Analysis of Flood Evacuation Mode and Route Choice Behavior of Households in a Developing Country

ANALYSIS OF FLOOD EVACUATION MODE AND ROUTE CHOICE BEHAVIOR OF HOUSEHOLDS IN A DEVELOPING COUNTRY

HECTOR RUIZ LIM, JR.

A DISSERTATION SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY (ENGINEERING AND TECHNOLOGY) SIRINDHORN INTERNATIONAL INSTITUTE OF TECHNOLOGY THAMMASAT UNIVERSITY ACADEMIC YEAR 2017

Ref. code: 25605522300184DMX ANALYSIS OF FLOOD EVACUATION MODE AND ROUTE CHOICE BEHAVIOR OF HOUSEHOLDS IN A

DEVELOPING COUNTRY

HECTOR RUIZ LIM, JR.

Ref. code: 25605522300184DMX

Abstract

ANALYSIS OF FLOOD EVACUATION MODE AND ROUTE CHOICE BEHAVIOR OF HOUSEHOLDS IN A DEVELOPING COUNTRY

HECTOR RUIZ LIM, JR.

Bachelor of Science in Chemical Engineering, Cagayan State University, 2002 Master of Engineering, Asian Institute of Technology, 2010 Doctor of Philosophy (Engineering and Technology), Sirindhorn International Institute of Technology, Thammasat University, 2017

Increasing frequency and severity of hazards that lead to devastating disaster impacts demand building substantial response capability. Evacuation is seen as one of the effective measures to avert disaster impacts. Planning and modeling of effective evacuation incorporate evacuation travel behavior. Evacuation modeling usually done in a sequential manner follows the classic four-step travel demand modeling. Essential decisions based from the classic four-step travel demand model include evacuation decision, departure time choice, destination choice, mode choice and route choice. Evacuation decision is the choice of households at risk to either leave their place and move to a safe place or stay at home. Departure time choice describes when the household actually leaves the area at risk. Destination choice describes where the households go when leaving the area at risk. Mode choice describes which mode of transport is preferred when leaving the area at risk. And route choice describes what route evacuees take when moving from area at risk to their chosen destination. All these decisions involve complex behavioral factors influencing each household of various characteristics and situations at the period of

iii

Ref. code: 25605522300184DMX

choosing (e.g. Simonovic and Ahmad, 2005). This study seeks to investigate the mode and route choice of evacuees as a continuation of the study made by M.B Lim (2016). Understanding the mode and the route choices of evacuees use when evacuating has implications on the traffic flow in road network available during evacuation. This however, are affected by certain factors such as household socio- demographic, travel related decisions and hazard-related information, among others. Identifying and understanding how these factors affect mode and route choices of evacuees can help government officials develop strategies whenever a disaster happens such who to prioritize, investments for disaster management that makes evacuation effective and efficient, resulting to reduction of casualties. There are limited studies on understanding the mode choice behavior. Sadri et al. (2014a) emphasized the need for more studies to investigate and reveal other factors that are important to evacuees’ decision making. Also, earlier evacuation planning studies were car-based and transit-based (e.g. Murray-Tuite, 2007; Balakrishna et al., 2008; Chen and Xiao, 2008; Klunder et al., 2009; Mastrogiannidou et al., 2009; Noh et al., 2009; Chan, 2010; Huibregste et al., 2010; Wang et al., 2010; Pel et al., 2011; An et al., 2013). Case studies in most developing countries are still lacking socio- demographic characteristics of the population, hazard-related information, and evacuees’ travel related decisions, in addition to available transportation infrastructures that are unique. Majority of people in developing countries do not have personal vehicles. They depend greatly on available modes of transport for evacuation. It was also recommended by Abdelgawad and Abdulhai (2010a) in their large scale evacuation multimodal study that modes such as walking and cycling that are readily available in urban cities could be integrated to evacuation planning. Allowing people to evacuate on foot is believed to be faster than vehicular evacuation within two-kilometer region (Shiwakoti et al., 2013). This is an important mode of evacuation at the onset of disasters especially when road networks are congested. For route choice, previous studies indicate that one of the core aspects of evacuation is routing. One of the impetus for this is that the travel demand during emergency evacuation exceed the capacity of the transportation networks (Pel et al., 2010), and congestion is likely to happen. Thus, a number of researches focused on

Ref. code: 25605522300184DMX

optimal evacuation considering different strategies on routes for evacuees to reach safety, were evaluated through simulation models (e.g. Ng and Waller, 2010; Sayyady and Eksioglu, 2010; Campos et al., 2012; Na et al., 2012; Bish and Sherali, 2013). Evaluated strategies with substantial reduction of clearance time against the lead time of the hazard are translated to evacuation plan. However, simulation models do not necessarily capture behaviors of the decision makers. This indicates that empirical studies on observed route choices of evacuees are then valuable. Travel behavior could be incorporated in evacuation simulations in order to identify optimal routing strategies (Fang and Edara, 2013). This recommendation is also consistent with previous studies that recognized route decisions of evacuees as an important part of evacuation simulation/traffic modeling (e.g. Dow and Cutter, 2002; Pel et al., 2010; Pel et al., 2011). Research attempts to fill this gap is done by shifting efforts to understanding how evacuees choose the route they take during evacuation (e.g. Akbarzadeh and Wilmot, 2015; Sadri et al., 2014). Researchers are becoming more adept to solve evacuation problems that will generalize how evacuees (e.g. individuals/households) decide on certain aspects (e.g. evacuation decision, departure time, destination, mode, and route) whenever a disaster strikes. However, in places where household characteristics are heterogeneous, households’ decisions might considerably vary. Household decisions can be affected by culture, demography, and environment, among others. Hence, it is an imperative to understand household’s preferences in different communities in making decisions during certain events such as evacuation. In line with the above gaps, the main goal of this study to understand determinants of household evacuation mode and route choice behavior. Understanding the variables that affect evacuees’ choice can assist emergency planners in preparing for future evacuation plans, for instance in determining what mode to use for evacuation, which route can be congested, and which routes to recommend for evacuees to take in order to reach destination in a timely manner. This study then seeks to identify and understand the determinants of households' mode and route choices in a developing country, particularly the Philippines. Discrete choice models were estimated and validated from original data collected in selected sub-

Ref. code: 25605522300184DMX

districts in Quezon City, Philippines. Face to face interviews were conducted with randomly selected households in sub-districts. Random selection of households was done using the cluster sampling technique. During the interview process, the respondents were given a brief introduction on the study being conducted. This was to ensure they understand the context which is the basis of their answers to the questions. The interviewers also made sure that the household experienced flood during the Mid-August 2013 before proceeding with the interviews. To do this, interviewees were first inquired if they experienced flood during the 2013 event. The flood details including the level of flood in their house, the number of days they were flooded, the level of damage the flood caused their house, and whether they received evacuation warning and its source. Information on the distance of house location from the flood hazard was also solicited. Then, information on evacuation-related decisions was also elicited including the type of evacuation decision, the timing of evacuation, their evacuation destination, the mode they used and the route they took when evacuating. Routing strategies available to households were identified and included in the survey instrument. The route recommended by the government, usually available in other studies was not made available to households in this current study during issuance of evacuation warning. During the interviews, a map of possible routes to identified evacuation centers were shown from which interviewees were asked which route they took during evacuation to their specific destinations. They were then asked of the reason of taking such route rather than taking other available ones. From these, three routing options were identified including the route that they usually take during normal days (most familiar route to them that was also confirmed as not the nearest to their destination), the nearest route or where they thought they can take to evacuate faster, and the route without flood waters. These were the list of options made available to households during the rest of interviews in addition to “others” in case of the existence of route option not identified during the pilot survey. After data collection, however, the third option was removed from analysis as it was found that such route was not available to them that time or it is the only route available. In addition, there were no “other” route options made available in the data collected. The

Ref. code: 25605522300184DMX

2 remaining routes are then used for analysis in this study. These route options are defined as nearest and familiar (the main road most familiar that they usually take during normal days but not the nearest). The second part of the interview solicited suggestions and comments for better situations in future evacuation from floods. The third part of the interview aided solicitation of socio-demographic information of the head of the household and other household information that includes age, gender, marital status, educational attainment and type of work of the head of the household, presence of health problem and insurance, household monthly income, number of household members, age of members, the presence of small children and senior citizens, number of years the household has been living in that residence, type of house ownership and materials, the number of house floor levels, vehicle ownership, and pet ownership. Six hundred thirty two interviews were completed out of 640 total number of households approached. Out of the 632 interviews, 340, 150, and 142 were from Bagong Silangan, Bahay Toro, and Sto. Domingo, respectively. All the data collected was tabulated in excel sheet. The data was verified and cross checked by ensuring that information provided by the households were according to the questions asked. Data obtained with a lot of missing information and inconsistency based from the needed information was excluded from the analysis. Also, data from households that decided not to evacuate was not included in the analysis. These were then processed to and the logit models used for analysing the mode and route choices of evacuees. Findings revealed important determinants that can help evacuation planners and managers develop strategies for future flood evacuation operations. For the mode that household choose, those who traveled on-foot take into account departure timing, destination type, age, gender and educational attainment of the head of the household, presence of small children, presence of health problem, house ownership, number of years living in the residence, vehicle ownership, source of warning, distance traveled to safety and cost of evacuation. These results provide insights that can be useful for the government to plan for future evacuations. For instance, the government can encourage the households with personal vehicle to use them in future evacuations, while providing for those without personal vehicle and

vii

Ref. code: 25605522300184DMX

needs to travel longer distances. The government can also encourage households living in high flood risk areas to prepare household evacuation plan. In terms of the route that households choose, the level of flood, vehicle ownership, mode chosen, departure timing, and house ownership show influence to decision making. Based on results, planners can analyze and evaluate appropriate actions such as which timing, evacuation routes, and destinations to advice households at risk in case of future floods. Given findings in this study, pedestrian and gender-based evacuation behavior can be modeled and included in the evacuation simulation framework understanding key determinants of gender-based evacuation travel behavior in the area of evacuation, departure time, destination, mode and route choice behavior can be investigated in the future. Increasing sample size and collecting data in other sub- districts may be helpful for further investigation. While transferability of estimated models to other hazards such as earthquake, is an area worthy to study.

Keywords: Evacuation, Modeling, Determinants, Route Choice, Mode Choice, Flood

viii

Ref. code: 25605522300184DMX

Acknowledgements

I would like to extend my sincere thanks to Sirindhorn International Institute of Technology (SIIT), Thammasat University, Thailand for the financial support for my Doctoral studies. This support would not be possible without the efforts and persistence of my adviser Dr. Mongkut Piantanakulchai. Additionally, I would like to thank Dr. Piantanakulchai for his support and guidance throughout my Doctoral studies. I am also thankful for my external reviewer Prof. Junyi Zhang and the members of Thesis Committee namely Dr. Pruettha Nanakorn, Dr. Chawalit Jeenanunta, Itthisek Nilkhamhang, and Dr. Sorntep Vannarat for their valuable comments and suggestions for the completion of my dissertation. Special thanks to my wife Ma Bernadeth Lim, my children David Joshua Lim and Luke Joseph Lim who serve as inspirations for me to persevere and joyfully complete my studies amidst of all the challenges and difficulties along the way. Lastly, I would like to praise and thank God for his divine providence and guidance.

Ref. code: 25605522300184DMX

Table of Contents

Chapter Title Page

Signature Page ii Abstract iii Acknowledgements ix Table of Contents x List of Tables xiii List of Figures xiv

1 Introduction 1

1.1 Rationale 1 1.2 Statement of the Problem 3 1.3 Objectives of the Study 5 1.4 Significance of the Study 6 1.5 Scope and Limitation 7 1.6 Outline of Dissertation 8

2 Literature Review 10

2.1 Evacuation and Evacuation Process 10 2.2 Evacuation Planning and Evacuation Travel Behavior 10 2.2.1 Socio-demographic information 15 2.2.2 Hazard information and evacuation warning 16 2.3 Evacuation Mode Choice Behavior and Related Determinants 18 2.4 Evacuation Route Choice and Related Determinants 20 2.5 Summary and Chapter Conclusions 24

3 Methodology 27

Ref. code: 25605522300184DMX

3.1 Research Framework 27 3.2 Study Area 28 3.3 Data Collection 30 3.4 Method Used for Variable Selection in both Mode and Route Choice Analysis 33 3.5 Modeling Framework for Mode and Route Choice Analysis and Model Estimation 33 3.6 Validation of Models 36 3.6.1 Receiver operating characteristics, area under curve and correct classification rates 36 3.6.2 Likelihood ratio-based validation test 37

4 Factors Determining the Evacuation Mode Choice of Residents in High Flood Risk Areas 38 4.1 Summary of Data and Descriptive Statistics of Variables for Mode Choice 38 4.2 Resulting Inter-Correlations of Mode Choice Model Variables 41 4.3 Estimated Parameters for Mode Choice Models 41 4.3.1 Model 1: Bagong Silangan 47 4.3.2 Model 2: Bahay Toro 47 4.3.3 Model 3: Sto. Domingo 48 4.3.4 Model 4: Pooled Model for Bahay Toro and Sto. Domingo 49 4.3.5 Model 5: Pooled Model for all sub-districts 50 4.4 Comparisons of Results of Estimated Mode Choice Models 54 4.5 Validation of the Estimated Mode Choice Models 52 4.6 Summary of Findings on the Mode Choice Behavior of Evacuees 53

5 Understanding Route Choice Behavior of Selected Population in Quezon City 55 5.1 Data Used for Route Choice Analysis and Resulting Descriptive Statistics 55

Ref. code: 25605522300184DMX

5.2 Inter-Correlation of Variables Included in the Route Choice Models 58 5.3 Resulting Parameter Estimates for Models of Route Choice 59 5.3.1 Model 6: Bagong Silangan 64 5.3.2 Model 7: Bahay Toro 64 5.3.3 Model 8: Sto Domingo 65 5.3.4 Model 9: Bahay Toro + Sto. Domingo 65 5.3.5 Model 10: Bagong Silangan + Bahay Toro + Sto. Domingo 66 5.4 Comparison of Results of Estimated Route Choice Models 67 5.5 Validation of Estimated Route Choice Models 68 5.6 Summary of Findings on Route Choice Behavior of Evacuees 69

6 Conclusions and Recommendations 72

6.1 Conclusions 72 6.1.1 Flood evacuation mode choice 72 6.1.2 Flood evacuation route choice 73 6.2 Recommendations and future research 77 6.2.1 Flood evacuation mode choice 74 6.2.2 Flood evacuation route choice 76

References 78 Appendices 90

Appendix A. Survey Questionnaire 91

xii

Ref. code: 25605522300184DMX

List of Tables

Tables Page 2.1 Route categories used in evacuation route choice literature 21 4.1 Variables included in each model, categories and corresponding percentage in the data 39 4.2 Inter-correlations of variables included in the logit models 42 4.3 Parameter estimates of models for households that walked in evacuating evacuation 44 5.1 Variables included in each model, categories and percentage 56 5.2 Inter-correlation of variables included in the models 60 5.3 Parameter estimates of models for households that evacuates using the nearest route 61

xiii

Ref. code: 25605522300184DMX

List of Figures

Figures Page 2.1 Evacuation Process Model 11 3.1 Research Framework 27 3.2 Flood map of Quezon City with number of households interviewed 30

xiv

Ref. code: 25605522300184DMX

Chapter 1 Introduction

1.1 Rationale

The past decade marks a period where large scale disasters caused devastating impacts in the world such as damage to properties, killed and affected millions of people (Gencer, 2013). These were exemplified by Typhoon Ketsana and Haiyan in 2009 and 2013 respectively (EM-DAT, 2014). The impact of disasters will continue to increase with the rising population and economic activities. As a result, governments all over the world are working on strengthening policies and programs to build resilience to the impact of disasters.

In March 2015, during the 3rd World Conference on Disaster Risk Reduction (WCDRR) in Japan, marks the adoption of Sendai Framework for Disaster Risk Reduction 2015-2030 (UNISDR, 2015). This new international agreement promotes greater disaster resilience in countries around the world, succeeding the Hyogo Framework for Action 2005-2015 (UNISDR, 2012) that helped governments establish comprehensive national disaster risk management strategies and legislations, resulting to a significant change from emergency response approach to increasing focus on disaster risk reduction (DRR). One of the four priorities for action in the Sendai Framework for DRR is enhancing disaster preparedness for effective response. One effective preparedness measure that helps minimize, if not eradicate, loss of lives is evacuation.

The Philippine government in particular strengthened its national disaster risk reduction system through the Republic Act 10121 (RA 10121) also known as the Philippine Disaster Risk Reduction and Management Act of 2010. Evacuation is one of the overt disaster preparedness measures in this law. Local governments are responsible to recommend forced and preemptive evacuation to local residents whenever they face hazards such as flooding or typhoon-induced flooding. Recently, 1

Ref. code: 25605522300184DMX

the local governments through the Community-Based Disaster Risk Management (CBDRM) program initiated the Flood Local Early Warning System (FLEWS) where evacuation is an integral part of the system. This effort is widely adapted and implemented in several cities and municipalities nationwide. However, evacuation researches to improve these programs and policies are imperative.

Effective evacuation depends greatly in careful planning which is being fleshed out in an evacuation plan. Evacuation strategies can be evaluated through models that demonstrate probable evacuation scenarios during the actual operations. In modeling evacuation scenarios, consideration of travel behavior of evacuees is deemed important (e.g. Sorensen 2000; Pel et al., 2010; Pel, et al., 2012; Fang and Edara, 2013; Huang et al., in press).

