r = .110, p < .01). Having a generator may contribute to a false sense of security that influences the intention to evacuate. This variable may also be functioning as a representation of planning behavior in the analysis. There were eight items that were listed in the analysis as things a person might have in preparation for a hurricane. The majority of respondents had six of these items. In his survey conducted for the Florida Association of Broadcasters, Baker (2006) also found the majority of respondents had six of eight preparedness supplies on hand. Baker labels these items as indicators of

PAGE 124

112 preparedness. This would support previous research linking evacuation intention with having an adaptive plan and risk perception. Figure 9 shows the frequency of planning variables in this study. Figure 9. Frequency of indicators of planning Strengths and Limitations This study was one of few that addressed future intentions to evacuate. The majority of studies on hurricane evacuation are retrospective in nature and do not incorporate future planning. This study is broad in scope and includes people in high risk areas whether they were exposed to a hurricane threat or not,

PAGE 125

113 making the data applicable t o a more general audience than those studies limited strictly to populations that were previously exposed to hurricanes. Although multiple studies have used multivariate statistics to analyze data, this is the first study to use classification trees to determine segments likely to evacuate or fail to evacuate in a future storm. Bivariate statistics and regression analysis have been used to determine correlates of evacuation, but they fail to take into consideration the multiple interactions explored through CHAID that lead to population segments that can be identified and targeted in education and marketing campaigns. The size of this study was also a strength. With over 5,000 participants in the main survey and over 1,000 participants in the Florida sur vey, this is one of the largest evacuation intention studies ever conducted. In comparison, the 2009 Citizen Corps National Survey contained interviews from 3,448 households nationally. There are weaknesses associated with the study as well. This survey relied on a random digit dialing telephone survey to collect the data. Traditional random digit dialing telephone surveys have come under question due to the proliferation of cell phone only households (Lee, 2010). The Consumer Expenditure survey has estimated that cell phone only households have grown from less than 1% in 2000 to 18.4% as of the second half of 2009. In addition to this population, there is the increase in cell phone mostly households. These households have a landline but are difficul t to reach because they mainly use cell phones. This may lead to a noncoverage error in the population surveyed. Telephone surveys are also subject to the practice of call screening by potential

PAGE 126

114 recipients. Early research into this issue found that ans wering machine owners were still reachable for survey purposes ( Tuckel and Feinberg, 1991) This can lead to a biased population of only those interested in the subject as participants in the survey. Florida is the part time home of many winter resident s as well. Many part time residents are only present during the winter months, traditionally coming to Florida around Thanksgiving and leaving for their homes in other states around Easter. The Harvard study was conducted in July of 2007, therefore missi ng the input of these winter only residents. Florida has not had a major hurricane landfall since 2005. Those residents who moved to Florida since that time have not experienced an active landfall year and thus do not have recent direct exp erience with hurricanes. Coming off of an active year, residents may be more likely to consider preparation and thus say they would evacuate. The use of secondary data limits inquiry to the variables available in the data set. These variables did not address the exact issues needed for hypothesis testing. In this study there w ere no detail s on family structure, such as number of children in the home or the ages of residents in the home, which limited the ability to address family accountability. In the same way, so cioeconomic status was not adequately defined by the available variables and so hypotheses regarding SES were not testable. Communications is known to play a part in the evacuation decision process but this study did not explore information source or credibility. This leaves out the social interaction of p lanning found in the warning and response model As in any secondary data analysis

PAGE 127

115 there are always issues of question wording and ordering. These data provide a snapshot of a single moment Evacuation decisions often occur as an optimal stopping problem where every potential evacuation time prior to the actual impact the household makes the decision to leave or stay versus the now or never one time choice (Czajkowski, 2007) By looking at a hypotheti cal future event, the stresses of the moment are not incorporated into the decision process. CHAID use s a stepforward model fitting method when not in automatic mode. As in other stepforward regression fitting models results depend on the order in whi ch the variables are entered into the model. With smaller samples an i mportant concern is the danger of over fitting the data. The training sample may contain random variations not in other samples that cause variations in the tree when new data are supplied to the algorithm (Neville, 1999). Over fitting is detected by applying the tree to new data, the test sample, and comparing the outcome. When over fitting occurs it can be addressed by pruning the tree and removing extraneous variables that may cause interference in the model. CHAID can be revised manually by either pruning extraneous variables or forcing a variable into the analysis at different stages. Since CHAID uses the chi squared statistic it is assumed to follow the chi squared distribut ion. This assumption requires a large sample size to ensure the validity of the test, though some authors have used samples of 500 cases with satisfactory results (van Diepen and Franses, 2006 ). Although this study was large, using split sample validation reduces the practical sample size to around 500 cases for training and 500 for the test sample. A larger sample size might

PAGE 128

116 have given different results. Therefore, the results of CHAID segmentation should be confirmed through another analysis approach. In this research logistic regression confirmed the significance of the CHAID variables. Classification accuracy in the CHAID model only reached approximately 70% for the model with significant variables from the logistic regression. This classification accuracy rate is greater than chance but less than optimal. Cramers V and the correlation coefficients were almost all in the small category for Cohens rule of thumb for interpreting effect sizes. Considering the model classification accuracy, Cramer s V and correlation coefficients, the results should be considered preliminary. Future R esearch The 2009 Citizens Corps National Survey recommends further research to explore the characteristics of groups who share similar attitudes and behaviors. This research has begun that task, but more needs to be done. The use of classification and regression tree methodologies such as CHAID reveals relationships that are not apparent in traditional regression analysis. This method of analysis should be further explored. One of the findings in previous research has been that the communication channel and format for hazard information influences risk perception and evacuation behavior. Traditional methods of hazard communication have used radio and television to disseminate information with varying success (Baker, 2006) In the 21st century the use of social networking websites and instant communications through texting and Twitter can change the face of information dissemination. As of April 2010 there were over 105 million Twitter users

PAGE 129

117 sending 55 million tweets per day (Bodnar, 2010) Facebook has reached the 500 million user mark and is used by governmental agencies to disseminate information. These instant communication media offer a new milling environment for the emergent norm process. The idea of appropriate response to an oncoming storm may now have the influence of potentially thousands of advice givers, sharing prior experience, up to the minute traffic delays and damage reports. The exchange of lay information, rumors and advice may impact the decision by those in the path or on the periphery of a storm to evacuate or not. Because the person texting or tweeting may be in a completely different impact zone for a storm, this may lead to misinformation based on personal experience. In this study a majority of people under the age of 50 intended to evacuate when told to do so (see Tables 5 and 6). However, there were still 30% of those surveyed under the age of 50 who did not intend to evacuate. More than 80% of Twitter users are under the age of 50 ( Bodnar, 2010) and a majority of online users on social media sites are under 40. Future research should investigate the use of Facebook, text ing and Twitter in disaster situations to see if this interacti on impacts the evacuation decision of this age group. In this analysis people over the age of 65 were less likely to evacuate when told to do so. This is a rapidly growing segment of the population in America In this survey, those over the age of 65 r epresented the largest segment of the population, 33%. As this age group retires and moves to coastal areas, they will be more likely to be impacted by hurricanes. In the Citizen Corps National Survey individuals over 55 years of age were less likely to prepare and

PAGE 130

118 more likely to rely on first responders for rescue than younger people. Knowing this population is the least likely to use social media, risk information channels should be developed that target this group especially through media they are mor e comfortable with, such as television. Further research can target this population to determine the reasons behind their intention not to evacuate and determine potential interventions This analysis was limited to the Florida population. Further analys is should be conducted on the complete Harvard survey data and compared with the Florida data to validate the results Using the entire sample will give a larger population for CHAID analysis and address some of the concerns with small sample sizes mentioned previously. Dissemination The report will be shared with the Apalachee Regional Planning Council and the Division of Emergency Management Operations in the Department of Health. The report will be summarized and presented to the Homeland Security Re gional Domestic Security Task Force region two emergency m anagement section. A journal article will be prepared and submitted for possible publication in a professional journal. Potential journals for publication include Environment and Behavior Sociolo gical Spectrum Natural Hazards Review, and Population and Environment Utilization. This is the first study to use classification and regression trees to determine segments of the population based on future evacuation expectation.