Evacuation travel behavior is an area that needs to be explored in evacuation modeling and management for major cities in different countries regardless of their economic status and geographical location. Evacuation travel behavior includes evacuation decision, departure time choice, destination choice, mode choice and route choice. Evacuation travel behavior is identified as: the decision to evacuate or stay, departure time, destination and shelter type, mode, and route choices. These have been the focus of evacuation modeling research (e.g. Sinha and Avrani, 1984; Revi and Singh, 2007; Paul and Dutt 2010; Paul 2012; Hasan et al. 2011; Mesa-arango et al., 2013; M.B. Lim et al., 2015a). Mode choice behavior of evacuees has been investigated in several studies (e.g. Perry et al., 1981; Lindell and Perry, 1992; Baker, 2000; Dash and Morrow, 2001; Dow and Cutter, 2002; Lindell and Prater, 2007; Siebeneck and Cova, 2008; Lindell et al., 2011; Wu et al., 2012; Wu et al., 2013; Sadri et al., 2014a). Investigation of the determinants of evacuation mode and route choice is the primary focus of this study.

Mode and route choice are essential parts of evacuation travel demand forecasting especially in evacuation traffic simulation. These allows planners and disaster managers to understand the preference of evacuating households on the mode and 2

Ref. code: 25605522300184DMX

route they use. Recognizing that these are significant inputs to the evacuation traffic simulation, strategies can be derived and proposed for the management of both demand and supply side of evacuation operations.

1.2 Statement of the Problem

Understanding the mode and the route choices of evacuees use when evacuating has implications on the traffic flow in road network available during evacuation. This however, are affected by certain factors such as household socio-demographic, travel related decisions and hazard-related information, among others. Identifying and understanding how these factors affect mode and route choices of evacuees can help government officials develop strategies whenever a disaster happens such who to prioritize, investments for disaster management that makes evacuation effective and efficient, resulting to reduction of casualties. In this section, I would like to attempt discussing some of the problems why we have to investigate mode and route choice behavior.

For mode choice, in one of the limited studies on understanding the mode choice behavior of residents in Miami Beach, United States of America (USA) during hurricane evacuation, Sadri et al. (2014a) emphasized the need for more studies to investigate and reveal other factors that are important to evacuees’ decision making. Also, earlier evacuation planning studies were car-based and transit-based (e.g. Murray-Tuite, 2007; Balakrishna et al., 2008; Chen and Xiao, 2008; Klunder et al., 2009; Mastrogiannidou et al., 2009; Noh et al., 2009; Chan, 2010; Huibregste et al., 2010; Wang et al., 2010; Pel et al., 2011; An et al., 2013). These developments on mass transit and car-based evacuation modeling; however, are large scale evacuations in developed countries such as the Netherlands and USA. Case studies in most developing countries are still lacking socio-demographic characteristics of the population, hazard-related information, and evacuees’ travel related decisions, in addition to available transportation infrastructures that are unique. Majority of people in developing countries do not have personal vehicles. They depend greatly on 3

Ref. code: 25605522300184DMX

available modes of transport for evacuation. It was also recommended by Abdelgawad and Abdulhai (2010a) in their large scale evacuation multimodal study that modes such as walking and cycling that are readily available in urban cities could be integrated to evacuation planning. Allowing people to evacuate on foot is believed to be faster than vehicular evacuation within two-kilometer region (Shiwakoti et al., 2013). This is an important mode of evacuation at the onset of disasters especially when road networks are congested. The earthquake and tsunami reported by Lindell et al. (2015) prompted pedestrian evacuation as well as vehicular evacuation.

For route choice, previous studies indicate that one of the core aspects of evacuation is routing. One of the impetus for this is that the travel demand during emergency evacuation exceed the capacity of the transportation networks (Pel et al., 2010), and congestion is likely to happen. Thus, a number of researches focused on optimal evacuation considering different strategies on routes for evacuees to reach safety, were evaluated through simulation models (e.g. Ng and Waller, 2010; Sayyady and Eksioglu, 2010; Campos et al., 2012; Na et al., 2012; Bish and Sherali, 2013). Evaluated strategies with substantial reduction of clearance time against the lead time of the hazard are translated to evacuation plan. However, simulation models do not necessarily capture behaviors of the decision makers. For example, Fang and Edara (2013) investigated that travel times realized during an evacuation is underestimated. This indicates that empirical studies on observed route choices of evacuees are then valuable. Travel behavior could be incorporated in evacuation simulations in order to identify optimal routing strategies (Fang and Edara, 2013). This recommendation is also consistent with previous studies that recognized route decisions of evacuees as an important part of evacuation simulation/traffic modeling (e.g. Dow and Cutter, 2002; Pel et al., 2010; Pel et al., 2011).

Recently, research attempt to fill this gap is done by shifting efforts to understanding how evacuees choose the route they take during evacuation (e.g. Akbarzadeh and Wilmot, 2015; Sadri et al., 2014). Researchers are becoming more adept to solve evacuation problems that will generalize how evacuees (e.g. individuals/households) 4

Ref. code: 25605522300184DMX

decide on certain aspects (e.g. evacuation decision, departure time, destination, mode, and route) whenever a disaster strikes. However, in places where household characteristics are heterogeneous, households’ decisions might considerably vary. Household decisions can be affected by culture, demography, and environment, among others. Hence, it is an imperative to understand household’s preferences in different communities in making decisions during certain events such as evacuation.

In line with the above gaps, this study aims to understand determinants of household evacuation mode and route choice behavior. Understanding the variables that affect evacuees’ choice can assist emergency planners in preparing for future evacuation plans, for instance in determining what mode to use for evacuation, which route can be congested, and which routes to recommend for evacuees to take in order to reach destination in a timely manner. In this study, variables were identified based on previous evacuation travel behavior studies. The variables included in the analysis are household socio-demographic information, hazard-related and travel-related information.

1.3 Objectives of the Study

Based on the background and statement of the problem discussed in earlier sections above, this study seeks to know the determinants of households’ flood evacuation mode and route choice behavior of households in the Philippines. In particular, Bagong Silangan, Bahay Toro and Sto Domingo Sub-districts of Quezon City, Metro Manila during the flood events in Mid-August of 2013. The flood was brought by the heavy monsoon rains and exacerbated by the Tropical Storm Trami.

The objective of this study is to understand the determinants that explains flood evacuation mode and route choices of selected households in Quezon City, Philippines. Specific objectives are to: • identify and understand the determinants of household flood evacuation mode and route choices;

Ref. code: 25605522300184DMX

• estimate and validate models of evacuation mode choice and route choice of each sub-district level as well as using pooled data of the whole district; • provide policy recommendations for planning future flood evacuations.

Based on the determinants to mode and route choice behaviors identified in earlier conducted evacuation behavior studies, in addition to some respondent related characteristics, this study will explore which are most important to people in the context of the Philippines. These, will be analyzed using the discrete choice modeling framework. The determinants will be identified first, then estimated models will be validated for potential policy recommendations.

1.4 Significance of the Study

This study contributes to the evacuation travel behavior modeling research, particularly, with consideration to flood hazard. It is a significant step towards understanding evacuation travel behavior in the context of Asian developing countries.

Specifically, this study identifies determinants that influence travel behavior of households when at risk of impending hazard. It is an effort towards bringing together sociologists, evacuation managers and transportation planners in an endeavor to work together for better evacuation planning. Before implemented in practice, research needs to prove that integration of the determinants from the viewpoint of these researchers can contribute to better understanding what is really happening during emergencies, hence better evacuation planning. Last but not least, this study is an initial step towards development of comprehensive evacuation plans in the onset of flood evacuation in the Philippines.

It is envisaged that the government of Quezon City can use the output of this study for the development for sub-district level and city level evacuation plans. This study is primarily helpful for the government to determine possible distribution of evacuation

Ref. code: 25605522300184DMX

mode and the probable evacuation route. These will practically help the local government to do the following: • allocate government vehicles for evacuees in need prior and/or during a flood disaster, • procurement of additional vehicles, coordination with public utility jeepneys and/or buses for evacuation purposes, • identify and install proper marking for evacuation routes leading to identified evacuation centers, and • allocation and coordination of traffic guides during evacuations

1.5 Scope and Limitation

This study focuses on understanding the determinants of travel decisions made at household level during evacuation. This study focuses on the demand side travel behavior that includes mode choice and route choice defined as follows: • Mode choice is the decision of households to either walk or use other modes such as personal vehicle, rented vehicle or public and government vehicle. Data used for analysis were collected from households in selected sub-districts in Quezon City, Philippines. Specifically, the behavior of walking evacuees in the event of flood was investigated. Other modes that were available to evacuees and the determinants that they considered upon choosing certain evacuation mode were also taken into account. • Route choice is the decision of households to take either the nearest (shortest route available to them) or familiar (defined as the usual main road taken during normal days but not the nearest which was identified by the evacuees).

This study does not cover the evacuation decision, departure time choice and destination choice of households. This study is a further investigation after a study covering evacuation decision, departure time choice and destination choice of households (see M.B. Lim, 2016). These choices were also treated as travel related exogenous variables in the analysis of this current study. Models were also developed 7

Ref. code: 25605522300184DMX

to quantify travel behavior. Households not living in high flood risk areas are not included in the analysis due to unavailable data.

1.6 Outline of Dissertation

This section introduces the content of the manuscript. It outlines the chapters and sections included here. The manuscript consists of 6 chapters. These are as follows:

Chapter 1. Introduction. This chapter presents the background and motivation of conducting this study, the objectives, study significance and the scope and limitation.

Chapter 2. Literature Review. This section gives an overview and relationship of evacuation planning and evacuation mode and route choices. It also provides a summary of the determinants that were investigated in the past studies for evacuees’ mode and route choices.

Chapter 3. Methodology. This chapter presents the methodology employed in conducting this research. Specifically, the study area, methods used for data collection, model estimation and model validation are presented in this chapter. The modelling framework used for analysis of mode and route choices is the discrete choice.

Chapter 4. Factors Determining the Evacuation Mode Choice of Residents in High Flood Risk Areas. This chapter presents the data used for analysis of the mode choice. It also discusses and presents the parameter estimates for each model, results of the validation and comparison of models.

Chapter 5. Understanding Route Choice Behavior of Selected Population in Quezon City. This chapter shows the analysis results for the route choice. First, the data is discussed followed by the variable selection. Then the parameter estimates as well as model comparisons are discussed. 8

Ref. code: 25605522300184DMX

Chapter 6. Conclusions and Recommendations. This is the last chapter that summarizes and presents what can be taken from the results of the study. Also, ways of how the results of this study can be used, in addition to some policy recommendations are done at the end.

Ref. code: 25605522300184DMX

Chapter 2 Literature Review

This section gives an overview and relationship of evacuation planning and evacuation mode and route choices. It also provides a summary of the determinants that were investigated in the past studies for evacuees’ mode and route choices.

2.1. Evacuation and Evacuation Process

Evacuation is an effective measure to minimize damages and losses in any disaster event. Evacuation means moving people at risk to safety (Na, Xueyen, and Mingliang, 2012) especially when preparedness measure is more appropriate due to the difficulty of retrofitting physical infrastructures in case of earthquake and implementing mitigation measures such as flood control system in case of flood. Evacuation follows a basic process that includes detection of hazard, warning, withdrawal, shelter and return-entry (EMA, 2005; Stepanov and Smith, 2009). This process involves important household decisions. Complex factors relevant to decisions include socio- demographic characteristics, hazard detection, issuance of warning and evacuation planning decisions (M.B. Lim, Lim and Piantanakulchai, 2013).

The figure 2.1 shows an integrated process model for evacuation process (EMA, 2005; Stepanov and Smith, 2009). The model shows important aspects of the models from where it came from, and emphasizes some essential logistical considerations in evacuation planning. This model also captures some complex decision making by both authorities and households at risk of hazard.

2.2 Evacuation Planning and Evacuation Travel Behavior

Evacuation planning is a proactive approach by modeling the hazard occurrence, redesigning and identifying shelters, conducting regular trainings to make sure that people understand their roles and know what to do, establishing contacts among all organizations involved in emergency management, conducting plan audits to ensure

Ref. code: 25605522300184DMX

procedures remain appropriate, and regular field assessment of the plan given diverse scenarios, among others (Lindell, 2013). Authorities, planners, and evacuation managers are assisted by evacuation models in employing strategies for effective and efficient operations. Alternative evacuation scenarios that can be integrated in the models are expected to establish suitable evacuation policies that help facilitate communication and transfer of information (Lumbroso et al., 2008). Evacuation modeling usually done in a sequential manner follows the classic four-step travel demand modeling (Abdelgawad et al., 2010b), described in general below.

Figure 2.1. Evacuation Process Model

The first stage is evacuation demand generation which determines the number of evacuees and their departure time patterns. The second stage is the evacuation distribution of which the origin-destination is either assumed using the potential locations of shelters/evacuation centers or estimated from the destination choices of evacuees gathered from past evacuation events (Mesa-arango et al., 2013). In the third stage, mode split is primarily important in the withdrawal stage of evacuation process. It is one of the essential logistical considerations in evacuation planning and operation, because resources are often scarce that requires careful planning. The last stage of evacuation modeling is the traffic assignment where the evacuation route is taken into consideration. This also has implications on the network flow during evacuation.

Ref. code: 25605522300184DMX

In view of transportation planning, essential decisions based from the classic four-step travel demand model are evacuation decision, departure time choice, destination choice, mode choice and route choice. Evacuation decision is the choice of households at risk to either leave their place and move to a safe place or stay at home. Risk may be recognized based on behavior and past experiences (Scolobig, Marchi, and Borga, 2012). Murray-Tuite and Wolshon (2013) and M.B. Lim et al. (2013) reviewed broad range of behavioral factors that affect the likelihood of evacuation decision. Knowing how many people decide to evacuate (partially or fully) or stay is a fundamental consideration for shelter and resource allocation (Lindell, Kang, and Prater, 2011). This is very important in the overall evacuation planning and management in the sense that planners can estimate the evacuation travel demand, vehicle allocation and carry out traffic assignment in order to reach identified shelters at the shortest possible time (Pel, Hoogendoorn, and Bliemer, 2010; Hsu and Peeta, 2013). Some of the examples of evacuation decision studies were conducted by Hasan, Mesa-Arango, Ukkusuri, and Murray-Tuite (2012) for Hurricane Andrew, Ivan and Katrina and Ricchetti-Masterson and Horney (2013) for Hurricane Irene.

The departure timing of evacuees is taken before or during the onset of the disaster striking the vulnerable area. Lindell, Lu, and Prater (2005) investigated departure timing of evacuees for Hurricane Floyd based on the issuance of voluntary and mandatory evacuation. The study indicates that the evacuees opt to evacuate prior to the issuance of evacuation notices. M.B. Lim et al. (2015) also investigated departure timing of households in the case of an urban context in the Philippines.

Evacuation destination is an important aspect of evacuation planning (Lindell and Prater, 2007). Most reports classify destination types as emergency or public shelters, hotels and motels, and friends and family. Previous studies on evacuation destination are mostly in the context of hurricane in developed countries. These studies indicate that a large number of evacuees went to their friends and families; followed by hotels and motels; and lastly public shelters (Mileti, Sorensen, and O’Brien, 1992; Chen, 2005; Modali, 2005; Cuellar, Kubicek, Hengartner, and Hansson, 2009; Wu, Lindell, and Prater, 2012). Mesa-arango, Hasan, Ukkusuri, and Murray-Tuite (2013) added church, workplace and others together with the three categories mentioned above. 12

Ref. code: 25605522300184DMX

Understanding the mode of transportation used by households in evacuation has implications on the traffic flow in road networks. Earlier evacuation planning studies have considered cars as mode of evacuation (e.g. Cova and Johnson, 2003; Jha et al., 2004; Han and Yuan, 2005; Kwon and Pitt, 2005; Mitchell and Radwan, 2006; Murray-Tuite, 2007; Williams et al., 2007; Balakrishna et al., 2008; Chen and Xiao, 2008; Klunder et al., 2009; Noh et al., 2009; Huibregste et al., 2010; Wang et al., 2010; Pel et al., 2011). According to studies such as that of Lindell and Prater (2007) and Deka and Carnegie (2010a), households with vehicles are more likely to evacuate using their own vehicles. However, not all people at risk have their own vehicles (Murray-Tuite and Wolshon, 2013). Recognizing this, researchers have investigated the needs of the carless people and those with special needs such as people with disabilities and the elderly (e.g. Renne et al., 2008; Renne et al., 2009). Results led to evacuation planning and modeling efforts toward multimodal evacuation plans (Wolshon et al., 2005; Elmitiny et al., 2007; Liu et al., 2008; Abdelgawad et al., 2010a; Sayyady and Eksioglu, 2010; Naghawi and Wolshon, 2012; VanLandegen, 2010). Evacuation studies expanded to other modes of evacuation such as mass transit (e.g. Mastrogiannidou et al., 2009; Chan, 2010; Bish 2011; An et al., 2013), a mode that has the capacity to move a large number of people to safety. It is seen to be effective in developed countries where it is established as a common mode of transportation in urban areas (Shiwakoti et al., 2013). Despite advances in multimodal evacuation models, evacuation travel behavior is not well represented or realistically assumed in modeling (e.g. Thomas et al., 2010; Naghawi and Wolshon, 2012). Development of multimodal evacuation plan can then be explored in the context of disaster-prone areas in developing countries where comprehensive evacuation plans may be limited and resources such as vehicles are scarce; but, the adaptive capacity of people is high. Extensive researches on this area are mainly on hurricane evacuation. The studies consistently showed that majority of evacuees use private vehicles against other modes (Lindell and Prater, 2007; Siebeneck and Cova, 2008; Deka and Carnegie, 2010a; Lindell et al., 2011; Wu et al., 2012; Sadri, Ukkusuri, Murray-Tuite, and Gladwin (2014a). Deka and Carnegie (2010) in their study found that usually people take their private vehicle when evacuating. In a study, Sadri et al (2014) investigated how evacuees from Miami Beach, chose the mode of evacuation. Data 13

Ref. code: 25605522300184DMX

was gathered through a stated preference survey from hypothetical category four (major) hurricane scenarios.

Route choice is essential during the withdrawal stage of evacuation process. This includes the movement of people from the area at risk to the identified shelters. Wu et al. (2012) in reviewing past studies related to evacuees’ route choices mentioned that relatively few evacuees rely on received written materials from local officials or the news media whenever they decide to leave. Instead, some rely on maps in choosing their routes; but, more rely on personal familiarity on their evacuation routes and prior expectations about time, safety or convenience. Those who chose an evacuation route based on previous experience were less likely to rely on other sources of route information. Moreover, they found out that in choosing their evacuation routes, evacuees relied either on previous experience or on traffic conditions en route. This is a significant issue for mathematical evacuation models that assume evacuees choose their evacuation routes based exclusively on conditions en route such as in Hobeika and Kim (1998) and Sheffi, Mahmassani, and Powell (1981) because it suggests that evacuees will not distribute themselves optimally over the available routes. As a result, they have suggested that more research is needed to understand evacuees’ route choice behavior. Sadri et al (2014c) investigated how residents at Miami Beach that needs to evacuate over one of the six major bridges or causeways because of geographical locations of residential homes. The estimation findings suggest that the preference over a given bridge involves complex interaction of variables, such as, distance to reach evacuation destination, evacuation specific characteristics like evacuation day, time, mode and destination, and evacuee-specific characteristics such as gender, race, evacuation experience and living experience. Akbarzadeh and Wilmot (2015) in their recent study on time-dependent route choice included attributes of specific routes in analyzing the route choices of evacuees during hurricane.

The following sub-sections describe categories of determinants and details that are important in travel behavior, according to findings in past studies.

2.2.1 Socio-demographic information

Ref. code: 25605522300184DMX

Socio-demographic information of the decision-maker such as age, gender, income, education, number of household members, and vehicle ownership are the basic factors normally investigated with regards to travel behavior decision making in transportation research (Train, 2009). This information has been used for investigation in the area of evacuation travel behavior especially in the context of evacuation decision, departure timing and destination choice (M.B. Lim et al., 2013).

Researchers have investigated the effects of socio-demographic details of the decision maker and found that these are important to decisions made. For instance, Lindell, Lu, and Prater (2005) who studied factors on evacuation decision using correlation analysis for Hurricane Lili in 2002 found that younger, female and respondents with children in the home are more likely to evacuate. Further, a study by Whitehead et al. (2000) on decision to evacuate or not for households in North Carolina based on 1998 Hurricane Bonnie, showed that households living in mobile homes, are more likely to evacuate than others. Also, those with White race have higher probability of evacuating than other groups. However, households with pets were less likely to evacuate.

In another study on evacuation decision by Fu and Wilmot (2004), the type of housing where households are living in mobile homes were more likely to evacuate than others. A household also decides to evacuate if living in an area at risk of flooding. Moreover, Stopher et al. (2004) in the case of bush fires found significant factors are age and gender of decision maker, number of vehicles, presence of younger children, presence of old age adults, and length of stay in the residence. They also found that if the household have more owned vehicles, the lesser the likelihood of evacuating. Also, older decision makers and those with older members in the home are more likely not to evacuate, while female household decision makers are more likely to evacuate when compared to male decision makers. Households with younger children are more likely not to evacuate, while those that have lived longer in the current residence have lower probability of evacuating.

Hasan et al. (2011) in another study identified number of children, house ownership status, type of housing (mobile), income and level of education are determinants to 15

Ref. code: 25605522300184DMX

evacuation decision making. A household who lives in a mobile house are more likely to evacuate while households that own their house are less likely to evacuate. High- income households are more likely to evacuate. Similarly, the higher the level of education a respondent from a household has, the more likely that the household evacuates. In addition, Hasan, Mesa-arango, Ukkusuri, and Murray-Tuite (2012) in their next study on transferability of models in different hurricanes but in the same geographical area found that households with more members preferred not to evacuate, while the more the children, the more likelihood of evacuation. Consistent findings in the previous research by Hasan et al. (2011) was found with regards to the effect of households’ type of housing, evacuation notice, pet ownership, house ownership and work during evacuation.

Results in M.B. Lim et al. (2015) in the study of evacuation decision of households in the Philippines found out that older household heads tend to stay at home compared to younger ones, females are more likely to evacuate than males. Households with greater number of households prefer not to evacuate. Also, education and the type of work of the head of the household, household income, house ownership type (rented or owned), and the number of house floor levels also affect evacuation decision.

More information on findings in the literature specific to respective evacuation behavior decisions such as mode and route choice are presented in Sections 2.3 and 2.4, respectively.

2.2.2 Hazard information and evacuation warning

Hazards are categorized as natural and man-made. Each of the hazards behaves differently: each has its own unique characteristics and impacts. For example, severity and impact of flood hazard can be accounted based on its level or depth and the affected segment of the road network respectively (Hsu and Peeta, 2013). Hazard- related information such as severity, existing vulnerable conditions, and complexities on the nature are essential factors. For these reasons, previous study underscores that future evacuation models should be specific to disaster type (Murray-Tuite and

Ref. code: 25605522300184DMX

Wholshon, 2013) and flexible enough to be utilized for other disaster type (Galindo and Batta, 2013).

Lindell, Lu, and Prater (2005) who studied factors on evacuation decision using correlation analysis for Hurricane Lili in 2002 found that social and environmental cues were significant to evacuation decision. Fu and Wilmot (2004) who developed a model for hurricane evacuation decision, identified distance to storm at a given time, forward speed at a given time, presence of an evacuation order, as influential factors to evacuation decision. The distance of the storm to the household location indicates that the nearer the storm, the more likely a household would evacuate. For households living in low risk area, the probability of evacuation increases day by day as the hurricane approaches. Stopher et al. (2004) also in the case of bushfire found that significant factors are wind speed, wind direction, fire type influence decision making. When all other variables are held constant, the higher wind speeds more likely result in evacuation. Similarly, an unfavorable wind direction and a hot fire are more likely to result in household evacuation.

Issuance of warning is another aspect of evacuation. This is done after the authorities or communities assessed the hazard. Important considerations on issuance of warning include the source of information; the warning message that specifies the characteristics of the hazard, vulnerability, and risk; and the communication channels, receiver of the information and feedback (EMA, 2005; Dash and Gladwin, 2007). The warning message specifically includes preparation and provision of shelters, description of the evacuation areas, and sequential order in which areas are to be evacuated (if staged evacuation is considered).

The source of warning and type of evacuation advice has been specifically found out to influence evacuation decision making. Lindell et al (2005) found that warning source including peers and local authorities strongly affect evacuation decision. Whitehead et al. (2000) revealed factors that influenced decisions of coastal North Carolina households to evacuate or not for Hurricane Bonnie in 1998 and future evacuations include the type of evacuation advice (mandatory or voluntary). Individuals who received a voluntary evacuation order were more likely to evacuate 17

Ref. code: 25605522300184DMX

than those who did not receive an order. Also, those who received a mandatory evacuation order were almost eight times more likely to evacuate than others. Fu and Wilmot (2004) also found that the presence of an evacuation order (either voluntary or mandatory evacuation) increases the probabilities of evacuation for either household. However, households living in high risk area have smaller probability to evacuate and tend to delay evacuation until a later time when there is no evacuation order. They also have high probability to evacuate and tend to evacuate early when given an evacuation order. In addition, Hasan et al. (2011) have identified factors including type of evacuation notice received. Household that receives mandatory evacuation notice are more likely to evacuate and the response is uniform across households. A household who first hears about the evacuation notice from a friend or relative instead of any other source such as television, radio, Internet, has a higher probability to evacuate. Moreover, previous experience with a major hurricane resulted in a lower probability to evacuate. In their next study, Hasan, Mesa-arango, Ukkusuri, and Murray-Tuite (2012) studied transferability of models in different hurricanes including Ivan, Andrew and Katrina, in the same geographical region. Previous hurricane experience negatively affects evacuation for Ivan hurricane dataset, and evacuation notice.

2.3 Evacuation Mode Choice Behavior and Related Determinants

The mode of evacuation is influenced by a number of determinants such as characteristics of the hazards, distance from areas at risk to safety, location of the evacuees when evacuation order is given, the capacity and size of the vehicle, type of vehicle (personal or transit), and population group such as tourists or people conducting intermediate trips (Murray-Tuite and Wolshon, 2013). It is also important to consider other intermediate trips such as child pick up, as well as activities of different groups of people such as tourists, reentry to home after evacuation. Background traffic is also to be considered and the behavior of evacuees fleeing from their place. Deka and Carnegie (2010a) highlighted that the choice of other modes excluding private vehicles is determined by the familiarity of evacuees with transit options and not having personal vehicle. In another study on the success of using shuttles during normal days, it can be learned from Deka and Carnegie (2010b) that if 18

Ref. code: 25605522300184DMX

authorities have to impose the use of public vehicles when evacuating, income, availability and parking costs may affect evacuees’ decision making in choosing the mode to take.

Sadri et al. (2014) investigated how evacuees from Miami Beach chose their mode of evacuation. Data was gathered from residents using survey data from a hypothetical hurricane in category four. In order to explain the results of evacuation mode choice behavior, a nested logit model was developed. With options including special evacuation bus, taxi, and regular bus that had been aggregated from evacuees riding with persons from another household with other type of mode, the authors found that evacuees have higher probability of taking special evacuation buses. In addition, single evacuees are less likely to take taxi, while evacuees with annual household income over $80,000 have higher probability of taking the taxi. Moreover, when a household has adults more than 65 years old, the higher is the likelihood of riding with someone from another household. Also, evacuees going to hotels are more likely to use the regular bus service while those going to public shelters are less probable to ride with someone else from another household. Also, evacuees going to their friends or relatives’ home are less likely to use special evacuation bus. Sadri et al. (2014) however, emphasized the small sample size used in the study. Hence, future studies were suggested to generate more data for the same kind of analysis. The authors also highlighted the need for more research effort to reveal and introduce new determinants which affect the mode choice behavior. However, findings are believed to be helpful for emergency planners and policy-makers to develop better evacuation plans and strategies for evacuees depending on others for their evacuation transportation.

Wu et al. (2012) in their study measured evacuation transportation through the number of registered vehicles, number of vehicles that evacuees took, number of trailers taken; and the mode evacuees took in case they did not have their own vehicle such rode with someone else, public transit, or other. The authors further exemplified that since older evacuees are less likely to have a registered vehicle, then they rely on carpooling. Carpooling carless evacuees also leave earlier than those with private

Ref. code: 25605522300184DMX

vehicles, travelled lesser distances and had higher likelihood of staying with their friends or relatives rather than in public shelters.

It was also found out that evacuating households take more than one vehicle. Yin et al. (2014) used data in their study collected from 3,200 households in Florida, Alabama, Mississippi, and Louisiana after Hurricane Ivan landfall in September 2004. Data from 1,443 households that evacuated were used for analysis where regular Poisson regression model was estimated to understand the effects of variables to the number of vehicles taken when evacuating. Yin et al. (2014) found out that there is a higher likelihood of taking fewer vehicles when households travel long distances and during late evacuation. Additionally, households that experienced hurricane before the event and have pets have higher likelihood of using more vehicles. Distance from the coast also indicated some level of significance to the number of vehicles taken.

2.4 Evacuation Route Choice and Related Determinants

In the past, studies in evacuation route choice were mostly considered under evacuation traffic simulation modeling, while very few are focused on modeling travel behavior. Evacuees in most studies consider nearest/shortest or familiar route during evacuation (Murray-Tuite et al., 2012). The former choice is particularly evident in most simulation modeling studies such as the works done by Huibregtse et al. (2010) and Pel et al. (2010). With the latter, Wu et al. (2012) in their review mentioned that evacuees rely on personal familiarity, time, safety or convenience. These similar conditions are assumed on en route scenario (Hobeika and Kim, 1998; Sheffi et al., 1981) as evacuees will not distribute themselves optimally over the available routes. In most cases, behavior and traffic simulation are modeled separately. Nevertheless, travel behavior such as evacuation route choice can give an added value to recent practices in traffic simulation studies by investigating the preferences of the evacuees in disaster events.

Explanatory variables to route choice behavior were investigated in the following studies. Three routing strategies was used for analysis in Sadri et al. (2014b)’s work in understanding route preferences of hurricane evacuees. These are usual/familiar route, 20

Ref. code: 25605522300184DMX

the route recommended by officials and updated route. Routes taken by interviewees in the Akbarzadeh and Wilmot (2015)’s hurricane study were categorized according to route attributes in terms of familiarity, accessibility, availability of services and road type. A common approach used in these studies in identifying choice sets for route choice analysis is that when interviewees have indicated that they would evacuate, they were asked to state the route they choose to reach their destination. The routes listed are then categorized according to similar attributes with reference to similar origins and destinations. The summary of the routing strategies used for analysis in evacuation route choice studies are presented in Table 2.1.

Table 2.1. Route categories used in evacuation route choice literature Authors Routes/routing Characteristics/description strategies Akbarzadeh Identified interstate Route-specific characteristics that were and Wilmot highways grouped investigated include familiarity with (2015) according to origin and route, availability of fuel and shelter, destination road type and accessibility Sadri et al. Usual/familiar route Usual route/familiar/shortest route (2014b) combined in one due to possible overlap. Recommended route Route recommended by the government officials Updated route For evacuees that updated route because of traffic congestion once they try initial route Sadri et al. 6 Bridge/Causeways McArthur Causeway, Venetian (2014c) Causeway, Julia Tuttle Causeway, John F. Kennedy Causeway, Broad Causeway and Haulover Bridge

Explanatory variables that determine route choice behavior of evacuees are related to transportation infrastructure and the available services that are located along given

Ref. code: 25605522300184DMX

routes. Akbarzadeh and Wilmot (2015) investigated the effect of travel time, familiarity with the route, availability of fuel and shelter, road type, and accessibility of the route on evacuees’ route choice. They also tested whether the importance given by evacuees assign to the variables determining route choice varies with time. By using multinomial logit model, findings show that the longer a route is, the less likely it will be selected. Interstate highways are expected to be more attractive for evacuation than other highways. Perceived service, which is the product of familiarity and service, will attract evacuees to that route. Accessibility, which shows the distance between a route and the residence suggest that evacuees prefer routes that are more accessible to them. Moreover, statistical comparison of parameters for risk- averse (evacuees during the first half) and risk-tolerant (evacuees during the second half) show that as the storm gets closer, people are more likely to take freeways rather than arterial routes. Also, as the storm gets closer, distance becomes less important, while perceived service becomes more important. This may indicate that as the storm gets closer, people become less discriminating about less significant issues and just want to get out of the area. On the other hand, because congestion is likely as the storm gets closer and being stranded on the road without services is highly undesirable, availability of service facilities becomes more important. In addition to these findings, the authors noted that the variables included in their model are not necessarily the only variables affecting route-choice behavior of evacuees. Moreover, Sadri et al (2014c) found that preference over a given bridge also involves the distance to reach evacuation destination, and evacuation specific characteristics like evacuation day, time to route decision making.

As mentioned earlier in section 2.2.1, the inclusion of socio-demographic information in transportation planning has become necessary in discrete choice modeling (Train, 2009). Researchers that studied the relationship of information to evacuation travel behavior includes Fu and Wilmot (2004), Stopher et al. (2004), Hasan et al. (2011), and Mesa-arango et al. (2013) to name a few as also mentioned in the earlier section. However, understanding how socio-demographic characteristics of evacuees affect their route choice decision is still understudied. Among the recent studies in line with this was conducted by Sadri et al. (2014b). In the study, the authors considered age, income and number of children in the family as variables that influence the route 22

Ref. code: 25605522300184DMX

decision from alternatives of familiar route and route recommended by authorities. Their findings showed older evacuees are likely to detour or update their routes. Evacuees having a high annual household income of USD 40,000 or more are less likely to prefer the recommended routes. High income households may prefer to stick with their plan instead of following recommendations because they have better access to in-vehicle and other real-time travel information. Evacuees having children less than 18 years old are more likely to follow the routes recommended by the emergency officials rather than any of the other two route options. Sadri et al (2014c) in their study on choices of households of which Bridge to take when evacuating, found that evacuee-specific characteristics such as gender, race, evacuation experience and living experience are important considerations to decision making.

Flood disaster warning plays an important role in the evacuation compliance. Studies show that warning in general may include the source, message and the characteristics and possible impact of the hazard (EMA, 2005; Dash and Gladwin, 2007). It was found out in Sadri et al. (2014b) that when a household receives an evacuation notice early enough, they could find more information about the traffic conditions and learn the evacuation routes specified for that particular area. Because of the predetermined routing strategy, evacuees are more likely to follow the recommended route or switch to the routes based on prevailing traffic conditions. However, the variable indicating the households receiving evacuation notice from radio or television instead of any other source (friend, relative, or newspaper) indicate that evacuees are more likely to take the route familiar to them rather than taking recommended evacuation routes.

Traffic simulation modeling studies for flood disasters used degeneration of road network over time in the event of flood hazards (Huibregtse et al., 2010). This has been applied over a number of simulation modeling studies such as Pel et al. (2011) and Lammel et al. (2010). Knowing that in reality this might happen regardless of the limitation of models, variables related to the hazard such as flood level, distance from source of flood, and distance to evacuation destination should also be considered in the analysis of evacuation route choice through discrete choice modeling technique. Such characteristics are attributed as environmental cues in the study conducted by Siebeneck and Cova (2013). The more distance the evacuees need to travel during 23

Ref. code: 25605522300184DMX

evacuation, the more likely that the evacuee will update or switch their route (Sadri et al., 2014b).