PAGE 131

119 The results of this s tudy suggest that there are segments of the population that may be open to social marketing interventions. This research presents a target for intervention. Current information dissemination on evacuation emphasizes storm predictions and hazard warning t o the general population. CHAID analysis indicates that the population over the age of 65 who had not evacuated previously was the population most likely to stay in their homes when told to evacuate. Using this research as a starting place, information c an be targeted at the segment least likely to evacuate in ways relevant to their concerns In keeping with the recommendations of the 2009 Citizen Corps survey, efforts to address self efficacy concerns and move this age group from awareness to preparedne ss should be implemented.

PAGE 132

120 References Aguirre, B E., Wenger, D., Vigo, G (1998) A Test of the Emergent Norm Theory of Collective Behavior Sociological Forum, (Vol. 13), 2, pp. 301320. Alfonso, M. L. (2007). The tip of the blade: Self injury among early adolescents (Doctoral dissertation). Available from http://digital.lib.usf.edu:8080/fedora/ get/usfldc:E14SFE0002096/DOCUMENT American Red Cross (2003), Fact sheet on shelter in place, Retrieved July 29, 2010 from http://www.redcross.org/preparedness/cdc_english/Sheltering.asp#top. Andreasen, A (1995) Marketing social change. San Francisco: Jossey Bass. Atkins, D. & Moy, E M (2005) Left behind: The legacy of hurricane Katrina. BMJ, 331; 9169 18. doi: 10.1136/bmj.331.7522.916 Arlikatti, S., Lindell, M. K., Prater, C. S., Zhang, Y. (2006) Risk Area Accuracy and Hurricane Evacuation Expectations of Coastal Residents Environment and Behavior, 38, 226. doi: 10.1177/0013916505277603 Bailey, M., Glo ver, R., & Huang, Y (2004 ) Epidemiologic assessment of the impact of four hurricanes ---Florida, 2004. MMWR, 54 (28), 693697. Baker, J (1979) Predicting Response to Hurricane Warnings: A Reanalysis of Data from Four Studies Mass Emergencies (Vol. 4), pp. 9 24.

PAGE 133

121 Baker, E J (1991) Hurricane e vacuation b ehavior International Journal of Mass Emergencies and Disasters (Vol. 9) No 2: 287310. Baker, E. J. (1995) Public response to hurricane probability forecasts. The Professional Geographer, 47:2, 137147. doi: 10.1111/j.00330124.1995.00137.x Baker, E J (2006, September) Florida statewide hurricane preparedness survey: Benchmarks for building a culture of preparedness Paper presented at the Florida Association of Broadcasters, Orlando, Florida. Bate man, J. M., & Edwards, B. (2002) Gender and evacuation: A closer look at why women are more likely to evacuate for hurricanes. Natural Hazards review, 3 10717. doi 10.1061/ 15276988. 3 3 107 Bennett, P and Calman, K., (Eds.) (1999) Risk communication and public health. NY: Oxford University Press Bernert, E H., Ikle, F C (1952) Evacuation and the cohesion of urban groups The American Journal of Sociology, 58., 2, pp 133138. Blake, E. S., Rappaport, E. N., Landsea, C. W. (2007) The deadliest, costl iest, and most intense United States tropical cyclones from 1851 to 2006. (NOAA Technical Memorandum NWS TPC 5). Miami, FL: National Hurricane Center. Blendon, R J., Buhr, T., Benson, J M., Weldon, K J. & Herrmann, M. J. (2007) Hurricane readiness in high risk areas Harvard School of Public Health. Retrieved from http://www.hsph.harvard.edu/news/press releases/files/Hurricane_release_2007_topl ine_total.doc

PAGE 134

122 Bodnar, Kipp (2010) The ultimate list: 100 Twitter statistics Retrieved from http://blog.hubspot.com/blog/tabid/6307/bid/6050/TheUltimate List 100Twitter Statistics.aspx Bolin, R. (1976) Family recovery from natural disaster: A preliminary model Mass Emergencies, 1, 267277. Bourque, L. B., Siegel, J M., Kano, M., & Wo o d, M M (2006) Weathering the storm: The impact of hurric anes on physical and mental health. The Annals of the American Academy, 604. Burnside, R., Miller, D S., & Rivera, J D (2007) The impact of information and risk perception on the hurricane evacuation and decision making of greater New Orleans residents Sociological Spectrum, 27: 727740. doi:10.1080/02732170701534226. Christensen, L. and Ruch, C E (1978) Assessment of Brochures & Radio & Television Presentations on Hurricane Awareness Mass Emergencies, 3, 209216. Covello, V.T (1991) Risk Compariso n and Risk Communication. In Kasperson, R.E. and Stallen, P.J (Eds.), Communicating Risks to the Public: International Perspectives Dordrecht, The Netherlands: Kluwer Academic Publishers. Czajkowski, J., Kennedy, E. (2010) Fatal tradeoff? Toward a better understanding of the costs of not evacuating from a hurricane in landfall counties. Population and Environment, 31: 121149. doi 10.1007/s111110090097x.

PAGE 135

123 Dash, N., & Morrow, B H (2001) Return delays and evacuation order compliance: the case of Hurric ane Georges and the Florida Keys Environmental Hazards 2, 119128. Deslatte, A (2006, June 14) Many shrug off evacuation ahead of Alberto The Tallahassee Democrat, pp. A5. Drabek, T. E & Boggs, K S (1968) Families in disaster: Reactions and relatives Journal of Marriage and the Family, 30, 443451 Drabeck, T. E. (2001) Disaster warning and evacuation responses by private business employees. Disasters, 25, 1, 7694. Dow, K & Cutter, S L., ( 1998) Crying wolf: Repeat responses to hurricane evacuation orders Coastal Management, 26 (4), pp 237252 Effects of Hurricane Katrina in New Orleans. In Wikipedia (2009, September 28) Retrieved from http://en.wikipedia.org/wiki/Effects_of_Hurricane_Katrina_ in_New_Orleans Effect size (n.d.) Retrieved from the Psychology Wiki: http://psychology. wikia.com/wiki/Effect_size_(statistical) Eisenman, D., Cordasco, K., Asch, S., Golden, J., & Glik, D. (2007). Disaster planning and risk communication with vulnerable communities: Lessons from hurricane Katrina. American Journal of Public Health: Supplement 1, 97(S1), S10915. Retrieved July 14, 2010, from ABI/INFORM Global. (Document ID: 1402418371). Elder, K., Xirasagar, S., Mil ler, N., Bowen, S. A., Glover, S., & Piper, C (2007) African Americans decisions not to evacuate New Orleans before

PAGE 136

124 Hurricane Katrina: A qualitative Study American Journal of Public Health, 97, S1. F ederal E mergency M anagement A gency (2004) Food and wat er in an emergency (FEMA Publication 477 A5055) Washington, D.C. : Author. Federal Emergency Management Agency (2009) Personal preparedness in America: Findings from the 2009 citizen corps national survey Washington, D.C.: Author. Florida Department of H ealth (2010) Hurricane resource center, Vulnerable populations. Retrieved from http://www.floridashealth.com/prepare/ VulnerablePopulations.html Harvard University (2008) Project on the public and biosecurity. Retrieved from http://www.hsph.harvard.edu/research/horp/project onthe public andbiosecurity/ Heath, S E., Kass, P. H., Beck, A M., & Glickman, L T., (2001) Human and pet related risk factors for household evacuation failure during a natural disaster, American Journal of Epidemiology, (153), 659665. Horney, J. A., MacDonald, P. D. M., Van Willigen, M., Berke, P. R., & Kaufman, J. S. (2010) Individual actual or perceived property flood risk: Did it predict evacuations from Hurricane Isabel i n North Carolina, 2003? Risk Analysis, 30, (3) 501 511. Kang, J E., Lindell, M K., & Prater, C S (2007) Hurricane evacuation expectations and actual behavior in Hurricane Lili Journal of Applied Social Psychology, 37 887 903.