The effect of variables related to evacuation such as evacuation decision, destination choice, departure time choice, and mode choice have also been investigated in studies on hurricane and flood (e.g. Sadri et al., 2014a). Evacuating households may decide to evacuate some members of the household (partial) or all members (full) (e.g. Stopher et al., 2004; Hasan et al., 2011). They can also choose to evacuate prior to or when the hazard strikes their vicinity (Lindell et al., 2005; Charnkol and Tanaboriboon, 2006; Akbarzadeh and Wilmot, 2015). The destination of the households may include public shelter, friends/relatives, hotels/motels, church, workplace and others (Mileti et al., 1992; Chen, 2005; Modali, 2005; Lindell and Prater, 2007; Cuellar et al., 2009; Wu et al., 2012; Mesa-arango et al., 2013). And in order to reach their destinations, evacuees could walk, cycle, or use their own car or vehicle provided by authorities (Lindell et al., 2007; Siebeneck and Cova, 2012; Deka and Carnegie, 2010a; Lindell et al., 2011; Wu et al., 2012). Sadri et al. (2014b) found that if an evacuee wants to evacuate to a friend or relatives’ house, majority of the evacuees, when they evacuate to a familiar destination, are likely to select their familiar routes from their previous visits to those destinations. In addition, when evacuees are departing well ahead of time, they do not necessarily follow recommended evacuation routes or switch routes; rather, they would prefer to drive through the routes which they are familiar with. Sadri et al (2014c) also in their study results indicate that route decision making is affected by mode and destination considerations of evacuees.

2.5 Summary and Chapter Conclusions

Evacuation is employed as an effective measure to minimize damages and losses in any disaster event. Evacuation planning through modeling helps authorities, planners, and evacuation managers to apply effective evacuation operations, as evacuation scenarios that can be integrated in the models are expected to establish suitable evacuation policies that help facilitate communication and transfer of information (Lumbroso et al., 2008). Evacuation modeling usually done in a sequential manner follows the classic four-step travel demand modeling. These include evacuation 24

Ref. code: 25605522300184DMX

decision, departure time choice, destination choice, mode choice and route choice. Evacuation decision is the choice of households at risk to either leave their place and move to a safe place or stay at home. Knowing how many people decide to evacuate (partially or fully) or stay is a fundamental consideration for shelter and resource allocation (Lindell, Kang, and Prater, 2011). This is very important in the overall evacuation planning and management in the sense that planners can estimate the evacuation travel demand, vehicle allocation and carry out traffic assignment in order to reach identified shelters at the shortest possible time (Pel, Hoogendoorn, and Bliemer, 2010; Hsu and Peeta, 2013). Understanding evacuation departure timing also is important in planning for demand at various time scales.

Moreover, understanding the mode of transportation used by households in evacuation has implications on the traffic flow in road networks. Most of early studies in this area and evacuation planning have considered cars as mode of evacuation. However, due to recognition that not all people at risk have their own vehicles, researchers have investigated the needs of the carless people and those with special needs, in addition to use of mass transit, walking and multimodal evacuations. Despite advances in multimodal evacuation models, evacuation travel behavior is not well represented or realistically assumed in modeling, especially in the context of developing countries. Evolution of multimodal evacuation plan can then be explored in the context of disaster-prone areas in developing countries. Comprehensive evacuation plans in developing countries may be limited and resources such as vehicles are scarce; but, the adaptive capacity of people is high. Extensive earlier researches on evacuation in developed countries are mainly on hurricane evacuation. Flooding is more prevalent in developing countries, hence can be explored.”

Route choice includes the movement of people from the area at risk to the identified shelters. It is recognized in research that relatively few evacuees rely on received written materials from local officials or the news media whenever they decide to leave, while some rely on maps in choosing their routes. More rely on personal familiarity on their evacuation routes and prior expectations about time, safety or convenience. Some evacuees also relied either on previous experience or on traffic conditions en route in choosing the route to take during evacuation. This is a 25

Ref. code: 25605522300184DMX

significant issue for mathematical evacuation models that assume evacuees choose their evacuation routes based exclusively on conditions en route. Hence, it was suggested that more research is needed to understand evacuees’ route choice behavior. Sadri et al (2014b) found that evacuees age, income, and number of children in the family are important to decision making in addition to the distance they need to travel and evacuation warning received. While in another study of investigating choice from 6 bridges to take when evacuating, Sadri et al. (2014c) found several determinants to the choice such as distance to reach destination, evacuation day, time, evacuation mode and destination, gender, race, evacuation experience and living experience. Akbarzadeh and Wilmot (2015) in their recent study on time-dependent route choice included attributes of specific routes in analyzing the route choices of evacuees during hurricane.

Many determinants that include the evacuee characteristics, ones that relate to the hazard itself, some other evacuation related decisions and in addition to evacuation, mode-related variables as well as route specific variables, are variables that are important to mode and route decision making as found in earlier studies. These are then the starting point for further investigation that is done in this current study. This is according to the need of more studies needed to further understand behavior of evacuees in terms of mode and route decision making. Specifically, behavior in this context is analyzed in the context of flooding. Whether variables that determine mode and route choice behavior are consistent with earlier findings is also taken into account.

Ref. code: 25605522300184DMX

Chapter 3 Methodology

This chapter presents the methodology employed in conducting this research. Specifically, the study area, methods used for data collection, model estimation and model validation are presented in this chapter.

3.1 Research Framework

Travel-related decisions

Evacuation decision

Departure time Household choice socio- Hazard-related demographic Destination information information choice

Mode choice

Route choice

Figure 3.1. Research Framework

In order to obtain the objectives of this study, the framework presented in Figure 3.1 is employed. In this study, the determinants that affect mode and route choice

Ref. code: 25605522300184DMX

behavior are investigated. The exogenous variables are categorized into three broad groups which are households’ socio-demographic information, hazard-related information and travel-related information. Socio-demographic information of household heads’ such as age, gender, educational attainment and type of work, household income, number of family members, number of years living in the residence, presence of small children that are 10 years old or more, presence of senior citizen in the home, type of house ownership, presence of vehicle, among others are considered as exogenous variables. In addition, information related to the flood hazard such as the distance from the home of the household, level of flood, among others, are also considered as exogenous variables in analyzing the mode and route choices of households. It can also be seen from the framework that in the analysis of mode and route choices, travel-related information such as the evacuation decision (full, partial or stay), departure timing (before or during the flood), and the type of destination the households choose to go (public shelters, church/seminaries, friends’/relatives’ home), are treated as exogenous variables. These are based from the fact that these travel-related decisions are sequentially modeled in the evacuation process, as discussed and presented in section 2.2.

While mode and route choice are considered travel-related information, the framework of research treats mode and route choice as the endogenous variables. Type of mode used by households in evacuating is considered as exogenous variable of route choice in the analysis of determinants that explains it, detailed in Chapter 5.

3.2 Study Area

Data used for analysis in this study was gathered from households in Quezon City, Philippines, the largest city among 16 in Metro Manila with a land area of about 16,112.58 hectares. The City is Metro Manila’s point of strategic convergence for various transportation infrastructures. As one of the key cities in the country, Quezon City can be considered as a melting pot of different cultures coming from other parts of the country. The City is located at the north-eastern portion of Metro Manila, 28

Ref. code: 25605522300184DMX

bordering Manila City to the southwest, Caloocan City and Valenzuela City to the west and northwest, San Juan and Mandaluyong City towards the south, and Marikina and Pasig City to the southeast (Figure 3.2). In 2012, the City’s population is approximately 3.18 million (706,564 households), which grows at the rate of 2.92% per annum (Quezon City Government Planning and Development Office, QCGPDO 2013). This population is about 25% of Metro Manila’s population making it one of the most populated cities in the Philippines. It is also considered vulnerable to different types of hazards, one of which is typhoon-induced flooding that occurs several times in a year affecting 700,000 people (Quezon City Government and Earthquake and Megacities Initiative, QCG and EMI 2013). A factor that contributes greatly to the flood risk is the fact that cities of Metro Manila are situated on a wide flood plain, the main river which is the Pasig river, filled with tributaries (the Marikina River and San Juan River as major tributaries), and canals that branch out in various cities and towns. One of the worst recorded massive flooding from excessive rains in Metro Manila happened in September 2009, affecting millions of people in the Philippines and resulted to hundred deaths in Quezon City alone. During this flood event, recorded cost of damage is approximately USD 275 million including infrastructures and the agriculture sector (QCG and EMI 2013). Another flood event happened in 2012 when about a thousand millimeters of rain fell in some parts of Metro Manila. Hundreds of residents were evacuated as water in affected villages reached up to the roofs of houses in many areas. This event was replicated in Mid- August 2013 brought about by the heavy monsoon rains and tropical cyclone Trami that hit the Philippines. The flood resulted to more than USD 14 Million worth of damages, and also, prompted the evacuation of less than nine thousand households in Quezon City to assigned public shelters (Social Services Development Department, SSDD 2013). This flood event was the basis of the information collected from households used for analysis in this study.

A number of sub-districts were affected by the 2013 flood event. Among the five sub- districts suitable for the study as recommended by the government are Bagong Silangan, Bahay Toro and Sto. Domingo sub-districts. The criteria for selection were 29

Ref. code: 25605522300184DMX

the history of evacuation and the number of flood vulnerable households in the sub- districts. Bagong Silangan has a population of 22,000 households, Bahay Toro has 74,987 households, and Sto. Domingo has around 15,560 households (QCGPDO 2013). During the flood, there were about 2864, 500, and 533 households that evacuated to public shelters from Bagong Silangan, Bahay Toro and Sto. Domingo sub-districts, respectively (SSDD 2013).

Figure 3.2. Flood map of Quezon City with number of households interviewed Source: Department of Science and Technology (2015)

3.3 Data Collection

Face to face interviews were conducted with randomly selected households in sub- districts. Random selection of households was done using the cluster sampling technique. Here, villages in high flood risk areas were selected. Within the villages, simple random selection of interview households was done.

Ref. code: 25605522300184DMX

Prior to the commencement of the data collection, the interviewers were trained by the research team to make sure that quality interviews are made. The interviews were conducted from October 2013 to May 2014. Households in villages located at flood prone areas in Bagong Silangan, Bahay Toro, and Sto. Domingo were randomly approached for interviews. During the interview process, the respondents were given a brief introduction on the study being conducted. This was to ensure they understand the context which is the basis of their answers to the questions. The interviewers also made sure that the household experienced flood during the Mid-August 2013 before proceeding with the interviews. To do this, interviewees were first inquired if they experienced flood during the 2013 event. The flood details including the level of flood in their house, the number of days they were flooded, the level of damage the flood caused their house, and whether they received evacuation warning and its source. Information on the distance of house location from the flood hazard was also solicited. Then, information on evacuation-related decisions was also elicited including the type of evacuation decision, the timing of evacuation, their evacuation destination, the mode they used and the route they took when evacuating. The type of evacuation was indicated as full when the whole members of the household moved to safety, and partial when most of the members moved to safety while leaving some others behind to look after the household’s belongings. For departure timing, households that are risk-averse or risk-tolerant were indicated as those that evacuated before or during the flood. The destination type choice included those that went to the public evacuation centers, church/seminary or to friends/relatives’ homes. The households’ modes of evacuation ranged from vehicles provided by the government, rented or personal vehicle, and walking. The distance traveled in evacuating, and the costs incurred during evacuation were also inquired.

Routing strategies available to households were identified and included in the survey instrument. The route recommended by the government, usually available in other studies was not made available to households in this current study during issuance of evacuation warning. During the interviews, a map of possible routes to identified evacuation centers were shown from which interviewees were asked which route they 31

Ref. code: 25605522300184DMX

took during evacuation to their specific destinations. They were then asked of the reason of taking such route rather than taking other available ones. From these, three routing options were identified including the route that they usually take during normal days (most familiar route to them that was also confirmed as not the nearest to their destination), the nearest route or where they thought they can take to evacuate faster, and the route without flood waters. These were the list of options made available to households during the rest of interviews in addition to “others” in case of the existence of route option not identified during the pilot survey. After data collection, however, the third option was removed from analysis as it was found that such route was not available to them that time or it is the only route available. In addition, there were no “other” route options made available in the data collected. The 2 remaining routes are then used for analysis in this study. These route options are defined as nearest and familiar (the main road most familiar that they usually take during normal days but not the nearest).

The second part of the interview solicited suggestions and comments for better situations in future evacuation from floods. The third part of the interview aided solicitation of socio-demographic information of the head of the household and other household information that includes age, gender, marital status, educational attainment and type of work of the head of the household, presence of health problem and insurance, household monthly income, number of household members, age of members, the presence of small children and senior citizens, number of years the household has been living in that residence, type of house ownership and materials, the number of house floor levels, vehicle ownership, and pet ownership. The questionnaire was prepared in English language and translated into Filipino. It was pilot tested prior to the full survey in order to make necessary adjustments according to the context of the study area.

Six hundred thirty two interviews were completed out of 640 total number of households approached. This shows a response rate of nearly 99%, equal to total interviews completed divided by the total number of approached households. The 32

Ref. code: 25605522300184DMX

high response rate shows the eagerness of the households in sharing their flood experiences with the hope that results of the endeavor can assist in better evacuation situations in future flood events. In addition, the target respondents are located in high flood risk area, where people are vulnerable to flood. Hence, they are already aware of the risks they face, understand the need for preparation and support. Details on the study area was discussed earlier in section 3.2. Above all, coordination with relevant officials from the city level down to the village level contributed to the high interview response rate. Out of the 632 interviews, 340, 150, and 142 were from Bagong Silangan, Bahay Toro, and Sto. Domingo, respectively. The locations of these sub- districts in Quezon City are indicated in Figure 3.2. All the data collected was tabulated in excel sheet. The data was verified and cross checked by ensuring that information provided by the households were according to the questions asked. Data obtained with a lot of missing information and inconsistency based from the needed information was excluded from the analysis. Also, data from households that decided not to evacuate was not included in the analysis.

3.4 Method Used for Variable Selection in both Mode and Route Choice Analysis

Determinants of respective choices included in the models were selected through stepwise backward elimination method. The stepwise selection method is an effective way to find the best subset of variables that predict an outcome (Steyerberg et al., 2004). Variables including evacuation-related decisions, socio-demographic characteristics of the head of the household, household characteristics, information related to the hazard and the mode, as mentioned earlier in this section, were included in the analysis. Data for each sub-district as well as pooled ones were subjected to stepwise selection of determinants.

3.5 Modeling Framework for Mode and Route Choice Analysis and Model Estimation

Ref. code: 25605522300184DMX

Mode choice models estimate how many people will use which available mode of transport. Similarly, route choice models are developed to predict the route that evacuees choose. As the considerations of households in choosing the mode and route they took in evacuating were investigated in this study, the logit model under the maximizing utility framework was used. Logit has been widely used in modeling mode choice during normal traffic conditions such as in studies conducted by Cherry and Cervero (2007), Ashalata et al. (2013), Miskeen et al. (2013), Danaf et al. (2014), and Mitra and Buliung (2015). Its usage is also becoming more popular in evacuation modeling due to its simplicity, closed form estimation, and its capability of capturing behavioral context of decision-making (e.g. Charnkol and Tanaboriboon, 2006; Hasan et al., 2011; Mesa-arango et al., 2013; M.B. Lim et al., 2015b).

The mode choice model postulates that the probability of choosing a specific mode for a journey is based on relative values of determinants (Ortuzar and Willumsen, 2011). The utility function in analyzing the mode choice behavior of households is shown in Equations 1 and 2, respectively for any household, h, choosing to walk, w, or by other modes of transport (owned/rented vehicle or vehicle provided by the government), v. The utility function is composed of observable determinants, Xwh and

Xvh, for mode outcomes of walking and other modes, respectively; the corresponding ′ ′ vector of parameters to be estimated, 휷푤 and 휷푣, for walking and any other modes, respectively; and random terms εw and εv, which represent unobserved determinants of decision making. ′ 푈푤ℎ = 휷푤푿푤ℎ + 휀푤ℎ (1) ′ 푈푣ℎ = 휷푣푿푣ℎ + 휀푣ℎ (2)

The probability that a particular mode is chosen is given in Equations 3 and 4. Pwh and

Pvh are the probabilities that a mode alternative, walking, w, or other modes, v, is chosen.

Ref. code: 25605522300184DMX

′ 푒훽푤 푋푤ℎ 푃푤ℎ = ′ ′ (3) 푒훽푤 푋푤ℎ+ 푒훽푣 푋푣ℎ

′ 푒훽푣 푋푣ℎ 푃푣ℎ = ′ ′ (4) 푒훽푤 푋푤ℎ+ 푒훽푣 푋푣ℎ

The log likelihood (LL) function is presented in Equation 5, where H is the number of households, and M is the type mode chosen by household, h. To fit the logit model, the maximum likelihood estimation was used with the aid of IBM SPSS Statistics.

푀 퐻

퐿퐿 = ∑ ∑ log ( 푃푚ℎ) (5) 푚=1 ℎ=1

Similarly, the modeling framework for the evacuation route decision making is developed using the binary logit model. The question of whether to evacuate through a familiar route (but not the nearest) and the nearest route involves a decision making between these two possible choices. The binary logit model for any household, h, choosing the nearest route, n or familiar route, f, is represented by the utility function as shown in equations 6 and 7, respectively.

′ 푈푛ℎ = 휷푛푿푛ℎ + 휀푛ℎ (6) ′ 푈푓ℎ = 휷푓푿푓ℎ + 휀푓ℎ (7)

′ ′ Where 휷푛 and 휷푓 are vectors of parameters to be estimated; Xnh and Xfh are vectors of the factors that determine the route choice of nearest, n, or familiar, f, of household, h, respectively; and εnh and εnh, accounts for the impact of unobserved attributes and unobserved taste variations on observed route choice n and f, respectively. The probability that nearest or familiar route is chosen, denoted by Pnh and Pfh respectively are presented in equations 8 and 9.

Ref. code: 25605522300184DMX

′ 푒훽푛 푋푛ℎ 푃 = (8) 푛ℎ ′ 훽′ 푋 푒훽푛 푋푛ℎ+ 푒 푓 푓ℎ

훽′ 푋 푒 푓 푓ℎ 푃 = (9) 푓ℎ ′ 훽′ 푋 푒훽푛 푋푛ℎ+ 푒 푓 푓ℎ

The LL function is presented in Equation 10, where H is the number of households, and M is the type mode chosen by household, h. To fit the logit model, the maximum likelihood estimation was used with the aid of IBM SPSS Statistics.

푀 퐻

퐿퐿 = ∑ ∑ log ( 푃푚ℎ) (10) 푚=1 ℎ=1

3.6 Validation of Models

3.6.1 Receiver operating characteristics, area under curve and correct classification rate

The ability of the model to distinguish correctly the different outcomes based on a specified cut-off point (discrimination) is evaluated using the area under the Receiver Operating Characteristics (ROC) curve (AUC). ROC analysis is being employed in many areas for as a common method for graphical analysis of classification models. Details about ROC have been documented in many literature and readers can refer to these for further details (e.g. Fawcett, 2006). The use of AUC has become an important method of assessing and building classification models. It indicates the probability of a model to indicate a randomly chosen positive case (sensitivity) higher than a randomly chosen negative case (specificity). AUC ranging from 0 to 1, a general rule by Hosmer and Lemeshow (2000) is outlined as follows: • 0.9≤AUC≤ 1 = outstanding discrimination • 0.8≤AUC<0.9 = excellent discrimination; and • 0.7≤AUC<0.8 = acceptable discrimination. 36

Ref. code: 25605522300184DMX

The base classification rate indicates the proportion of correct classification that occurs by chance alone is also presented in this study. It is determined by summing up the squares of the percentage of outcomes in the data. The correct classification rate (CCR) is compared to the base rate to evaluate the predictive performance of the models. The increase in the CCR compared to the base rate indicates how much the model can improve predictive accuracy with the addition of significant variables in the model.

3.6.2 Likelihood ratio-based validation test

To further statistically investigate the validity of the models’ specifications, a likelihood ratio (LR)-based test was employed. The test was conducted to check the significant difference between the parameters of the models estimated using pooled data and the models estimated using parts of the whole data (that is, using the sub- district data). The null hypothesis is that the pooled models are the same between sub- districts. The procedure as detailed in Sadri et al. (2014b) is adopted in this study.

Equation 6 shows the formula for calculating the LR, where LL (βpool), LL (βsub1) and

LL (βsub2) are the log-likelihood at convergence of the model estimated using the pooled data, and the separate sub-district data (sub1 and sub2), respectively. The LR value which is χ2-distributed with degrees of freedom equal to the number of parameters estimated (not including the model constant), is compared to the critical value at 5% significance level. If the resulting LR is less than the critical value, the null hypothesis is not rejected. Hence, the validity of the pooled model specification is established. Otherwise, the results indicate that sub-district models may be more useful than the pooled model.

LR= -2[LL(βpool )-[LL(βsub1)+LL(βsub2)] (6)

Ref. code: 25605522300184DMX

Chapter 4 Factors Determining the Evacuation Mode Choice of Residents in High Flood Risk Areas

4.1 Summary of Data and Descriptive Statistics of Variables for Mode Choice

The total number of observations used for analysis of mode taken during evacuation is 427 of which 212, 132, and 83 are from Bagong Silangan, Bahay Toro and Sto. Domingo, respectively. From the data gathered, households’ modes of evacuation include vehicles provided by the government, rented or personal vehicle, and walking. Vehicles provided by the government, rented, or personal vehicle were merged as one category, “others”, due to their small percentage of data. These two categories (walking and others) were included in the analysis of mode choice behavior. Mode indicated as “others” served as the reference category for parameter estimation.

The percentage of those who walked when evacuating and those who evacuated by other modes of transport in Bagong Silangan are 82.1% and 17.9%, respectively. Bahay Toro had about 65.9% households that walked, while 34.1% reported using other modes of transportation. In Sto. Domingo, 48.2% reported evacuating by walking and the other 51.8% took other available modes of transport. Due to the small number of cases for Bahay Toro and Sto. Domingo, these were combined and was used to estimate and validate a pooled model (Model 4). Grouping these data may yield to reliable results, particularly if the sub-districts have similar determinants with the same effects to the outcomes. A model is also estimated using the combined cases for the three sub-districts (Model 5).

For each mode choice model, the backward elimination stepwise selection of variables was performed to identify appropriate explanatory variables to be included in the model. Sixteen variables were tested for significance and inclusion in the models. These are age, gender, type of work, and marital status of the household head, number of household members, presence of small children aged less than or equal to 10 years old, and presence of elderly aged greater or equal to 60, presence of pet and

Ref. code: 25605522300184DMX

vehicle ownership, flood level, distance of household from the source of flood, and the evacuation household decisions variables (evacuation decision, departure time, destination, and mode choice). The variables were assessed for inclusion in the model using statistical test with resulting p-values≤0.05. Insignificant variables were removed one at a time. After a variable is removed, the variables left were subjected to statistical test. The process is repeated until the desired combination of variables that gave a significant model is met. Resulting variables included in the models are presented in Table 4.1. Information on the percentage under specified categories is also provided for each determinant with the corresponding models.

Determinants included in Bagong Silangan model (Model 1) are evacuation departure timing, the level of education of the head of the household, presence of small children aged less than or equal to 10 years old, presence of personal vehicle and evacuation distance traveled. Bahay Toro model (Model 2) includes evacuation departure timing, the age of the head of the household, the level of education of the head of the household, the number of years living in the residence, presence of personal vehicle, and the source of warning. Determinants for Sto. Domingo model (Model 3) include, age and gender of the household head, presence of small child, presence of health problem for the head of the household and the cost of evacuation. The determinants for the pooled model for Bahay Toro and Sto. Domingo data (Model 4) are evacuation departure timing, their destination, age, and level of education of the head of the household, house ownership type, presence of vehicle and the source of warning. Finally, for the model using pooled data from the three sub-districts (Model 5), determinants include presence of vehicle, source of evacuation warning and evacuation distance traveled.

Table 4.1. Variables included in each model, categories, and corresponding percentage in the data Percentage in the data (%)

Variables Categories Model Model Model Model Model 1 2 3 4 5

Dependent variable

Mode choice Evacuated by other modes 17.9 34.1 51.8 40.9 29.5

Ref. code: 25605522300184DMX

(MDEC) Evacuated by walking 82.1 65.9 48.2 59.1 70.5

Evacuation-related decisions

Departure timing Evacuated during the flood 34.4 84.1 89.2 86.1 60.4 (TDEC) Evacuated before the flood 65.6 15.9 10.8 14.0 39.6

Destination choice Evacuation Center 61.3 50.8 8.4 34.4 47.8 (DDEC) Church/seminary 14.6 9.9 72.3 34.0 24.4

Friends/relatives 24.1 39.4 19.3 31.6 27.9

Characteristics of the head of household and other household information

Age of the head of 20-30 years old 16 20.4 16.9 19.1 17.6 the family (AGE) 31-40 years old 36.8 28.0 20.5 25.1 30.9

41-50 years old 22.6 20.5 22.9 21.4 22.0

>50 years old 24.5 31.1 39.8 34.4 29.5

Gender of the head of Female 12.7 22.0 22.9 22.3 17.6 household (GEN) Male 87.3 78.0 77.1 77.7 82.4

Educational Elementary 28.8 20.5 18.1 19.5 24.1 attainment of the head of the household High School 49.5 54.6 56.6 55.4 52.5 (EDUC) Diploma/Undergraduate 14.2 13.6 20.5 16.3 15.2

Graduate 7.6 11.4 4.8 8.8 8.2

Presence of small No small child 37.3 38.6 16.9 30.2 33.7 children (CHILD) Small child is present 62.7 61.4 83.1 69.8 66.3

Health problem No health problem 74.5 86.4 73.5 81.4 78.0 (HPROB) With health problem 25.5 13.6 26.5 18.6 22.0

House ownership Rented 14.1 25.0 50.6 34.9 24.6 (HOWN) Owned 85.9 75.0 49.4 65.1 75.4

Years of residence <10 years 30.2 26.5 33.7 29.3 29.7 (YLIVE) 10-20 years 58.5 21.2 26.5 23.3 40.8

>20 years 11.3 52.3 39.8 47.4 29.5

Vehicle ownership No personal vehicle 85.8 89.4 77.1 84.7 85.3 (VEH) With personal vehicle 14.2 10.6 22.9 15.4 14.8

Hazard-related variables

Source of warning Other sources 30.2 47.7 69.9 56.3 43.3 (SWARN) (tv/radio/friends/relatives)

Ref. code: 25605522300184DMX

From Sub-district official 69.8 52.3 30.1 43.7 56.7

Mode-related variables

Evacuation distance <200 meters 33.0 0.8 6.0 2.8 17.8 (EDIST) 200-400 meters 16.0 1.5 9.6 4.7 10.3

>400 meters 50.9 97.7 84.3 92.6 71.9

Cost of evacuation No cost spent with mode 85.8 40.2 83.1 56.7 71.2 (ECOST) Spent cost in evacuating 14.2 59.9 16.9 43.3 28.8 through mode

Note: Model 1 = Model for Bagong Silangan Model 4= Model for combined Bahay Toro and Sto. Domingo Model 2 = Model for Bahay Toro Model 5= Model for combined Bagong Silangan, Bahay Toro and Sto Domingo Model 3 = Model for Sto. Domingo

4.2 Resulting Inter-correlations of Mode Choice Model Variables

The inter-correlations of the variables included in the logit models are first presented in Table 4.2, for two main reasons as outlined by Huang et al. (in press). First is to determine possibilities of better model fit of a model to quite different estimated models. Gordon (1968) noted that the best model estimated on a sample could fit significantly less well in the population and thus, in other samples. Second, the inter- correlation among all variables included in the logit is key information for conducting the quantitative meta-analyses, where the relative strength of effects among variables such as demographic and other behavioral ones can be determined. As presented in Table 4.2, the mode of evacuation is significantly correlated with departure timing (r=0.104), vehicle ownership (r=-0.180), source of warning (r=0.118), evacuation distance (r=-0.257), and cost of evacuation (r=-0.121). Low to medium level inter- correlations also exists among variables (0.096 ≤ r ≤ 0.309). In the next section, logit models are estimated in order to evaluate the effects of multiple variables on mode choice.

4.3 Estimated Parameters for Mode Choice Models

Parameter estimates for the logit models for households that walked during evacuation are presented in Table 4.3 and are discussed here. The zero-order

Ref. code: 25605522300184DMX

correlations between the alternative mode and the determinants (riY) for all the models estimated are also reported in Table 4.3.

Ref. code: 25605522300184DMX

Table 4.2. Inter-correlations of variables included in the logit models Variables 1 2 3 4 5 6 7 8 9 10 11 12 13 14

1. MDEC 1

2. TDEC 0.104* 1

3. DDEC 0.036 -0.189** 1

4. AGE 0.080 -0.050 -0.049 1

5. GEN -0.002 0.072 -0.050 -0.224** 1

6. EDUC 0.009 0.002 0.080 -0.198** 0.019 1

7. CHILD -0.060 0.040 0.025 -0.227** 0.126** 0.057 1

8. HPROB -0.016 0.021 -0.002 0.278** -0.096* -0.027 -0.027 1

9. HOWN 0.036 0.151** -0.051 0.199** 0.008 -0.001 -0.131** 0.159** 1

10. YLIVE -0.049 -0.078 0.024 0.294** -0.089 -0.029 -0.092 0.097* 0.309** 1

11. VEH -0.180** -0.026 0.074 -0.024 0.036 0.127** 0.101* 0.034 0.008 -0.033 1

12. SWARN 0.118* 0.157** -0.189** 0.059 0.106* 0.005 -0.074 0.111* 0.170** 0.071 0.004 1

13. EDIST -0.257** -0.224** -0.256** -0.027 -0.075 0.010 -0.045 -0.043 -0.113* 0.057 0.008 -0.006 1

14. ECOST -0.121* -0.050 0.198** -0.153** -0.033 0.066 0.005 -0.101* 0.003 0.116* -0.046 -0.039 0.183** 1

* Correlation is significant at the 0.05 level (2-tailed). ** Correlation is significant at the 0.01 level (2-tailed).

Ref. code: 25605522300184DMX

Table 4.3. Parameter estimates of models for households that walked in evacuating Model 1: Bagong Silangan Model 2: Bahay Toro Model 3: Sto. Domingo Model 4: Bahay Toro + Sto. Model 5: Bagong Silangan + Determinants Domingo Bahay Toro + Sto. Domingo riY b S.E. riY b S.E. riY b S.E. riY b S.E. riY b S.E.

- Intercept - 4.093*** 0.707 - -0.253 0.588 - 0.849 - -1.013* 0.58 - 2.459*** 0.408 2.203***

Evacuation-related decisions

Indicators for TDEC (1 for households that 0.050 0.882** 0.44 -0.343** -2.848 0.695 - - - -0.129 -0.861* 0.441 - - - evacuated before the flood, 0 otherwise)

Indicators for DDEC (1 for households that evacuated to ------0.111 0.314* 0.189 - - - friends/families’ homes or to church/seminary, 0 otherwise)

Characteristics of the head of household and other household information

Indicators for AGE (1 for head ≥ 30 years, 0 - - - 0.085 0.426* 0.221 0.118 0.681*** 0.259 0.089 0.308** 0.143 - - - otherwise)

Indicators for GEN (1 for male household head, 0 ------0.146 1.123* 0.624 - - - otherwise)

Indicators for EDUC (1 if educational attainment of household head is higher -0.127 -0.497** 0.236 0.258** 0.602** 0.307 - - - 0.156* 0.511** 0.203 - - - than elementary level, 0 otherwise)

Ref. code: 25605522300184DMX

Indicators for CHILD (1for households with -0.131 -1.235*** 0.479 ------small children ≤ 10 years old, 0 otherwise)

Indicators for HPROB (1 if household head have ------0.142 -1.137* 0.587 ------health problem, 0 otherwise)

HOWN (1 if household own their house, 0 if ------0.073 -0.597* 0.333 - - - renting)

Indicators for YLIVE (1 for households living in - - - -0.083 -0.518* 0.285 ------the residence ≥ 10 years, 0 otherwise)

Indicators for VEH (1 if household have their own -0.234** -1.436*** 0.474 -0.064 -1.571** 0.761 -0.144* -1.157*** 0.419 -0.180** -1.106*** 0.299 vehicle, 0 otherwise)

Hazard- related variables

Indicators for SWARN (1 if the source of warning - - - 0.241** 20.047*** 0.579 - - - 0.181** 0.907*** 0.323 0.118* 0.584*** 0.227 are authorities, 0 otherwise)

Mode-related variables

Indicators for EDIST (1 if household traveled a -0.249** -1.038*** 0.276 ------0.257** -1.028*** 0.209 distance ≥ 200 meters, 0 otherwise)

Indicators for ECOST (1 if household spend cost ------0.209 1.705** 0.717 ------in evacuating by mode, 0 45

Ref. code: 25605522300184DMX

otherwise)

Number of Observations 212 132 83 215 427

-2Log-likelihood at zero 132.544 140.538 53.664 239.805 102.048

-2Log-likelihood at 96.082 96.878 39.407 207.594 48.565 convergence

Akaike Information 108.082 110.878 49.407 223.594 56.565 Criterion (AIC)

Bayesian Information 128.221 131.057 61.501 250.559 72.792 Criterion (BIC)

Likelihood ratio (LR) chi- 36.462 43.660 14.257 32.211 53.483 square (χ²)

Degrees of Freedom 5 6 4 7 3 (DF)

Prob>χ² 0.000 0.000 0.007 0.000 0.000

McFadden Pseudo-R² 0.183 0.258 0.124 0.111 0.103

Correct classification 81.6 73.5 69.9 66.0 72.1 rate (CCR) (%)

CCR for the base rate 70.6 55.1 50.1 51.6 58.4 (%)

Area under curve (AUC) 0.801 0.823 0.726 0.720 0.711

- Indicates that the independent variable is not a determinant in corresponding model; riY indicates the correlation, r, of the independent variable i, to mode outcome, Y; * indicates significance at 95%; ** indicates significance at 99%; *** indicates significance at 99.9%

Ref. code: 25605522300184DMX

4.3.1 Model 1: Bagong Silangan

Model 1 shows that significant variables include evacuation departure timing, the level of education of the head of the household (significant at p<0.01), presence of small children, vehicle ownership and evacuation distance traveled (significant at p<0.001). Households that evacuated before the floodwaters reached their place are more likely to walk (b=0.882, OR=2.415). Household heads with educational attainment higher than elementary level are less likely to walk compared to those who attained only elementary level education (b=-0.497; OR=-0.392). These households have higher probability of evacuating through a vehicle. Also, households with children that are less than or equal to 10 years old are less likely to walk compared to households that do not have small children. This is indicated by its coefficient (b= - 1.235). An addition of a child in the household will reduce the odds of walking by 70.9% (OR=-0.709). These households have higher probability of evacuating through a vehicle available to them such as the one provided by the government or their personal or rented vehicle. Moreover, those that have personal vehicle are more likely to use them when evacuating (b=-1.436; OR=-0.762). Similarly, households that traveled a distance of 200 meters or longer, are less likely to walk when going to their destinations (b=-1.038; OR=-0.646).

The ability of Model 1 to discriminate was measured by the AUC, which is 0.801. This indicates an excellent level of discrimination (Hosmer and Lemeshow, 2000). The CCR of 81.6% indicates that the model is better than prediction by chance indicated by the increment from the base CCR which is 70.6%.

4.3.2 Model 2: Bahay Toro

The significant determinants at 0.05 of Model 2 are the age of the head of the household, and number of years living in the residence. In addition, the level of education of the head of the household, and vehicle ownership are significant at 0.01, while the source of warning is significant at 0.001. The evacuation departure timing, although not significant is still included as it is believed to have an effect to mode outcome (based from Ben-Akiva and Lerman, 1985). 47

Ref. code: 25605522300184DMX

Households that evacuated before the flood are less likely to walk (b=-2.848; OR=- 0.942). This indicates that households, after heeding to the evacuation warning from the government, may have also traveled either by the vehicle provided by the government or their personal or rented vehicle. In addition, household heads older than 30 years have higher likelihood of walking when evacuating compared to younger ones (b=0.426; OR=1.531). Those that have been living in their residence for more than 10 years have lower probability of walking (b=-0.518, OR=-0.404). This indicates familiarity of the households on available transport mode and/or the provision of modes of transport by the government during evacuation, hence avail of it. Further, household heads with educational attainment higher than elementary level are more likely to walk when evacuating (b= 0.602, OR=1.825). Those with personal vehicle are less likely to walk (b=-1.571, OR=-0.792). This means that those who have vehicles take them when evacuating. Households that received evacuation warning from authorities are more likely to walk (b=2.047, OR=7.745).