PAGE 137

125 Kotler, P & Andreasen, A (1991) Strategic marketing for nonprofit organizations, 163200. Prentice Hall, Englewood Cliffs, N.J. Koutnik, F. (2000) Task force examines current systems InDepth, 3 (2), 4 7. Lee, S., Brick, M., Brown, R., & Grant, D. (2010) Growing cell phone population and non coverage bias in traditional random digit dial telephone health surveys. Health Services Research 45(4) 11211139.. doi: 10.1111/j.14756773.2010.01120.x Lindell, M. K & Perry, R.W (1992) Behavioral foundations of community emergency planning. Washington, D.C.: Hemisphere Pub. Lindell, M. K. & Perry, R. W. (2004) Communicating environmental risk in multiethnic communities. Thousand Oaks, CA: Sage. Lindell, M K., Lu, J., & Prater, C S (2005) Household Decision Making and Evacuation in Res ponse to Hurricane Lili Natural Hazards Review 6 (4) 171179 Lindell, M. K., Prater, C. S., Sanderson, Jr., W. G., Lee, H. M., Yang, Z., Mohite, A., & Hwang, S. N. (2001) Texas gulf coast residents expectations and intentions regarding hurricane evacuat ion. Texas A&M University Retrieved from Texas A&M University, Hazard Reduction & Recovery Center, College of Architecture website: http://archone.tamu.edu/hrrc/ Publications Luhmann, N (1993) Risk: A sociological theory New York: Walter de Guyter, I nc.

PAGE 138

126 Magidson, J (1994) The CHAID approach to segmentation modeling: Chi squared automatic interaction detection. In R Bagozzis (Ed.) Advanced methods of marketing research (pp 119 159) Cambridge, MA: Blackwell Publishers. Maibach, E.W., Maxfield, K .L., Ladin, K & Slater, M (2000) Translating health psychology into effective health communication. Journal of Health Psychology, 1 261277. Malilay, J (1997) Tropical Cyclones In E Noji (Ed.) The public health consequences of disasters (pp. 207 227) Oxford: Oxford University Press. Mardberg, B (1996) Forming homogeneous clusters for differential risk information. Radiation Protection Dosimetry, 68(3) 227230. Mayhorn, C P (2005) Cognitive aging and the processing of hazard information and disaster warnings Natural Hazards Review, 6, (4) p 165 Mileti, D.S. and Peek, L. (2000) The social psychology of public response to warnings of a nuclear power plant accident Journal of Hazardous Materials, 75, 181194. Mozumder, P., Raheem, N., Talberth, J., & Berrens R.P (2008) Investigating intended evacuation from wildfires in the wildland urban interface: Application of a bivariate probit model Forest Policy and Economics, 10, 415 423.

PAGE 139

127 National Oceanic and Atmospheric Administration (2006) Hurrican e Katrina, A climatological perspective. (Technical Report 200501). Retrieved from http://www.ncdc.noaa.gov/oa/reports/techreport 200501z.pdf Nielsen Company (2010) Segmentation and marketing. Retrieved November 24, 2010 from http://en us.nielsen.com/content/nielsen/en_us/ expertise/segmentation_and_targeting.html Neville, P. G. (1999) Decision trees for predictive modeling. The SAS Institute. Retrieved October 9, 2009 from http://www.sasenterpriseminer.com/ documents/ Decision%20Trees%20for%20Predictive%20Modeling.pdf The New York Times (2007) Class Matters Retrieved September 25, 2010 from http://www.nytimes.com/packages/html/n ational/20050515_CLASS_GRA PHIC/index_01.html Norusis, M J. (2008) SPSS statistics 17.0: Statistical procedures companion Upper Saddle River, N.J.: Prentice Hall. OLeary, M. (2004) The first 72 hours: A community approach to disaster preparedness. Lincoln, Nebraska: iUniverse Publishing. Retrieved from http://www.iuniverse.com/bookstore/book_detail.asp?&isbn=0595 310842 Perry, R.W (1979) Incentives for evacuation in natural disaster Journal of the American Planning Association, 45, 440447. doi: 10. 1080/01944367908976988 Perry, R.W (1985) Comprehensive Emergency Management: Evacuating Threatened Populations Greenwich, Ct: Jai Press.

PAGE 140

128 Pielke, R. A., & Landsea, C W (1998) Normalized hurricane damages in the United States: 192595. American Meteorological Society. Prochaska, J. O., & DiClemente, C. C. (1983) Stages and processes of self change of smoking: Toward an integrative model of change. Journal of Consulting and Clinical Psychology, 51, 390395 Rappaport, E N & Fernandez Partagas, J (1995, M ay 28, updated 1997, April 22) The Deadliest Atlantic Tropical Cyclones, 1492 Present Retrieved from http://www.nhc.noaa.gov/pastdeadly.html Regnier, E. (2008). Public Evacuation Decisions and Hur ricane Track Uncertainty. Management Science, 54(1), 1628. Retrieved from Business Source Complete database. Riad, J. K., Norris, F. H., & Ruback, R. B. (1999) Predicting evacuation in two major disasters: Risk perception, social influence, and access to resources. Journal of applied social Psychology, 29 (5), pp. 918934. Rodeghier, M. (2007) Extending your survey research results with decision trees. Retrieved from http://www. SPSS.com Sattler, D N., Kaiser, C F & Hittner, J B (2000) Disaster prepar edness: relationships among prior experience, personal characteristics, and distress Journal of Applied Social Psychology, 30 (7), pp. 13961420. Sheppard, B., Hartwick, J., & Warshaw, P. (1988). The theory of reasoned action: A metaanalysis of past research with recommendations for modifications and future research. Journal of Consumer Research, 15, 325343.

PAGE 141

129 Slater, M. D (1995) Choosing audience segmentation strategies and methods for health communication. In Maibach, E & Parrot, R (Eds.) Designing health messages: Approaches from communication theory & public health practice. Thousand Oaks, CA: Sage. Slater, M. D (1996) Theory and method in health audience segmentation. Journal of Health Communication, 1: 267283. Smith, S. K. & McCarty, C. (2009) F leeing the storm(s): An examination of evacuation behavior during Floridas 2004 hurricane season. Demography, (46) 1. 127145. Smith, W R., (1956) Product differentiation and market segmentation as alternative marketing strategies Journal of Marketing (21), 38. Sorensen, J and Mileti, D (1991) Risk communication in emergencies In Kasperson, R.E and Stallen, P.J (Eds.), Communicating Risks to the Public: International Perspectives Dordrecht, The Netherlands: Kluwer Academic Publishers. South Florida Regional Planning Council (20 07), Shelter in place. Retrieved July 29, 2010 from http://www.sfrpc.org/data/lepc/sip/ShelterInPlaceSlid eShow.pps#277 Tabachnick, B. G. and Fidell, L. S. (2001). Using multivariate statistics, (4th ed)., Needham Heights, MA: Allyn & Bacon. The Measurement Group. (19992005) CHAID. Retrieved October 9, 2009 from http://www.themeasurementgroup.com/Definiti ons/CHAID.htm.