The AUC of Model 2 which is 0.823, indicates an excellent discrimination ability (Hosmer and Lemeshow, 2000). Also, the addition of the variables in the model results to a CCR of 73.5%, which means that the model is better than predicting by chance. This is measured by the increment from the base CCR which is 55.1%.

4.3.3 Model 3: Sto. Domingo Model

In the case of Model 3, estimation results show that the gender and the presence of health problem of the head of the household are significant at 0.05, while the cost of evacuation and age of the head of the household are significant at 0.01 and 0.001, respectively. Males have higher likelihood of walking compared to females (b=1.123, OR=3.074). On the other hand, those that have health problem are less likely to walk (b=-1.137, OR=-0.679). This is reasonable as they have constraints in walking a distance considering they have to carry some personal belongings with them. Household heads more than 30 years of age are more likely to walk (b=0.681, OR=1.976). If there is a cost associated with evacuating through a mode, the more likely that households choose to walk (b=1.705, OR=5.501). As shown in Table 4.1, 48

Ref. code: 25605522300184DMX

the cost of evacuation is positively correlated with evacuation distance (r=0.183) indicating that households that travel longer distances are more likely to spend in evacuating through a vehicle.

The ability of Model 3 to discriminate (AUC = 0.726) is at an acceptable level (Hosmer and Lemeshow, 2000). Also, the addition of the variables in the model results to a CCR of 69.9%, which is higher than 50.1%. This shows that the model is better than prediction by chance.

4.3.4 Model 4: Pooled Model for Bahay Toro and Sto. Domingo

The determinants in Model 4 are evacuation departure timing, their destination, house ownership type (significant at 0.05); age and level of education of the head of the household (significant at 0.01); and presence of vehicle and the source of warning (significant at 0.001). Households that evacuated before the floodwaters reached their place are less likely to walk (b= -0.861, OR=-0.577). Those that evacuated to their friends’/relatives’ homes or to church/seminaries are more likely to walk (b=0.314, OR=1.369). This explains that fact that the government usually provides vehicles to evacuees they haul to public evacuation centers which are government-designated facilities. Moreover, households that own their own house are less likely to walk (b=- 0.597, OR=-0.450). It can be observed from Table 4.1, that house ownership is positively correlated with the presence of health problem (r=0.159). This explains their higher probability not to walk. In addition, household heads who are 30 years or older have higher likelihood of walking when evacuating (b= 0.308, OR=1.361). Similarly, household heads with educational attainment higher than elementary level have higher probability to walk when evacuating (b= 0.511, OR=1.667). Consistent with the effect in the models presented earlier, those with personal vehicle are less likely to walk (b= -1.157, OR=-0.686). An additional vehicle to the household reduces the odds of walking by 68.6%. Also, those that received evacuation warning from authorities are more likely to walk (b= 0.907, OR=2.477).

The model’s ability to discriminate (AUC = 0.720) indicates an acceptable level (Hosmer and Lemeshow, 2000). The addition of variables in the model resulted to a 49

Ref. code: 25605522300184DMX

CCR of 66% with a significant increase from the base CCR of 51.6%. This shows that the model is able to predict better than just by chance.

4.3.5 Model 5: Pooled Model for all sub-districts

For Model 5, determinants include presence of personal vehicle, source of evacuation warning and evacuation distance traveled. Those with personal vehicle are less likely to walk (b=-1.106, OR=-0.669), while households that received evacuation warning from authorities are more likely to walk (b=0.584, OR=1.793). Households traveling a distance greater than or equal to 200 meters when going to their destinations have lower likelihood of walking when compared to those that travel shorter distances (b=- 1.028, OR=-0.642). Traveling more than 200 meters reduces the odds of walking by 64.2%.

The model’s ability to discriminate also indicates an acceptable level (AUC = 0.711) (Hosmer and Lemeshow, 2000). The resulting CCR of 72.1% is higher than the base CCR of 58.4%. This means that the model is better than predicting by chance.

4.4 Comparisons of results of the estimated mode choice models

The following observations can be highlighted in comparing the model results. • Vehicle ownership is a significant determinant to all models, except for Model 3. It is significant at 0.01 in all models with the same negative effects. This shows similar behavior of households with personal vehicle that use them when evacuating. This finding goes with Deka and Carnegie (2010a) and Lindell and Prater (2007). • Common to Models 1, 2, 4 is the educational level attained by the household head, significant at 0.01 in all three models. However, the effect of the variable to Models 2 and 4 are positive while having negative effect to Model 1. Households in Model 1 which are educated higher than elementary level have lower likelihood of walking contrary to households in Models 2 and 4. The difference may be due to the correlation existing between the level of

Ref. code: 25605522300184DMX

education and vehicle ownership (r=0.127) as shown in Table 4.2. These are both related to the economic status of the household. • A determinant common to Models 2, 3, 4 is the age of the household head with the same effects. Household heads that are 30 years or older have higher likelihood of walking when evacuating. • Departure timing is significant in Model 1 and 4 with opposite effects, while included in Model 3 although it is not significant, but believed to have an influence in the decision making. Opposite effects in Model 1 and 4, indicates difference in behavior of risk-averse households in Bagong Silangan against Bahay Toro and Sto. Domingo. Households in Bagong Silangan are more likely to walk while less likely in the other 2 sub-districts. The difference may be related to the greater distance that households in the other 2 sub-districts have to travel when going to their destinations. It is shown in Table 4.1 that almost 93% of the households in Bahay Toro and Sto. Domingo travel distances greater than 400 meters, while only more than 50% in Bagong Silangan. Also in Table 4.2, there exists a negative correlation (r=-0.224) between departure timing and evacuation distance. • Distance traveled during evacuation is common to Models 1 and 5 with the same level of significance and negative effect. The result is plausible as the greater the distance households have to travel, the more likelihood that they do not evacuate on-foot. • The destination type and presence of small children is unique in Model 1. Result regarding households’ destination as a determinant to mode choice behavior supports findings in Sadri et al. (2014a). With regards to having small children in the household, it is logical that these households require assistance in evacuation through a vehicle available. • The number of years the households have been living in the residence is a unique determinant to households in Bahay Toro. The households living in the area for more than 10 years are less likely to walk. It should also be noted that the number of years in the residence is positively correlated with the presence of health problem (r=0.097) and house ownership (r=0.309). This specifies that those living in the area for 10 years or more, may have health problem,

Ref. code: 25605522300184DMX

own their house and are also aware of mode of evacuation that are available, and take them when evacuating instead of walking. On the other hand, the presence of health problem, the gender of the head of the household and cost of evacuation are determinants taken into account only by households in Sto. Domingo, but not in other sub-districts. In addition, type of house ownership is a determinant to Model 4, where households owning the house have lower probability of walking. • The R2 of separate Models for each sub-district: Model 1 (0.183), Model 2 (0.258), Model 3 (0.124) are higher than that of the R2 of pooled models (Model 4=0.111; Model 5=0.103). Model 2 has slightly higher value compared to that of Sadri et al. (2014a) with R2=0.236, while the rest of the models have lower values. However, AUCs for Model 1 (0.801) and Model 2 (0.823) indicate excellent level while for Model 3 (0.723) is at acceptable level slightly higher than those of the pooled models (Model 4 = 0.720; Model 5=0.711). This shows the usefulness of the models estimated.

4.5 Validation of the Estimated Mode Choice Models

The LR based validation was conducted to test the validity of the pooled model specifications versus separate sub-district models. The procedure described in section 3.4 was employed. Model 4 was the first case tested against the parts of the data (Bahay Toro and Sto. Domingo) estimated with the same model specifications. The log-likelihood at convergence is recorded for each estimation result. The LR is then calculated using Equation 6, where -2LL(βmodel4)=207.59, -2LL(βSto Domingo)=84.41 and

-2LL(βBahay Toro)=107.12, are the log-likelihood at convergence of the model estimated using Model 4, Sto. Domingo and Bahay Toro data, respectively. This results to an LR of 16.07, which is χ2-distributed with degrees of freedom equal to 7. The critical 2 2 value of χ for significance level of 5%, χ 0.05,7 equals 14.07, which is less than the resulting LR. Hence, the null hypothesis is not rejected. Therefore, the validity of the model specification is supported. This indicates that Model 4 specification is established over sub-districts Sto. Domingo and Bahay Toro.

Ref. code: 25605522300184DMX

The second case tested was the Model 5, estimated using the pooled data of 427 observations against the data used for Model 4 and Model 1, estimating it with the same model specification with that of Model 5. The resulting -2LL at convergence for the estimations for βmodel5, βSto. Domingo+Bahay Toro and βBagong Silangan are 48.565, 31.867 and 30.776, respectively. Using the same Equation 6, resulting LR with degrees of 2 freedom equal to 3, χ 0.05, 3 is -14.078. This indicates that the parameters of Model 5, is not established over subsamples from the 3 sub-districts. Therefore, separate models 1 and 4 are more useful than the pooled Model 5.

4.6 Summary of Findings on the Mode Choice Behavior of Evacuees

In this chapter, the determinants of evacuees’ mode choice in the context of Quezon City, Philippines is investigated. Specifically, determinants that households take into account when evacuating on foot or by taking other modes of transport, such as their own vehicle, rented vehicle or vehicle provided by the government were studied. Analysis using separate data for sub-districts as well as pooled ones was presented and discussed.

Findings show that pooled Model for Bahay Toro and Sto. Domingo (Model 4), and separate model for Bagong Silangan (Model 1) are more useful than the pooled Model for all 3 sub-districts (Model 5). Bagong Silangan households’ mode choice is determined by a combination of determinants including evacuation departure timing, the level of education of the head of the household, presence of small children that are 10 years or younger, presence of personal vehicle and evacuation distance traveled. Pooled Bahay Toro and Sto. Domingo model (Model 4) determinants are evacuation departure timing, their destination, age, and level of education of the head of the household, house ownership type, presence of vehicle and the source of warning. Similar behavior of households with personal vehicle is observed, that is, taking their vehicles when evacuating. This finding is consistent with earlier studies such as Deka and Carnegie (2010a) and Lindell and Prater (2007).

Results in this study provide insights that can be used by evacuation planners and managers in preparing plans for future evacuation. First, the government can 53

Ref. code: 25605522300184DMX

encourage the households that have their own vehicle to use them in future evacuations. For those lacking vehicles, but needing them because of disability or evacuation distance, the government can arrange for their rides with peers (e.g. friends, relatives, neighbors). Second, the government can encourage those that evacuate later to do pre-emptive evacuation and encourage those that travel shorter distances to walk, while providing vehicle for those that need to travel longer distances.

Ref. code: 25605522300184DMX

Chapter 5 Understanding Route Choice Behavior of Selected Population in Quezon City

The results including the details of the set of data that were used in the analysis, the parameter estimates, validation, and model performance are presented and discussed in this section.

5.1. Data Used for Route Choice Analysis and Resulting Descriptive Statistics

For analysis of route choice in this chapter, a total of 440 observations was utilized: 254, 103, 83 are from Bagong Silangan, Bahay Toro and Sto. Domingo, respectively. Data were used to estimate various models for route choice of households. These different types of models were estimated in order to see the difference in results from a disaggregated to a more aggregated level of Model.

For the selection of variables for route choice models, the same procedure that was done in selecting the variables in the mode choice models as discussed earlier was done. First, all variables including the respondents’ age, gender, type of work, and marital status of the household head, number of household members, presence of small children aged less than or equal to 10 years old, and presence of elderly aged greater or equal to 60, presence of pet and vehicle ownership, flood level, distance of household from the source of flood, and the evacuation household decisions variables (evacuation decision, departure time, destination, and mode choice), were included. Then each variable’s p-values were assessed. The variables with p≤0.05, were removed one at a time, until the variables included in the model. Insignificant variables were removed one at a time. After a variable is removed, the variables left were subjected to statistical test. The process is repeated until the desired combination of variables that gave a significant model is met. Table 5.1 presents the resulting variables included in the models including categories and percentage of data.

Ref. code: 25605522300184DMX

Result of the model fit was investigated to check whether the number of sample affected the model results and hence combined data to improve results. Models 6, 7 and 8 were estimated using data from Bagong Silangan, Bahay Toro and Sto. Domingo sub-districts, respectively. Model 9 was estimated by combining the data due to the small number of samples for Sto. Domingo, while Model 10 is a combined model using data from the 3 sub-districts. The purpose was for estimating the Model 10 by combining all data from the three sub-districts.

Table 5.1 presents the percentage of data and the categories selected for analysis. From the data, 2 route options were available to the households. These route options are defined as nearest and familiar (the main road most familiar that they usually take during normal days but not the nearest). In Bagong Silangan data (used to estimate Model 6), there were more than 34% that took the familiar route while the rest of more than 65% took the nearest route. More than 31% and 68% of the respondents took the familiar and nearest route respectively, used for Model 7 estimation. For Model 8, data shows more than 35% went through familiar route while more than 64% went through the nearest route. Also, data for estimation of Models 9 and 10 show that more than 33% and 66% evacuated through the familiar and nearest routes respectively. Table 5.1 shows details of the independent variable categories and the percentage in the data.

Table 5.1. Variables included in each model, categories, and percentage Percentage in Data (%)

Variables Categories Model Model Model Model Model 6 7 8 9 10

Dependent Variable

Familiar 34.6 31.1 35.4 33.0 33.9 Route choice (ROUTE) Nearest 65.4 68.9 64.6 67.0 66.1

Characteristics of the head of household and other household information

Age of the head of the 20-30 years old 11.8 23.3 17.1 20.5 15.5 family (AGE) 31-40 years old 34.6 27.2 20.7 24.3 30.3

Ref. code: 25605522300184DMX

41-50 years old 24.0 18.4 23.2 20.5 22.6

>50 years old 29.5 31.1 39.0 34.6 31.7

Number of household 1-4 39.0 35.0 41.5 37.8 38.5 members (MEM) >5 61.0 65.0 58.5 62.2 61.5

Presence of senior No senior citizen 88.2 85.4 87.8 86.5 87.5 citizen (SEN) Presence of senior 11.8 14.6 12.2 13.5 12.5

Presence of small No small child 40.9 37.9 17.1 28.6 35.8 children (CHILD) Small child is present 59.1 62.1 82.9 71.4 64.2

Elementary 30.7 23.3 18.3 21.1 26.7

Educational attainment High School 46.9 51.5 56.1 53.5 49.7 of the head of the household (EDUC) Diploma/Undergraduate 15.0 12.6 20.7 16.2 15.5 Graduate 7.5 12.6 4.9 9.2 8.2

Type of work of the Part time 37.4 36.9 39.0 37.8 37.6 head of the household (TWORK) Full time 62.6 63.1 61.0 62.2 62.4

House ownership Rented 12.6 24.3 51.2 36.2 22.6 (HOWN) Owned 87.4 75.7 48.8 63.8 77.4

Vehicle ownership No personal vehicle 86.6 85.4 78.0 82.2 84.7 (VEH) With personal vehicle 13.4 14.6 22.0 17.8 15.3

Hazard-related variables

Less than or equal to 1 44.1 8.7 3.7 6.5 28.2 Level of flood meter (FLEVEL) Greater than 1 meter 55.9 91.3 96.3 93.5 71.8

Evacuation-related decisions

Evacuated during the 34.3 78.6 89.0 83.2 54.9 Departure timing flood (TDEC) Evacuated before the 65.7 21.4 11.0 16.8 45.1 flood

Evacuated by other 14.6 42.7 51.2 46.5 28.0 Mode of evacuation modes (MDEC) Evacuated by walking 85.4 57.3 48.8 53.5 72.0

The explanatory variables were classified under socio-demographic information, hazard-related information and evacuation household decisions. The socio- 57

Ref. code: 25605522300184DMX

demographic variables constitute age, gender and marital status of the household head, number of household members, presence of small children aged less than or equal to 10 years old, and presence of senior citizen aged greater or equal to 60, pet and vehicle ownership. Hazard-related variables were level of flood and distance from the source of flood. The evacuation household decisions variables included evacuation decision (whether the household fully, partially or did not evacuate at all), departure time (whether the household evacuated before or during the flood), destination choice (whether the household evacuated to public evacuation shelter, church/seminary, or to friends/relatives’ house) and mode choice (whether they evacuated by walking, owned/rented vehicle, or vehicle provided by the government). These sets of explanatory variables were subjected to backward stepwise elimination to select the ones included in the models. Table 5.1 presents the summary of data and all selected explanatory variables in all models. Selected explanatory variables are age, marital status and vehicle ownership under socio-demographic information; flood level for hazard related factors; and destination and mode choices on evacuation household decisions.

5.2 Inter-correlation of Variables Included in the Route Choice Models

From Table 5.2, it can be seen that the variable strongly correlated with the route decision making is the mode of evacuation, with correlation coefficient, r, of 0.104. Other variables are also inter-correlated among each other which may indicated the level of connection among the selected variables included in the models. For instance, age of the household head is correlated with presence of senior citizen in the house (r=0.232) and the type of house ownership (r=0.197). The older the heads of the households living in the residences are, the more probable that they have a senior citizen in the home as well as they already own the residence. If the household owns the house, the more the number of household members (r=0.120). Also, the higher the level of education of the household head, having a full time work (r=0.226) has the likelihood of having a vehicle (r=0.099). Anyhow, these results show relationship of the independent variables to route choice at a time, hence the parameters estimates

Ref. code: 25605522300184DMX

taking into account interrelationships among variables are presented and discussed in the next section.