PAGE 142

130 Tierney, K J., Lindell, M K., & Perry, R. W (2001) Facing the unexpected Disaster preparedness and response in the United States Washington, D.C.: Joseph Henry Press. Tobin, G.A., Bell, H.M., Montz, B.E., Hughey, E.P., Whiteford, L.M., Everist, M.P., Kelsey, C., & Miller R (2005) Hurricane Charley The Aftermath: Impacts and responses (Technical Report Feb, 2005) Tampa: University of South Florida, Global Center for Disaster Management and Humanitarian Action, Natural Hazards Research an Applications Information Center, USF Department of Geography, USF Department of Anthropology, Binghamton University Department of Geography. Turner, R and Killian, L. (1957) Collective Behavior New York Prentice Hall University of Nebraska (2010) Effect size overheads. Retrieved September 20, 2010 from http://psych.unl.edu/psycrs/971/meta/effect_sizes.ppt#291,23 Interpreting Effect Size Results. U.S. Army Corps of Engineers South Atlantic Division (1993) Hurricane Andrew assessment Florida. Washington DC: Government Printing Office. U.S. Census Bureau (2010) State & County QuikFacts. Retrieved July 29, 2010, from http://quickfacts.census.gov/qfd/states/12000.html. Van Diepe n, M. & Franses, P.H. (200 6 ). Evaluating chi squared automatic interaction detection. Information Systems 31 pp 814 831 doi:10.1016/j.is.2005.03.002

PAGE 143

131 Wedel, M., & Kamakura, W (2000) Market segmentation: Conceptual and methodological foundations, 2nd ed. N orwell, MA. Kluwer Academic Publishers. Weinstein, A (1994) Market segmentation: Using demographics, psychographics and other niche marketing techniques to predict and model customer behavior New York, McGraw Hill. Whitehead, J. C. (2003) One million dol lars per mile? The opportunity costs of hurricane evacuation. Ocean and Coastal Management 46, 10691083. doi:10.1016/j.ocecoaman.2003.11.001 Whitehead, J.C., Edwards, B, Van Willigen, M., Maiolo, J.R., Wilson, K., & Smith, K.T. (2000) Heading for higher ground: factors affecting real and hypothetical evacuation behavior Environmental Hazards 2, 133 142. Widener, A (Ed.) (1979) Gustave Le Bon. The man and his works Indianapolis, IN Liberty Press.

PAGE 144

132 APPENDICES

PAGE 145

133 Appendix A : Hurricane Readiness in Hi gh Risk Areas Survey Harvard School of Public Health Project on the Public and Biological Security HURRICANE READINESS IN HIGHRISK AREAS June 18July 10, 2007 N=5,046 adults in coastal counties (within 20 miles of the coastline) of Alabama, Fl orida, Georgia, Louisiana, Mississippi, North Carolina, South Carolina, and Texas Research Team: Robert J. Blendon, Harvard School of Public Health, Project Director Tami Buhr, Harvard School of Public Health John M. Benson, Harvard School of Public Health Kathleen J. Weldon, Harvard School of Public Health Melissa J. Herrmann, ICR/International Communications Research Contact: Robert J. Blendon, 6174324502

PAGE 146

134 Hurricane Readiness in High -Risk Areas Overall Survey Results The study was conduc ted for the Harvard School of Public Health via telephone by ICR an independent research company. Interviews were conducted from June 18 to July 10, 2007, among a representative sample of 5046 respondents age 18 and older in coastal counties of Texas, Lou isiana, Mississippi, Alabama, Florida, Georgia, South Carolina, and North Carolina. Interviewing was conducted in all counties within twenty miles of the coastline for each of these states. The survey included 502 residents of the New Orleans metropolitan area where interviews were conducted with adults from cellphone -only households, as well from households with landline telephones. The margin of error for total respondents is +/ 2.6 percentage points at the 95% confidence level. 1. How worried are you that a major hurricane will hit your community during the next 6 months? 2. Overall, how prepared are you if a major hurricane were to strike your community during the next 6 months? 3. Im going to read you a list of things some people have in the ir homes that could be used in case of a hurricane emergency. For each one, please tell me if that is something you currently have or do not have. A battery operated radio that you know works A flashlight that you know works A first aid kit Extra ba tteries A cell phone At least $300 in cash to take with you if you had to leave your home Water purifying supplies such as chlorine or iodine tablets A generator (Asked of total with a generator) 4. Have you heard or read about the dangers from car bon monoxide due to running your generator in an area that isnt properly ventilated? 5. If grocery stores in your community were closed due to a major hurricane and you had lost electricity, for how many days could you feed your family on the non-perish able food you currently have in your home? 6. How many days of clean water do you think you would have on hand if the running water in your home was cut off by a major hurricane? 7. How much water do you think you should have on hand for each member of your family? About a quart of water per day, about a half -gallon of water per day, about a gallon of water per day, or about two gallons of water per day?

PAGE 147

135 8. If there were a power outage, how long do you think the perishable food in your refrigerator li ke milk and meat would remain safe to eat? A few hours, one day, two days, or three days or more? 9. Do you or does anyone else in your household take prescription drugs on a regular or ongoing basis, or not? (Asked of total who take Rx drugs on a regu lar basis) 10. In the event of a major hurricane, do you and other household members have at least an extra three week supply of the prescription drugs you take regularly, or not? 11. If government officials said that you had to evacuate the area because there was going to be a major hurricane in the next few days, would you leave the area or would you stay? 12. If you had to evacuate the area where you live because of a major hurricane, would you need help to do so? 13. Do you have that help lined up? (Asked of total who would/might evacuate) 14. If you had to evacuate the area where you live because of a major hurricane, where would you go? Stay with friends or family members in another area Go to a hotel or motel Go to an evacuation center ru n by the Red Cross or government Sleep in a car or outdoors Dont know where you would go Refused 15. If you had to evacuate because of a major hurricane, how far away would you go? Less than 10 miles 10 to 50 miles 50 to 100 miles 100 to 200 miles More than 200 miles Dont know Refused 16. If you had to evacuate, how would you get to where you are going? Go in your car In a friends car Use public transportation Walk or ride a bike Dont know how you would evacuate Refused 17. If you had to evacuate because of a major hurricane, when would you return home? Would you Return to your home as soon as the hurricane is over Wait until officials say its safe to go back Dont know Refused

PAGE 148

136 (Asked of total who would/might stay in area if there wer e an evacuation) 18. Im going to read a list of reasons some people might have for not evacuating the area where they live if there were a major hurricane. For each one, please tell me if it is a reason why you would not /might not evacuate. You don t know where to go You dont have a car or know anyone who could give you a ride You have medical or physical problems that would make it difficult to leave You have to take care of someone who would be physically unable to leave You would be worried y our possessions would be stolen or damaged if you left You would not want to leave your pet You would not be able to afford to leave You would not be able to leave your job You think your home is well built and you will be safe at home. You think eva cuating would be dangerous You think the roads would be too crowded to leave 19. If a major hurricane were to hit your community and for whatever reason you did not leave your home, how confident are you that you would be rescued if you needed to be? 20. Do you or any other household members have any pets in your home, such as dogs, cats, birds and the like? (Asked of total who have pets) 21. If you had to evacuate because of a hurricane, do you have a place you can go where you can take your pet, or not? 22. Has your family agreed on a phone number outside the region that all members of your immediate family could call in the event of a hurricane if you are unable to communicate, or havent you done that? 23. Has your family agreed on a place you could meet after a hurricane is over if you got separated and could not go back home, or havent you done that? 24. Do you know the location of an evacuation center in your community where you could go if you had to? (Asked of total who know the lo cation of an evacuation center) 25. Do you know if this evacuation center is considered strong enough to withstand a hurricane rated category 3 or higher? 26. If you had to go to an evacuation shelter because of a hurricane, how worried would you be about the conditions and your safety?

PAGE 149

137 27. Im going to read a list of concerns people sometimes have about going to a hurricane evacuation center or shelter. If you had to go to a shelter because of a hurricane, please tell me how worried you would be about each one. a. You wouldnt have enough clean water to drink b. You wouldnt have enough food to eat c. You wouldnt have the prescription drugs or medicines that you need d. You would be threatened by violence e. You would need medical care and wouldnt be able to get it f. The conditions of the shelter would be unsanitary g. You would be exposed to sick people and could catch their illness h. The shelter would be too crowded and you would not have any privacy i. You would have a hard time communicating with family outside of the shelter j. Your valuables might be stolen 28. Thinking back over the past three years was your community threatened or hit by a major hurricane, or not? (Asked of total whose community was threatened/hit by a major hurricane in last 3 years 29. Before the hurricane hit, were you able to get the information you needed to keep yourself and your family safe, or not? 30. Was your community damaged by this hurricane, or not? 31. Was there major flooding associated with this hurricane in your community or not? 32. Because of this hurricane, did you leave your home where you lived, or did you stay in your home? (Asked of total whose community was threatened/hit by a major hurricane in last 3 years and left home where lived) 33. When you left your home, do you happen to remember if you brought each of the following documents with you? A. Proof of health insurance 33aa. Was that because you didnt have health insurance, or because you didnt think to bring proof of health insurance with you? B. Proof of prescription drugs you and your family were taking 33bb. Was that because no one in your family was taking prescription drugs, or because you didnt think to bring proof of the prescription drugs with you? C. Proof of homeowners or renters insurance 33cc. Was that because you didnt have homeowners or renters insurance, or because you didnt think to bring proof of it with you? 34. If you had needed your Social Security number while you were away from your home, would you have been able to provide it?

PAGE 150

138 35. How would you rate the response of government and voluntary agencies to the problems created by this storm? 36. During and immediately following this hurricane, was there a time when you had any of the following problems? You didnt have enough fresh water to drink You didnt have enough food to eat You didnt have the prescription drugs or medicines that you needed You were threatened by violence You needed medical care and couldnt get it You had problems getting gas to evacuate You had other problems evacuating You didnt have enough money You had problems because you were disabled or chronically ill You had problems caring for a disabled, chronically ill or elderly member of your household You suffered from heat exhaustion due to power failure You were injured as a result of the storm 37. Thinking about where your home is located, how likely is your home to be flooded or damaged due to wind in a major hurricane? (READ S CALE) 38. Is your home located in an evacuation zone or not, or dont you know if it is in an evacuation zone? 39. Do you think your home would withstand a major hurricane of Category 3 or higher without significant damage? 40. Now thinking about your own health status In general, would you say your health is excellent, very good, good, fair, or poor? 41. Do you or does anyone in your household have a chronic illness or disability that would require you to get help if you had to evacuate because of a major hurricane, or not? (Asked of total who have or someone in household has a chronic illness or disability) 42. Do you have help lined up for this person with the chronic illness or disability if you need to evacuate because of a major hurricane or not? 43. Do you live in a home you or your family own, are you renting a house or apartment, or do you live somewhere else? 44. Do you live in a single family home, a duplex or multi -family home, an apartment building or condominium, or a mobile hom e? 45. How long have you lived in your community? Less than 1 year 1 -5 years 6 -10 years 11-20 years More than 20 years Your whole life Dont know Refused

PAGE 151

139 46. Are you currently married, living with a partner, widowed, divorced, separated, or h ave you never married? 47. Are you, yourself now covered by any form of health insurance or health plan, or do you not have health insurance at this time? 48. Do you have homeowners or renters insurance or dont you have this insurance at this time? D01. Including yourself, how many adults, 18 or older, are there living in your household? D02. Are there any children under the age of 18 living in your household? DO3. Besides the telephone number I reached you on, how many other telephone number s, if any, does your household have that I could have reached you on? D03a. In total, how many cell phones do you and all the other members of your household have? D04. What is the last grade or class that you completed in school? HS or Less (net) Less than HS (sub net) None, or grade 1 8 High school incomplete (grades 9 11) HS Grad (sub net) High school graduate (grade 12 or GED certificate) Business, technical, or vocational school AFTER high school Some college or more (net) So me college College graduate+ (sub net) College graduate Post graduate training or professional schooling after college Dont Know Refused D05. What is your age? D06. Are you, yourself, of Hispanic or Latino background, such as Mexican, Puer to Rican, Cuban, or other Latin American background? (Asked of total Hispanics) D06a. Are you White Hispanic or Black Hispanic? (Asked of total non -Hispanics) D07. Do you consider yourself to be white, black or African-American, Asian -American, or so me other race?

PAGE 152

140 D08. IS YOUR TOTAL ANNUAL HOUSEHOLD INCOME FROM ALL SOURCES, AND BEFORE TAXES: Less than $40,000 NET Less than $15K $15K but less than 20K $20K but less than $25K $25K but less than $30K $30K but less than $40K Less than $40K (unspecified) $40,000 or more NET $40K but less than $50K $50K but less than $75K $75K but less than $100K $100K + $40K + (unspecified) Dont know Refused RECORD GENDER FROM OBSERVATION

PAGE 153

141 Appendix B : Population Statistics for Surveyed Counties total population 2008 % population under 18 2007 % population <65 years 2007 % females, 2007 % White, 2007 % Black, 2007 % Hispanic 2007 % AIAN, 2007 % Asian, 2007 % high school graduate or higher 2000 % bachelor's degree or hig her 2000 Persons per household 2000 Median household income 2007 Population per square mile 2000 Bay, FL 163,946 23.00 14.20 50.50 83.60 11.60 3.50 0.80 1.90 81.00 17 .70 2.43 47,495.00 194.00 Brevard, FL 536,521 20.20 20.10 50.90 85.90 10.00 6.90 0.40 2.00 86.30 23.60 2.35 50,261.00 467.80 Broward, FL 1,751,234 23.60 14.30 51.40 69.60 25.30 23.40 0.40 3.00 82.00 24.50 2.45 52,504.00 1,346.90 Charlotte, FL 150,060 16.20 30.30 51.90 91.70 5.70 5.30 0.30 1.30 82.10 17.60 2.18 46,328.00 204.10 Citrus, FL 141,416 16.50 30.20 52.00 94.20 3.20 4.20 0.30 1.30 78.30 13.20 2.20 36,979.00 202.20 Collier, FL 315,258 20.70 25.20 49.40 91.90 5.90 25.50 0.40 1.10 81.80 27.90 2.39 58,519.0 0 124.10 DeSoto, FL 33,991 22.50 15.90 41.90 83.70 12.00 33.60 2.80 0.60 63.50 8.40 2.70 35,988.00 50.60 Dixie, FL 14,957 20.30 19. 20 46.40 88.80 9.50 2.60 0.40 0.20 65.90 6.80 2.44 31,018.00 19.60 Duval, FL 850,962 25.90 10.50 51.50 64.50 29.90 6. 00 0.40 3.50 82.70 21.90 2.51 49,175.00 1,006.30 Escambia, FL 302,939 22.30 14.60 50.30 71.70 22.50 3.60 0.90 2.50 82.10 21.00 2.45 41,772.00 444.70 Flagler, FL 91,247 18.80 23.90 51.30 86.20 10.40 8.00 0.20 2.00 85.90 21.20 2.32 45,639.00 102.70 Franklin, FL 11, 202 19.90 17.20 48.80 86.70 11.30 1.70 0.50 0.40 68.30 12.40 2.28 35,182.00 20.30 Gulf, FL 15,667 18.60 16.60 45.50 79.60 17.60 2.70 0.70 0.40 72.60 10.10 2.42 38,160.00 24.00