5.3 Resulting Parameter Estimates for the Models of Route Choice

In estimating the parameters of each model, familiar route, f, was considered as the basis of estimation. Parameters were estimated that constitutes all the socio- demographic and other explanatory variables. Then, backward elimination stepwise binary logit was employed to select the explanatory variables included in the model. The stepwise binary logit is known to be an efficient way to reduce a large number of variables by removing the least significant explanatory variable at a time based on the statistical test results with the highest p-value. A variable is eliminated at each step until the desired level of significance of the model is achieved. All analyses were set at 95% confidence level.

Table 5.3 shows the parameter estimates for the different Models of route choice. The

Table also presents the zero-order correlations (riY) between route options and the determinants. Models 6, 7 and 8 were estimated using data from Bagong Silangan, Bahay Toro and Sto. Domingo sub-districts, respectively. Model 9 was estimated by using combined data from Bahay Toro and Sto. Domingo. While Model 10 is a combined model using data from the 3 sub-districts. These different types of Models were estimated in order to see the difference in results from a disaggregated to a more aggregated level of Model. Model 9 was estimated by combining the data due to the small number of samples for Sto. Domingo. Result of the Model fit was investigated to check whether the number of sample affected the model results and hence combined data to improve results. The purpose was for estimating the Model 10 by combining all data from the three sub-districts. The following sub-sections presents the results of specific Models estimated.

Ref. code: 25605522300184DMX

Table 5.2. Inter-correlation of variables included in the models Variables 1 2 3 4 5 6 7 8 9 10 11 12

1. ROUTE 1

2. AGE 0.016 1

3. MEM -0.059 0.086 1

4. SEN 0.011 0.232** 0.074 1

5. CHILD -0.013 -0.268 0.298** -0.077 1

6. EDUC 0.082 -0.204 -0.011 -0.095 0.078 1

7. TWORK 0.014 -0.040 -0.001 -0.103 0.108* 0.226** 1

8. HOWN 0.083 0.197** 0.120* 0.138** -0.107 -0.005 -0.038 1

9. VEH 0.055 -0.058 0.029 0.029 0.092 0.164** 0.099* 0.005 1

10. FLEVEL 0.082 0.120* -0.041 0.053 -0.067 0.073 0.002 -0.059 -0.026 1

11. TDEC 0.082 0.023 -0.033 0.045 0.017 -0.002 0.025 0.192** -0.071 -0.195 1

12. MDEC 0.104* 0.096* 0.001 -0.023 -0.095 0.003 0.044 0.074 -0.193 -0.123 0.160** 1

* Correlation is significant at the 0.05 level (2-tailed). ** Correlation is significant at the 0.01 level (2-tailed).

Ref. code: 25605522300184DMX

Table 5.3. Parameter estimates of models for households that evacuates using the nearest route Model 10: Bagong Model 6: Bagong Model 9: Bahay Toro+Sto. Model 7: Bahay Toro Model 8: Sto. Domingo Silangan+Bahay Silangan Domingo Determinants Toro+Sto. Domingo

riY b S.E. riY b S.E. riY b S.E. riY b S.E. riY b S.E.

Intercept -0.143 0.433 -0.088 0.399 -0.423 0.645 -0.298 0.513 -0.387 0.292

Evacuation related decision

TDEC (1 for households that evacuated before the - - - 0.145 1.190** 0.645 - - - 0.133 1.106** 0.518 0.082 0.400* 0.214 flood, 0 otherwise)

MDEC indicator variable (1: households 0.075 0.771** 0.400 0.099 0.896** 0.478 0.224* 1.383* 0.560 0.160* 0.812** 0.336 0.104 0.567** 0.231 that evacuated by walking, 0: otherwise)

Characteristics of household head and other household information

AGE indicator variable (1: household head -0.076 -0.250 0.139 ------aged>30, 0: otherwise

MAR indicator variable (1: household head is ------0.107 -0.811* 0.432 - - - married; 0 otherwise)

EDUC (1 for higher than elementary graduate, 0 ------0.254* 1.057* 0.423 ------otherwise)

TWORK (1 if head - works full time; 0 ------0.574 ------1.111** otherwise)

Ref. code: 25605522300184DMX

VEH indicator variable (1: households with 0.116 1.125* 0.481 ------0.055 0.540* 0.302 personal vehicle, 0: otherwise)

CHILD indicator variable (1: households ------0.064 0.788** 0.392 - - - with children ≤ 10 years old, 0: otherwise)

SEN (1 if household have a senior aged 60 - - - 0.158 1.442** 0.820 ------and above, 0 if none)

MEM (1 for household heads with 1-4 members; 0 for household with ------0.084 -0.909 0.565 ------greater number of members)

HOWN (1 for owned ------0.194 1.059** 0.570 0.158* 0.833** 0.349 - - - house, 0 otherwise)

Hazard related variables

FLEVEL indicator variable (1: >1 meter, 0: 0.153* 0.800* 0.281 ------0.082 0.558** 0.230 otherwise)

Number of Observations 254 103 83 186 440

-2Log-likelihood at zero 80.164 41.53 81.945 96.567 84.546

-2Log-likelihood at 63.38 32.5 62.837 76.826 69.248 convergence

Akaike Information 73.38 40.85 74.837 88.826 79.248 Criterion (AIC)

Ref. code: 25605522300184DMX

Bayesian Information 91.07 51.389 89.35 108.148 99.682 Criterion (BIC)

Likelihood ratio (LR) 16.78 8.68 19.11 19.741 15.298 chi-square (χ²)

Degrees of Freedom 4 3 5 5 4 (DF)

Prob>χ² .000 .034 .002 .001 .004

McFadden Pseudo-R² .050 0.070 0.180 0.084 0.027

Correct classification 5.70% 74.80% 78.30% 71.40% 66.60% rate (CCR)

CCR for the base rate 4.74% 57.14% 54.26% 55.37% 55.16%

Area under curve (AUC) .647 0.680 0.719 0.679 0.606

- Indicates that the independent variable is not a determinant in corresponding model; riY indicates the correlation, r, of the independent variable i, to route outcome, Y; * indicates significance at 95%; ** indicates significance at 99%; *** indicates significance at 99.9%

Ref. code: 25605522300184DMX

5.3.1 Model 6: Bagong Silangan

The results of Model 6 estimates using the data from Bagong Silangan households indicate that the mode of evacuation (significant at 0.01), vehicle ownership and level of flood (significant at 0.05), and the age of the household head are significant to the decision making. The coefficient of the age of the household head (b= -0.250, OR=0.779) indicates that older household heads are more than 3 times more likely to evacuate through the nearest route than the younger ones. The presence of personal vehicle with coefficient of (b=1.125, OR=3.080), indicates that households having their own vehicle are more likely to evacuate through the nearest route when compared to those who do not have vehicle. This is reasonable according to the culture in the Philippines. The coefficient of the level of flood (b=0.800, OR=2.226) indicates that those who experienced the flood higher than 1 meter are more likely to evacuate through the nearest route when compared to others. Additionally, those who walk during evacuation are more likely to take the nearest route as indicated by the evacuation mode coefficient of 0.771 (OR=2.162).

Model 6’s ability to discriminate, with AUC of 0.647 indicates a level below acceptable according to Hosmer and Lemeshow (2000). In addition, the CCR of 65.70%, when compared to the base rate of 54.74% shows an increase in the value. This indicates that the model’s ability to discriminate is better than prediction by chance.

5.3.2 Model 7: Bahay Toro

Significant variables for Model 7 according to the estimates are departure timing and vehicle ownership, and presence of senior citizen in the home, all significant at 0.01. Those that depart to their destinations before the flood reaches their homes are more likely to take the nearest route, as indicated by the coefficient of 1.190 (OR= 3.287). The presence of vehicle in the home also increases the probability of taking the nearest route as can be seen by the coefficient of 0.896 (OR= 2.450). The presence of

Ref. code: 25605522300184DMX

senior citizen in the household (b=1.442, OR=4.229) increases the likelihood of taking the nearest route when evacuating.

The Model’s AUC of 0.68, indicating the level of discrimination it can perform, is at a level below acceptable of 0.70 (Hosmer and Lemeshow, 2000). The CCR of 74.8%, is higher than the base rate of 57.14% shows a much improved value, which indicates that the model with additional parameters is better than prediction by chance.

5.3.3 Model 8: Sto. Domingo

In this Model 8, results show that mode of evacuation and the educational attainment (significant at 0.05), type of work of the household head and the house ownership type (significant at 0.01), and number of household members are variables that are significant to route choice. Households that walk during evacuation are more likely to take the nearest route as indicated by the coefficient (b=1.383; OR=3.987). The higher the level of education of the household head than elementary level, with the coefficient of 1.057 (OR=2.877), the more likely that the household chooses to take the nearest route. When the household head works full time, the less likely that the household will take the nearest route in evacuating, as indicated by the coefficient of - 1.111 (OR=0.329). If the household owns the house they are living in, the more likely the members take the nearest route when evacuating. This is shown in the coefficient of 1.059 (OR=2.883). At the same time, the more the number of household members, the less likely that they will take the nearest route (b= -0.909; OR=0.403).

The Model’s AUC of 0.719, indicates its level of discrimination which is at an acceptable level (Hosmer and Lemeshow, 2000). The CCR of 78.3%, greater than the base rate of 54.26% shows a much improved value. This means that the model is able to do better prediction than doing just by chance.

5.3.4 Model 9: Bahay Toro + Sto. Domingo

Ref. code: 25605522300184DMX

For Model 9, the evacuation departure timing, mode of evacuation, presence of child with age equal or less than 10 years, and type of house ownership (significant at 0.01) and the marital status of the household head (significant at 0.05), are variables that are significant to route choice. Households that evacuate before the flood reach their homes have higher probability of taking the nearest route (b=1.106; OR=3.022). Households that walk when evacuating also increases the probability of taking the nearest route as can be seen by the coefficient of 0.812 (OR=2.252). Also, the presence of at least a child less than or equal to 10 years old, increases the likelihood of taking the nearest route. This is indicated by coefficient of 0.788 (OR=2.200). In addition, if the household owns the house they are living in, the more likely that members take the nearest route when evacuating (b=0.833; OR=2.300). When the household head is married, (b= -0.811; OR=0.444), the higher the likelihood of taking the nearest route.

The Model has an AUC of 0.679. This shows that the level of discrimination it can perform is lower than an acceptable level (Hosmer and Lemeshow, 2000). The CCR of 71.4%, compared to the base rate of 55.37% shows a much improved value. This means that the model with additional parameters is better than prediction by chance.

5.3.5 Model 10: Bagong Silangan+ Bahay Toro + Sto. Domingo

Model 10 shows the significant variables to route choice are the mode of evacuation and the flood level (significant at 0.01), as well as departure timing and vehicle ownership (significant at 0.05). Households that walk when evacuating increases the probability of taking the nearest route as can be seen by the coefficient of 0.567 (OR=1.763). The coefficient of the level of flood (b=0.558, OR=1.747) indicates that those who experienced the flood higher than 1 meter are more likely to evacuate through the nearest route when compared to others. Departure timing with b=0.4 (OR=1.492) means that the household is more likely to take the nearest route when evacuating before the flood reach their homes. Vehicle ownership with coefficient of (b=0.54, OR=1.716) indicates that households having their own vehicle are more

Ref. code: 25605522300184DMX

likely to evacuate through the nearest route when compared to those who do not have vehicle.

The Model’s AUC of 0.606, indicates a level of discrimination lower than acceptable according to Hosmer and Lemeshow (2000). The CCR of 66.6% is also higher than the base rate of 55.16%. This shows that the model with additional parameters is better than just predicting by chance.

5.4 Comparison of Results of the Estimated Route Choice Models

Comparing the various model results, main observations and conclusions are made as follows: • A significant variable common to all Models is the mode of evacuation. This is an indication that the mode available to households when evacuating can be one of the most dominating explanatory variables to be included in a generalized route choice model for the flood affected households in Quezon City. Their effect to the household decision on route choice seems to be consistent regardless of the geographical location of the data source. The mode of evacuation was also found as an important determinant to route choice in the study conducted by Sadri et al. (2014c). They found that evacuees taking a car to evacuate indicate that their probability of taking specific route through the McArthur Causeway decreases by 0.152. • Common to Models 7 and 9 are the departure timing and mode of evacuation. Households that depart before the flood reaches their homes are most likely to take the nearest route than those that are departing when floods had reached their homes. Sadri et al. (2014b) also found in their study that evacuees departing well ahead of the hurricane landfall choose to take the route familiar to them, rather than the government recommended route. Moreover, mode of evacuation is consistent with findings in other models, as mentioned in the earlier point above.

Ref. code: 25605522300184DMX

• Common to Models 8 and 9 is the house ownership. The more likely the households take the nearest route when they own the house. This variable is a new determinant found in this study. • Common to Models 6 and 10 is the level of flood factor. Households that have experienced flood level more than 1 meter in their previous flood experience have higher probability of taking the nearest route when evacuating. • The R2 for separate Models (Model 6= 0.050, Model 7= 0.070, Model 8= 0.180, Model 9=0.084, Model 10=0.027), are much lower than that of the models in Sadri et al. (2014c) of 0.236. This indicates very low model fit except for the Sto. Domingo Model (Model 8), which is at an acceptable level. • Findings in this study also indicates some level of significance for some other determinants including the marital status and the level of education of the household head, age, presence of children and senior in the home, number of household members and the flood level. Some of these such as the age, presence of children and income of evacuees were also found a significant determinant in a study by Sadri et al. (2014a) on mode choice. These variables were not found significant in the study of Sadri et al. (2014c) on route choice. Hence finding results in this current study.

5.5 Validation of Estimated Route Choice Models

The LR based validation as described in section 3.4, was also done to test the validity of the pooled model specifications versus separate sub-district models. Model 9 was the first case tested against the parts of the data (Bahay Toro and Sto. Domingo) estimated with the same model specifications. The log-likelihood at convergence is recorded for each estimation result. The LR is then calculated using Equation 6. The log-likelihood at convergence of the model estimated using Model 9, Sto Domingo and Bahay Toro data, are 76.83, 55.95 and 44.06, respectively. Calculated LR is equal to -20.19. This is χ2-distributed with degrees of freedom equal to 5. The critical value 2 2 of χ for significance level of 5%, χ 0.05,5 equals 11.07, is less than the resulting negative LR. Therefore, Model 7 and Model 8 are established over Model 9.

Ref. code: 25605522300184DMX

For the second case, Model 10 was tested against Model 6 and 9 estimated with the same model specifications. The log-likelihood at convergence is recorded for each estimation result are 69.25, 39.72 and 46.58 for Models 10, 6 and 9, respectively. The resulting LR is equal to -17.05, with degrees of freedom of 4. The critical value of χ2 2 for significance level of 5%, χ 0.05,4 equals 9.49. Result shows that the critical value is less than the resulting negative LR. Therefore, Model 6 and Model 9 are established over Model 10.

This farther indicates that the since Model 7 and 8 are established over Model 9. Thus, the models of individual sub-districts (models 6, 7 and 8) are preferred models over pooled models (models 9 and 10).

5.6 Summary of Findings on Route Choice Behavior of Evacuees

In order to further investigate travel behavior of people during emergency evacuation, route choice behavior is studied here in the context of households in selected sub- districts in Quezon City, Philippines that were affected during the flood event in 2013. These are Bagong Silangan, Bahay Toro and Sto. Domingo. After data was verified and cleaned, data used for analysis of route choice included 254, 103 and 83 for Bagong Silangan, Bahay Toro and Sto. Domingo respectively. Explanatory variables explored for route choice modeling include the type of evacuation of household whether partial or full evacuation, their departure timing (whether they evacuated before or during the flood), the mode of transport they used to evacuate (whether they walked or evacuated by a vehicle) and their destination (whether they went to public evacuation center, church/seminary or to their friends’ or relatives’ house) in addition to their socio-demographic information and some hazard related factors.

Verified on the map of available routes to possible evacuation destinations for households, the route choices include the nearest and familiar routes. Familiar route is the main road that households usually take during normal days and not the nearest

Ref. code: 25605522300184DMX

route. While the nearest route is the shorter route available to them. In estimating Model parameters, familiar route was the basis of estimation. Five models were estimated. The first three was developed by using data from each of the sub-district levels including Bagong Silangan, Bahay Toro and Sto. Domingo, respectively. Model 9 was investigated using combined data for Sto. Domingo and Bahay Toro. Model 10 was estimated using the combined data from the three sub-districts. These different models were investigated to ensure that the smaller number of samples for data especially from households in Sto. Domingo and Bahay Toro does not affect model fit.

The analysis results show that a common significant variable to all models is the mode of evacuation, indicating this is one of the most dominating explanatory variables to be included in a generalized route choice model for the flood affected households in Quezon City. Their effect to the household decision on route choice seems to be consistent regardless of the geographical location of the data source. Also, common to Models 7 and 9 are the departure timing and mode of evacuation where households that depart before the flood reaches their homes are most likely to take the nearest route than those that are departing when floods had reached their homes. Models 8 and 9 has a common significant variable of house ownership, where households more likely take the nearest route when they own the house. Models 6 and 10 shows level of flood as a common significant factor, where households have higher likelihood of taking the nearest route when evacuating. Further, age, income, presence of children less than 18 years old and the destination type do not significantly affect the route choice behavior of households. These variables were found to significantly affect the route choice behavior in the context of hurricane in Sadri et al. (2014b). Although these variables are also investigated in this study, some other variables resulted to have significant effect to route choice decision-making in this context. The difference in the context as well as the hazard type might have contributed to difference in results. Also, the thresholds used in some investigated variables in this study are very much different compared to past studies. However, some of the variables found in this study that are consistent with earlier findings include the

Ref. code: 25605522300184DMX

timing of evacuation and mode of evacuation. Sadri et al. (2014b) found that those that evacuate well ahead of time choose to take routes that are familiar to them.