PAGE 154

142 total population 2008 % population under 18 2007 % population <65 years 2007 % females, 2007 % White, 2007 % Black, 2007 % Hispanic 2007 % AIAN, 2007 % Asian, 2007 % high school graduate or higher 2000 % bachelor's degree or higher 2000 Persons per household 2000 Median household income 2007 Population per square mile 2000 Hernando, FL 171,689 19.20 25.70 52.10 92.10 5.40 8.90 0.30 1.10 78.50 12.70 2.32 43,208.00 273.60 Hillsborough, FL 1,180,784 24.80 11.70 50.70 78.20 16.60 22.40 0.50 3.00 80.80 25.10 2.51 50,485.00 950.50 Indian River, FL 132,315 19.10 25.30 51.10 88.90 8.80 9.80 0.30 1.0 0 81.60 23.10 2.25 47,563.00 224.50 Jefferson, FL 14,547 18.80 14.80 45.70 62.70 35.40 3.40 0.50 0.50 73.20 16.90 2.53 40,217.00 21.60 Lafayette, FL 8,013 19.70 12.10 37.80 81.00 17.10 12.20 0.80 0.30 68.20 7.20 2.66 36,855.00 12.90 Lee, FL 593,136 21.00 22.20 50.50 89.30 7.90 17.20 0.40 1.30 82.30 21.10 2.31 50,750.00 548.40 Leon, FL 264,063 20.70 8.70 51.70 64.50 31.10 4.40 0.30 2.60 89.10 41.70 2.34 48,739.00 359.00 Levy, FL 39,460 21.90 18.40 51.50 87.10 10.80 5.50 0.40 0.50 73.90 10.60 2. 44 34,499.00 30.80 Manatee, FL 315,766 21.40 22.40 51.20 88.10 9.00 13.10 0.30 1.50 81.40 20.80 2.29 48,940.00 356.30 Martin, FL 138,660 18.30 26.10 50.40 91.10 5.90 10.00 1.00 1.00 85.30 26.30 2.23 55,229.00 227.90 Miami Dade, FL 2,398,245 22.80 14.80 51.50 77.20 19.80 62.00 0.40 1.50 67.90 21.70 2.84 43,495.00 1,157.90 Monroe, FL 72,243 15.80 15.60 46.80 91.60 5.40 18.50 0.50 1.30 84.90 25.50 2.23 55,054.00 79.80 Nassau, FL 69,835 22.20 14.90 50.50 89.50 8.10 2.50 0.40 0.90 81.00 18.90 2.59 56,500.00 88.40 Okaloosa, FL 179,693 23.80 13.20 50.00 83.50 9.90 5.70 0.60 2.80 88.00 24.20 2.49 54,633.00 182.20 Orange, FL 1,072,801 25.30 9. 60 50.20 72.30 20.80 24.30 0.50 4.40 81.80 26.10 2.61 50,988.00 988.30

PAGE 155

143 total population 2008 % population under 18 2007 % population <65 years 2007 % females, 200 7 % White, 2007 % Black, 2007 % Hispanic 2007 % AIAN, 2007 % Asian, 2007 % high school graduate or higher 2000 % bachelor's degree or higher 2000 Persons per household 2000 Median household income 2007 Population per square mile 2000 Osceola, FL 263,676 26.00 11.30 50.10 82.80 11.30 40.50 0.70 3.20 79.10 15.70 2.79 46,599.00 130.50 Palm Beach, FL 1,265,293 21.20 21.70 51.20 79.70 16.40 17.30 0.50 2.20 83.60 27.70 2.34 53,500.00 573.00 Pinellas, FL 910,260 19.00 20.80 51.90 85.10 10.30 7.00 0.40 2.80 84.00 22.90 2.17 44,325.00 3,291.00 Putnam, FL 73,459 23.50 18.80 50.70 80.90 17.00 8.20 0.50 0.60 70.40 9.40 2 .48 33,282.00 97.50 Santa Rosa, FL 150,053 23.30 11.90 50.00 89.70 5.40 3.90 0.90 1.80 85.40 22.90 2.63 50,935.00 115.80 Sarasota, FL 372,057 16.40 29.70 52.00 92.90 4.70 7.00 0.30 1.20 87.10 27.40 2.13 50,031.00 569.90 Seminole, FL 410,854 23.00 11.20 50.60 83.10 11.30 15.10 0.40 3.50 88.70 31.00 2.59 56,315.00 1,185.70 St. Johns, FL 181,540 20.90 14.60 50.80 90.40 6.40 4.50 0.30 1.90 87.20 33.10 2.44 63,728.00 202.20 St. Lucie, FL 265,108 22.70 19.60 50.80 79.30 17.40 15.20 0.30 1.60 77.70 15.10 2.47 46,127.00 336.90 Taylor, FL 21,546 21.80 15.40 48.80 78.00 19.20 1.90 1.00 0.60 70.00 8.90 2.51 38,056.00 18.50 Volusia, FL 498,036 19.70 20.40 51.00 86.50 10.50 10.30 0.40 1.50 82.00 17.60 2.32 42,268.00 401.90 Wakulla, FL 31,089 21.30 12.90 47.90 85.20 12.70 2.90 0.50 0.60 78.40 15.70 2.57 46,997.00 37.70 Walton, FL 53,837 21.00 15.20 49.00 89.10 7.40 3.40 1.10 0.70 76.00 16.20 2.35 45,288.00 38.40 Washington, FL 23,928 21.30 14.40 47.50 81.70 14.30 3.00 1.50 0.70 71.20 9.20 2.46 34,535.00 36.20

PAGE 156

144 Appendix C : C orrelation Matrix of Significant Variables in Bivariate Analysis 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 2 0 2 1 1 Correlation 1.000 p 2 Correlation .100 1.000 p .002 3 Correlation .022 .278 1.000 p .489 .000 4 Correlation .003 .171 .204 1.000 p .930 .000 .000 5 Correlation .026 .272 .179 .150 1.000 p .406 .000 .000 .000 6 Correlation .026 .291 .276 .269 .281 1.000 p .414 .000 .000 .000 .000 7 Correlation .014 .115 .112 .059 .147 .134 1.000 p .668 .000 .000 .063 .000 .000 8 Correlation .062 217 .058 .032 .105 .098 .058 1.000 p .054 .000 .070 .325 .001 .002 .070 9 Correlation .043 .203 .119 .008 .186 .152 .089 .105 1.000 p .178 .000 .000 .791 .000 .000 .005 .001 10 Correlation .037 .265 .139 .065 .160 .128 .169 .093 .132 1.000 p .248 .000 .000 .039 .000 .000 .000 .004 .000 11 Correlation .114 .066 .003 .007 .034 .005 .057 .032 .042 133 1.000 p .000 .043 .916 .829 .301 .877 .082 .342 .201 .000 12 Correlation .100 .187 .025 .079 .112 .049 .032 .057 .029 .034 .093 1.000 p .002 .000 .447 .014 .000 .128 .315 .081 .361 .2 87 .005 13 Correlation .038 .061 .059 .050 .093 .053 .122 .018 .017 .201 .084 .037 1.000 p .235 .055 .064 .116 .003 .091 .000 .582 .584 .000 .010 .253 14 Correlation .022 .121 .062 .017 .097 .05 5 .096 .017 .022 .026 .249 .125 1.000 p .620 .005 .152 .697 .025 .202 .026 .696 .604 .552 .000 .004 . 15 Correlation .036 .221 .069 .035 .131 .050 .019 .147 .133 .110 .040 .074 .023 .166 1.000 p .257 .000 .031 .264 .000 .118 .552 .000 .000 .001 .220 .022 .463 .000

PAGE 157

145 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 16 Correlation .030 .260 .105 .033 .131 .098 .016 .169 .156 .086 .005 .060 .028 .172 .426 1.000 p .345 .000 .001 .297 .000 .002 .615 .000 .000 .007 .885 .063 .382 .000 .000 17 Correlation .151 .050 .051 .030 .049 .054 .001 .064 .020 .102 .003 .009 .030 .045 .031 .011 1.000 . p .000 .120 .110 .339 .123 .092 .986 .048 .521 .001 .928 .77 4 .348 .301 .339 .735 . . 18 Correlation .052 .137 .076 .064 .033 .086 .002 .029 .007 .126 .014 .060 .038 .074 .039 .012 1.000 p .188 .000 .052 .101 .396 .028 .951 .474 .853 .001 .724 .133 .328 .163 .315 .764 . 19 Correlation .012 .072 .013 .021 .024 .000 .014 .007 .058 .126 .047 .095 .042 .047 .063 .003 . 1.000 p .782 .092 .766 .632 .583 .991 .751 .876 .177 .003 .288 .029 .326 .407 .140 .950 . 20 Correlation .040 .005 .059 .035 .060 .000 .041 .058 .077 .015 .274 .039 .008 .086 .061 .073 .024 .080 1.000 p .316 .898 .133 .377 .127 .993 .297 .149 .051 .706 .000 .332 .847 .106 .124 .066 .535 .066 21 Correlation .182 .131 .072 .030 .081 .067 .023 .031 .034 .063 .132 .176 .019 .014 045 .052 .115 .097 .097 .133 1.000 p .000 .000 .024 .354 .011 .036 .465 .346 .288 .048 .000 .000 .559 .755 .163 .108 .000 .014 .024 .001 22 Correlation .078 .010 .037 .038 .027 .036 .016 .102 .026 .083 .072 .001 .003 .170 .142 .101 .004 .025 .143 .242 .217 p .036 .796 .315 .305 .473 .334 .667 .007 .485 .026 .062 .987 .928 .001 .000 .007 .919 .582 .004 .000 .000 23 Correlation .124 .132 .001 .012 .088 .072 .019 .023 .034 .110 .125 .177 .004 .036 .036 .090 .050 .068 .103 .128 .265 p 000 .000 .972 .714 .009 .032 .577 .504 .310 .001 .000 .000 .909 .434 .289 .008 .146 .098 .022 .002 .000 24 Correlation .005 .070 .070 .047 .028 .116 .087 .049 .039 .092 .142 .027 .070 .010 .025 .003 .015 .030 .034 .068 .074 p .879 .029 .027 .138 .384 .000 .006 .128 .222 .004 .000 .402 .026 .812 .430 .934 .636 .451 .428 .083 .021 25 Correlation .026 .067 .002 .018 .040 .028 .110 .018 .016 .089 .110 .091 .115 .010 .028 .014 .053 .003 .106 .072 .030 p .420 .037 .947 .575 .209 .377 .001 .586 .621 .005 .001 .005 .000 .819 .387 .660 .096 .933 .014 .070 .353 26 Correlation .066 .137 .054 .014 .146 .151 .020 .003 .043 .069 .088 .056 .104 .072 .006 .026 .004 .005 .084 .030 .042 p .040 .000 .098 .655 .000 .000 .528 .925 .189 .033 008 .086 .001 .102 .864 .418 .891 .907 .054 .462 .200 27 Correlation .098 .142 .107 .099 .153 .149 .269 .160 .032 .194 .007 .056 .158 .087 .050 .030 .054 .026 .041 .026 .026 p .005 .000 .002 .004 .000 .000 .000 .000 .356 .000 .842 .111 .000 .063 .1 53 .387 .123 .548 .374 .540 .456 28 Correlation .028 .103 .044 .045 .032 .095 .162 .022 .037 .048 .143 .174 .146 .087 .046 .007 .056 .020 .104 .019 .136 p .381 .001 .161 .156 .312 .002 .000 .498 .238 .127 .000 .000 .000 .045 .144 .838 .077 601 .015 .636 .000 29 Correlation .042 .160 .096 .087 .177 .142 .256 .126 .092 .114 .006 .077 .110 .088 .015 .018 .093 .049 .045 .029 .035 p .260 .000 .009 .018 .000 .000 .000 .001 .012 .002 .884 .038 .003 .075 .678 .630 .012 .277 .359 .529 .343 (C ontinued)

PAGE 158

146 22 23 24 25 26 27 28 29 22 Correlation p 1.000 23 Correlation .075. 1.000 p .054 24 Correlation .050 .011 1.000 p .181 .740 25 Correlation .045 .017 .134 1.000 p 231 .614 .000 26 Correlation .015 .048 .157 .164 1.000 p .700 .169 .000 .000 27 Correlation .026 .077 .094 .176 .158 1.000 p .529 .037 .007 .000 .000 28 Correlation .035 .028 .299 .488 .240 .224 1.000 p .343 .407 .000 .000 .000 .000 29 Correlation .024 .059 .112 .134 .135 .807 .178 1.000 p .575 .132 .002 .000 .000 .000 .000 (Continued)

PAGE 159

147 Appendix C : Correlation Matrix of Significant Variables in Bivariate Analysis (continued) # in Correlati on Matrix # in Survey 1 1. How worried are you that a major hurricane will hit your community during the next 6 months? 1 2 2. Overall, how prepared are you if a major hurricane were to strike your community during the next 6 months? 2 3 I'm going to read you a list of things some people have in their homes that could be used in case of a hurricane emergency. Do you have A battery operated radio that you know works? 3a 4 Do you have A flashlight that you know works? 3b 5 Do you have A first aid kit? 3c 6 Do you have Extra batteries? 3d 7 Do you have A cell phone? 3e 8 Do you have At least $300 in cash to take with you if you had to leave your home? 3f 9 Do you have Water purifying supplies such as chlorine or iodine tablets 3g 10 Do you have A g enerator? 3h 11 If government officials said that you had to evacuate the area because there was going to be a major hurricane in the next few days, would you leave the area or would you stay? 11 12 If a major hurricane were to hit your community and for whatever reason you did not leave your home, how confident are you that you would be rescued if you needed to be? 19 13 Do you or any other household members have any pets in your home, such as dogs, cats, birds and the like? 20 14 If you had to evacuat e because of a hurricane, do you have a place you can go where you can take your pet, or not? 21 15 Has your family agreed on a phone number outside the region that all members of your immediate family could call in the event of a hurricane if you are una ble to communicate, or haven't you done that? 22 16 Has your family agreed on a place you could meet after a hurricane is over if you got separated and could not go back home, or haven't you done that? 23 17 Thinking back over the past three years was your community threatened or hit by a major hurricane, or not? 28 18 Was your community damaged by this hurricane, or not? 30 19 Was there major flooding associated with this hurricane in your community or not? 31 20 Because of this hurricane, did you lea ve your home where you lived, or did you stay in your home? 32 21 Thinking about where your home is located, how likely is your home to be flooded or damaged due to wind in a major hurricane? 37 22 Is your home located in an evacuation zone or not, or do n't you know if it is in an evacuation zone? 38 23 Do you think your home would withstand a major hurricane of Category 3 or higher without significant damage? 39 24 How long have you lived in your community? 45

PAGE 160

148 25 Are there any children under the age o f 18 living in your household? D02 26 Race Summary D07R 27 Income Summary D14c 28 Age Categories AgeCat 29 SES

PAGE 161

149 Appendix D: Correlation matrix of preparedness variables 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 1 1.000 2 100 ** 1.000 3 .022 .278 ** 1.000 4 .003 .171 ** .204 ** 1.000 5 .026 .272 ** .179 ** .150 ** 1.000 6 .026 .291 ** .276 ** .269 ** .281 ** 1.000 7 .014 .115 ** .112 ** .059 .147 ** .134 ** 1.000 8 -.062 .217 ** .058 .032 .105 ** .098 ** .058 1.000 9 .043 .203 ** .119 ** .008 .186 ** .152 ** .089 ** .105 ** 1.000 10 .037 .265 ** .139 ** .065 .160 ** .128 ** .169 ** .093 ** .132 ** 1.000 11 .100 ** .187 ** .025 .079 .112 ** .049 .032 .057 .029 .034 1.000 12 .182 ** -.131 ** -.072 -.030 -.081 -.067 -.023 -.031 -.034 -.063 -.176 ** 1.000 13 .078 .010 .037 .0 38 .027 .036 .016 .102 ** .026 .083 .001 .217 ** 1.000 14 .124 ** .132 ** .001 .012 .088 ** .072 .019 .023 .034 .110 ** .177 ** .265 ** .075 1.000 15 .005 .070 .070 .047 .028 .116 ** .087 ** .049 .039 .092 ** .027 .074 .050 .011 1.000 (Continued)

PAGE 162

150 Appendix D: Correlation matrix of preparedness variables (continued) Variable key. # in Correlation Matrix Variable question # in Survey 1 How worried are you that a major hurricane will hit your community during the next 6 month s? 1 2 Overall, how prepared are you if a major hurricane were to strike your community during the next 6 months? 2 3 I'm going to read you a list of things some people have in their homes that could be used in case of a hurricane emergency. Do you have A battery operated radio that you know works? 3a 4 Do you have A flashlight that you know works? 3b 5 Do you have A fi rst aid kit? 3c 6 Do you have Extra batteries? 3d 7 Do you have A cell phone? 3e 8 Do you have At least $300 in cash to take with you if you had to leave your home? 3f 9 Do you have Water purifying supplies such as chlorine or iodine tablets 3g 10 Do you have A generator? 3h 11 If a major hurricane were to hit your community and for whatever reason you did not leave your home, how confident are you that you would be rescued if you needed to be? 19 12 Thinking about where your home is located, how lik ely is your home to be flooded or damaged due to wind in a major hurricane? 37 13 Is your home located in an evacuation zone or not, or don't you know if it is in an evacuation zone? 38 14 Do you think your home would withstand a major hurricane of Categ ory 3 or higher without significant damage? 39 15 How long have you lived in your community? 45

PAGE 163

About the Author Homer J. Rice was born in Ohio but raised in Florida. He received his B.S. in Biological Sciences from Florida State University and an M .P.H. in Environmental and Occupational Health from the University of South Florida. Homer has worked in the Department of Health for the State of Florida since 1978, beginning his career in Environmental Health at the Sarasota County Health Department. During his career with the Department of Health he has responded to multiple disasters in various counties in the state. He was stationed in Homestead after Hurricane Andrew and coordinated the environmental health response in five counties after Hurricane Charley. He worked with the Centers for Disease Control in the development and teaching of the Environmental Health Training in Emergency Response (EHTER) course and to develop guidelines for emergency sheltering in radiological events. Homer is current ly the Administrator of the Leon County Health Department in Tallahassee, Florida. He serves as the Health and Medical cochair for the Region Two Domestic Security Task Force, working with law enforcement and emergency responders to plan, train, and prepare against terrorism and disaster.


xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim
leader nam 22 Ka 4500
controlfield tag 007 cr-bnu---uuuuu
008 s2010 flu s 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0004706
035
(OCoLC)
040
FHM
c FHM
049
FHMM
090
XX9999 (Online)
1 100
Rice, Homer.
0 245
Before the storm :
b evacuation intention and audience segmentation
h [electronic resource] /
by Homer Rice.
260
[Tampa, Fla] :
University of South Florida,
2010.
500
Title from PDF of title page.
Document formatted into pages; contains X pages.
502
Dissertation (PHD)--University of South Florida, 2010.
504
Includes bibliographical references.
516
Text (Electronic thesis) in PDF format.
538
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
3 520
ABSTRACT: The purpose of this study was to describe the predictors of evacuation intention among coastal residents in the State of Florida and to determine if there are meaningful segments of the population who intend to evacuate when told to do so by governmental officials because of a major hurricane. In the America's and the Caribbean, 75,000 deaths have been attributed to hurricanes in the 20th century. A well planned evacuation can reduce injury and death, yet many people do not have an evacuation plan and do not intend to evacuate when told to do so. The study used secondary data from the Harvard School of Public Health, Hurricane in High Risk Areas study, a random sample of 5,046 non-institutionalized persons age 18 and older in coastal counties of Texas, Louisiana, Mississippi, Alabama, Georgia, North Carolina, South Carolina and Florida. Surveys for the State of Florida were segregated and used in this analysis, resulting in a study sample of 1,006 surveys from 42 counties. When asked if they would evacuate in the future if told to by government officials, 59.1% of Floridians surveyed said they would leave, 35.2% said they would not leave and 5.6% said it would depend. In Florida, 65.7% of the population had been threatened or hit by a major hurricane in the last three years and 26.6% of those had left their homes because of the hurricane. Of those whose communities were threatened by a hurricane, 83.3% of the communities were damaged and 33.8% experienced major flooding associated with the hurricane. Bivariate statistics and logistic regression were used to explore the interactions of predictors and evacuation intention. The best predictor of evacuation intention was prior evacuation from a hurricane (chi-square= 45.48, p < .01, Cramer's V = 0.266). Significant relationships were also demonstrated between evacuation intention and worry a future hurricane would hit the community (chi-square = 22.75, p < .01, Cramer's V = 0.11), the presence of pets (chi-square = 6.57, p < .01, Cramer's V = 0.084), concern the home would be damaged (chi-square = 19.41, p < .01, Cramer's V = 0.10), belief the home would withstand a major hurricane (chi-square = 19.55, p < .01, Cramer's V = 0.10), length of time in the community (chi-square = 26.59, p < .01, Cramer's V = 0.12), having children in the household (chi-square = 11.13, p < .01, Cramer's V = 0.11), having a generator (chi-square = 17.12, p < .01, Cramer's V = 0.13), age (chi-square = 24, p < .01, Cramer's V = 0.16) and race (chi-square = 12.21, p = .02, Cramer's V = 0.12). Logistic regression of the predictors of evacuation intention resulted in significant relationships with previous evacuation experience (OR = 4.99, p < .001), age 30 to 49 compared to age over 65 (OR = 2.776, p < .01), the presence of a generator (OR = .447, p < .01), having a home not very likely to be damaged compared to a home very likely to be damaged (OR =.444, p = .018), and experiencing poor prior government and voluntary agency response to previous hurricanes compared to excellent response (OR = .386, p < .027). Chi-squared Automatic Interaction Detection (CHAID) was used to identify segments of the population most likely and least likely to evacuate when told to do so. Those most likely to evacuate had evacuated due to a previous hurricane. Those least likely to evacuate when told to do so had not evacuated in a previous storm, do not own a generator and are over the age of 65. Information from this study can be used in planning for evacuation response by governmental entities. Available demographic information can be used to determine numbers of persons likely to evacuate before a storm. The results of this study can be used to inform a marketing strategy by government officials to encourage evacuation among those who say they would not evacuate when told to do so. Further research is needed to determine additional characteristics of the populations who say they will and will not evacuate when told to do so.
590
Advisor: Carol Bryant, Ph.D.
653
Disaster
Planning
Hurricanes
Chi-Square
CHAID
690
Dissertations, Academic
z USF
x Dean's Office
Masters.
773
t USF Electronic Theses and Dissertations.
4 856
u http://digital.lib.usf.edu/?e14.4706