Ref. code: 25605522300184DMX

Chapter 6 Conclusions and Recommendations

The frequency and severity of hazards that lead to considerable disaster impacts are continuously increasing. The need of devising preparedness plans for future evacuations is imperative in order to minimize disaster impacts to people. Modeling helps in evacuation planning as it will predict the most probable evacuation outcomes (e.g., cumulative percentage of the risk area population who evacuate over time, location of bottlenecks, etc.) if the analysts use accurate inputs for evacuation demand and capacity. Within the modeling efforts, evacuation behavior has been realized as an important consideration in modeling, hence, this has been the focus of evacuation modeling research in the past. Incorporating evacuees’ behavior is helpful for efficient evacuation operations in the future. Evacuation behavior aspects are identified as the decision to evacuate or stay, departure timing, destination and shelter type choice, mode choice, and route choice. The two latter decisions including the mode and route choices are the focus of investigation in this study.

6.1 Conclusions

Results in this study provide insights that can be used by evacuation planners and managers in preparing plans for future evacuation. Details for each evacuation related decision investigated here are summarized in the sub sections below.

6.1.1 Flood evacuation mode choice

From the results of mode choice analysis, vehicle ownership is a significant determinant to all models, except for Model 3. This is consistent with findings in other studies such as that of Deka and Carnegie (2010a) and Lindell and Prater (2007) that households with personal vehicle use them when evacuating. Also, a determinant common to 3 of the models include educational level of the household head, although the effects are not the same in all models. The difference may be due to the

Ref. code: 25605522300184DMX

correlation existing between the level of education and vehicle ownership (r=0.127) as shown in Table 4.2. These are both related to the economic status of the household. Another determinant common to Models is the age of the household head where those that are 30 years or older have higher likelihood of walking when evacuating. Departure timing is also a determinant to 2 models but with opposite effects. The difference may be related to the greater distance that households in the other 2 sub- districts have to travel when going to their destinations. In addition, the distance traveled during evacuation is common to Models 1 and 5, indicating that the greater the distance households have to travel, the more likelihood that they do not evacuate on-foot. The type of destination is also a determinant that supports findings in Sadri et al. (2014a). The presence of small children also is a determinant and it is logical that these households require assistance in evacuation through a vehicle available. The number of years living in the residence are less likely to walk and it should be noted that this is correlated with the presence of health problem (r=0.097) and house ownership (r=0.309), meaning that those living in the area for 10 years or more, may have health problem, own their house and are also aware of mode of evacuation that are available, and take them when evacuating instead of walking. On the other hand, the presence of health problem, the gender of the head of the household and cost of evacuation are determinants taken into account only by households in Sto. Domingo, but not in other sub-districts. Type of house ownership is a determinant to Model 4, where households owning the house have lower probability of walking.

6.1.2 Flood evacuation route choice

Study findings indicate that different sub-districts decide differently based on some specific variables that determine household evacuation route choice. This is an indication that households have their own uniqueness based on their characteristics and the circumstance they face during an emergency. However, the results show the mode of evacuation available to the households, consistent in all sub-districts, is a variable that is dominating factor to households’ decision making. Households that walk when evacuating prefer to take the nearest route. Implication of this in future

Ref. code: 25605522300184DMX

evacuation is that, the nearest routes to destinations can be designated for those who will walk. Evacuees on other modes (e.g. by motorcycle, tricycle, vehicles provided by the government) can be designated to main roads or other longer routes available. This is to avoid traffic congestion of multi modes in the nearest routes.

6.2 Recommendations and Future Research

This sub section presents some recommendations that can be drawn from results of this study as well as for future research.

Although results in this study provide useful insights that can help planners and governments prepare plans for future evacuation, some limitations exist. In general, increasing sample size and collecting data in other sub-districts may be helpful. Studying the transferability of the models to other hazards (e.g. earth quake typhoon etc.) as well as other geographical areas can also be investigated next.

Also, since on pedestrian evacuation focuses at the moment on optimization and simulation based model especially focuses on building evacuation and less had been done on large scale pedestrian evacuation and travel behavior, it is worthwhile to investigate what determines evacuation decision, departure timing, pedestrian and gender-based evacuation behavior. These can be modeled and included in the evacuation simulation framework of understanding key determinants of gender-based evacuation travel behavior in the area of evacuation, departure time, destination, mode and route choice behavior. Obviously, male and female household heads behaves differently. Factors that affect their decisions may differ.

6.2.1 Flood evacuation mode choice

From results of analyzing the determinants of households’ mode choice, the government can encourage the households that have their own vehicle to use them in future evacuations. For those lacking vehicles, but needing them because of disability

Ref. code: 25605522300184DMX

or evacuation distance, the government can arrange for rides with peers (friends, relatives, neighbors). Also, the government can encourage those that evacuate later to do pre-emptive evacuation and encourage those that travel shorter distances to walk, while providing vehicles for those that need to travel longer distances. Overall, the government can mandate the households living in high flood risk areas to prepare a household evacuation plan that includes the details such as the mode that they will take or need when evacuating. Home owners in specific villages can devise their own plan that can assist the sub-district level government to implement effective evacuation in the future.

Although the Bagong Silangan Model shows an excellent predictive accuracy, the Model for Bahay Toro and Sto. Domingo is only at an acceptable level. This indicates that there might still be determinants that are not captured in the pooled Bahay Toro and Sto. Doming model. Hence, further investigation is needed in order to develop a model that can be used to predict future percentage of evacuees that use which type of evacuation mode. For further analysis, applying methodologies in handling missing data, may add significant value to model estimates. Johnson and Young (2011) found that modern methodologies of handling missing data yield better results than using traditional ones. In addition, Graham (2009) gives a literature overview of the missing data, issues and challenges and suggests practical ways to handle bias. These useful insights and methodology recommendations may be applied in future analysis. Predictive models can help evacuation planners and managers to plan for appropriate type and number of modes to provide for moving people to safety. These models can also help governments assess whether their resources in providing government vehicles are sufficient or not. If not enough, procurement of additional vehicles might be needed. If procurement is not possible, collaborations with other local government or carefully planning the number of trips during evacuation can be done. Practical example of integrating mode choice model in the evacuation planning model is detailed in Yin et al. (2014).

Ref. code: 25605522300184DMX

In the context of this study, a study is needed to explore the possibility of developing models that can be generalized to cities in the Philippines. In addition, the growing need to understand how different genders are affected by disasters and growing vulnerability of women is increasingly interesting.

6.2.2 Flood evacuation route choice

Empirical findings provide insights in route choice behavior of evacuees that are useful for emergency planning and management. An understanding of how evacuees choose the route they take during flood events help evacuation planners and managers predict behavior that can be incorporated in evacuation traffic models. Particularly, it can assist emergency planners and evacuation managers to plan and prepare according to existing facilities. Specifically, it can help them determine which route can be congested and which routes to recommend to evacuees when providing the evacuation warning. This further helps in effective evacuation operations in the future.

These results of preference on the nearest routes of specific characteristics of households indicate possible congestion on nearest routes. The government officials can plan accordingly, designate nearest to those that walk, and provide vehicles to those with elderly as possible in future evacuations. Motorized vehicles used during evacuation should be designated to the main road or other longer routes. In preparation, signage should be put in place and evacuation drills to different routes conducted for more organized evacuations in the future.

Characteristics of routes in this study are not available in the data and hence not investigated here. Also small number of samples used for data analysis might have contributed to the insignificance of these variables. Collecting more data will be helpful in developing useful prediction models. Also, the small number of samples used for data analysis might have contributed to the insignificance of these variables. Collecting more data will be helpful in developing useful prediction models. Since the routes recommended by the government officials are not made available to

Ref. code: 25605522300184DMX

households in the study area, this information should be included in evacuation warning messages in future evacuations. It is important that once routes that may be congested during evacuation are identified, distribution of evacuees to other routes available should be done to avoid delays in future evacuation.

Future research may include a comparative study of the evacuation route choice behavior of other flood affected areas in Quezon City, Philippines. Explanatory variables might vary due to the different characteristics and decision-making of different households. Other explanatory variables that were not significant and consequently removed from the analysis might be useful in understanding behavior in other areas and/or much broader context. Moreover, transferability of the models to other hazards and simultaneously modeling route choice and mode can be worth investigating. External validation of the model developed in this study could also be explored in research.

Specific studies may focus on pedestrian and gender-based flood evacuation route choice behavior. Moreover, other significant explanatory variables which are attributed as socio-demographic information are likely due to the ethnic majority in the area which can be further investigated. This information is not available and can be collected in future studies. For further modeling research, integration of some other decisions including the evacuation decision, departure time, destination, mode and route choices can be explored.

Ref. code: 25605522300184DMX

References

Books and Book Articles

Baker, E.J. (2000). Hurricane evacuation in the United States. In R. Pielke, Jr. & R. Pielke, Sr. (Eds.). Storms (Vol. 1, pp. 308–319). London: Routledge Ben-Akiva, M., Lerman, S. (1985). Discrete choice analysis. Cambridge, MA: MIT Press. Hosmer, D.W., Lemeshow, S. (2nd Ed.). (2000). Applied Logistic Regression. New York: John Wiley and Sons. Lindell, M.K., Perry, R.W. (1992). Behavioral foundations of community emergency planning. Washington, DC: Hemisphere Press Ortuzar, M., Willumsen, C. (2011). Activity based models. In Modelling Transport (4th ed). John Wiley and Sons: United Kingdom Perry, R.W., Lindell, M.K., Greene, M.R. (1981). Evacuation planning in emergency management. Lexington, MA: Heath Lexington Books Train, K. (2nd Ed.). (2009). Discrete choice methods with simulation. New York: Cambridge University Press.

Articles

Abdelgawad, H., Abduhai, B. (2010a). Managing large-scale multimodal emergency evacuations. Journal of Transportation Safety and Security, 2, 122–151. Abdelgawad, H., Abdulhai, B. (2010b). Towards a complete evacuation demand and supply modeling and management process. Paper presented at the 12th World Conference on Transportation Studies, Lisbon Portugal, July 11-15. Akbarzadeh, M., Wilmot, C. (2015). Time-dependent route choice in hurricane evacuation. Natural Hazards Review, 16(2). doi:10.1061/(ASCE)NH.1527- 6996.0000159

Ref. code: 25605522300184DMX

An, S., Cui, N., Li, X., Ouyang, Y. (2013). Location planning for transit-based evacuation under the risk of service disruptions. Transportation Research Part B, 54, 1-16 Ashalatha, R., Manju, V.S., Zacharia, A.B. (2013). Mode choice behavior of commuters in Thiruvananthapuram City. Journal of Transportation Engineering, 139, 494-502 Balakrishna, R., Wen, Y., Ben-Akiva, M., Antoniou, C. (2008). Simulation-based framework for transportation network management for emergencies. Transportation Research Record, 2041, 80–88 Bish, D.R. (2011). Planning for a bus-based evacuation. OR Spectrum 33(3), 629-654 Bish, D., Sherali, H. (2013). Aggregate-level demand management in evacuation planning. European Journal of Operational Research, 224(1), 79–92. Campos, V., Bandeira, R., Bandeira, A. (2012) A method for evacuation route planning in disaster situations. Procedia - Social and Behavioral Sciences, 54, 503–512. Chan, C.P. (2010). Large scale evacuation of carless people during short-and long- notice emergency. Dissertation, the University of Arizona Charnkol, T., Tanaboriboon, Y. (2006). Tsunami evacuation behavior analysis-one step of transportation disaster response. IATSS Research, 30(2), 83-96. Chen, Y., Xiao, D. (2008). Emergency evacuation model and algorithms. Journal of Transportation Systems Engineering and Information Technology, 8(6), 96– 100 Cherry, C., Cervero, R. (2007). Use characteristics and mode choice behavior of electric bike users in China. Transport Policy, 14, 247–257 Cova, T., Johnson, J. (2003). A network flow model for lane-based evacuation routing. Transportation Research Part A: Policy and Practice 37(7), 579–604 Cuellar, L., Kubicek, D., Hengartner, N., Hansson, A. (2009). Emergency relocation: population response model to disasters. Proceedings of the IEEE Conference on Technologies for Homeland Security, Boston, MA, USA.

Ref. code: 25605522300184DMX

Danaf, M., Abou-Zeid, M., Kaysi, I. (2014). Modeling travel choices of students at a private, urban university: Insights and policy implications. Case Studies on Transport Policy 2, 142–152 Dash, N., Gladwin, H. (2007). Evacuation decision making and behavioral responses: individual and household. Natural Hazards Review, 8, 69-77. Dash, N., Morrow, B.H. (2001). Return delays and evacuation order compliance: The case of Hurricane Georges and the Florida Keys. Environmental Hazards, 2, 119–128 Deka, D., Carnegie, J.A. (2010a). Analyzing evacuation behavior of transportation- disadvantaged populations in Northern New Jersey. Proceedings of the 89th Annual Meeting of the Transportation Research Board, Washington, DC, USA. Deka, D., Carnegie, J., Bilton, P. (2010b). What does it take for shuttles to succeed? Comparison of stated preferences and reality of shuttle success in New Jersey. Transportation Research Record, 2144, 102-110. DOI: 10.3141/2144-12 Dow, K., Cutter, S.L. (2002). Emerging hurricane evacuation issues: Hurricane Floyd and South Carolina. Natural Hazards Review, 3(1), 12–18 Elmitiny, N., Ramasamy, S., Radwan, E. (2007). Emergency evacuation planning and preparedness of transit facilities: traffic simulation modelling. Transportation Research Record, 1992, 121-126 Emergency Management Australia (2005). Evacuation planning. Australian Emergency Manual Series, 11. Fang, L., Edara, P. (2013). Sensitivity of evacuation performance estimates to evacuee route choice behavior. Paper presented at the 92nd Annual Meeting of the Transportation Research Board, Washington D.C, January 13-17. Fawcett, T. (2006). An introduction to ROC analysis. Pattern Recognition Letters, 27, 861-874 Fu, H., Wilmot, G. (2004). Sequential logit dynamic travel demand model for hurricane evacuation. Transportation Research Record, 1882, 19–26. doi:10.3141/1882-03

Ref. code: 25605522300184DMX

Galindo, G., Batta, R. (2013). Review of recent developments in OR/MS research in disaster operations management. European Journal of Operational Research, 230(2), 201-211. Gencer, E.A. (2013). Natural Disasters, Urban Vulnerability, and Risk Management: A Theoretical Overview. Springer Briefs in Environment, Security, Development and Peace, 7-43. Graham, J.W. (2009). Missing data analysis: making it work in the realworld. Annual Review of Psychology, 60, 549-576 Han, L.D., Yuan, F. (2005). Evacuation modeling and operations using dynamic traffic assignment and most desirable destination approaches. Paper presented at the 84th Annual Meeting Transportation Research Board, Washington DC, United States, 9-13 January 2005 Hasan, S., Mesa-Arango, R., Ukkusuri, S. (2013). A random-parameter hazard-based model to understand household evacuation timing behavior. Transportation Research Part C, 27, 108–116. Hasan, S., Mesa-Arango, R., Ukkusuri, S., Murray-Tuite, P. (2012). Transferability of hurricane evacuation choice model: joint model estimation combining multiple data sources. Journal of Transportation Engineering, 138(5), 548–556. Hasan, S., Ukkusuri, S., Gladwin, H., Murray-Tuite, P. (2011). Behavioral model to understand household-level hurricane evacuation decision making. Journal of Transportation Engineering, 137(5), 341–348. Hobeika, A. G., Kim, C. (1998). Comparison of traffic assignments in evacuation modeling. IEEE Transactions on Engineering Management, 45, 192–198. Hsu, Y., Peeta, S. (2013). An aggregate approach to model evacuee behavior for no- notice evacuation operation. Transportation, 40(3), 671-696. Huang, S.K., Lindell, M.K., Prater, C.S. (in press). Who leaves and who stays? A review and statistical meta-analysis of hurricane evacuation studies. Environment and Behavior. DOI: 10.1177/0013916515578485 Huibregtse, L., Bliemer, M., Hoogendoorn, S. (2010). Analysis of near-optimal evacuation instructions. Procedia Engineering, 3(0), 189–203.

Ref. code: 25605522300184DMX

Jha, M., Moore, K., Pashaie, B. (2004). Emergency evacuation planning with microscopic traffic simulation. Transportation Research Record, 1922, 40–48 Johnson, D.R., Young, R. (2011). Toward best practices in analyzing datasets with missing data: comparisons and recommendations. Journal of Marriage and Family, 73, 926-945 Klunder, G., Terbruggen, S., Mak, J., Immers, B. (2009). Large-scale evacuation of the Randstand: evacuation simulations with the dynamic traffic assignment model Indy. Paper presented at the 1st International Conference on Evacuation Modelling and Management, the Hague, the Netherlands Kwon, E,, Pitt, S. (2005). Evaluation of emergency evacuation strategy for downtown event traffic using a dynamic network model. Transportation Research Record, 1922, 149–155 Lämmel, G., Grether, D., Nagel, K. (2010). The representation and implementation of time-dependent inundation in large-scale microscopic evacuation simulations. Transportation Research Part C, 18, 84–98. Lindell, M.K., Prater, C.S., Gregg, C.E., Apatu, E., Huang, S.K., Wu, H.C. (2015). Households' immediate responses to the 2009 Samoa earthquake and tsunami. International Journal of Disaster Risk Reduction, 12, 328-340. Doi: 10.1016/j.ijdrr.2015.03.003 Lim, Ma. Bernadeth (2016). Understanding evacuation decision, departure timing and destination choice in high flood risk areas using discrete choice model. Doctoral Dissertation submitted to the Department of Civil Engineering and Technology, Sirindhorn International Institute of Technology, Thailand. Lim, M.B., Lim, H.J., Piantanakulchai, M. (2013). Factors affecting flood evacuation and its implication to transportation planning. Journal of the Eastern Asia Society for Transportation Studies, 10, 163-177 Lim, M.B., Lim, H. Jr., Piantanakulchai, M., Uy, F.A. (2015). Flood evacuation decision modeling for high risk urban area in the Philippines. Paper presented at the 3rd international conference on evacuation modeling and management, Tainan, Taiwan.

Ref. code: 25605522300184DMX

Lindell, M., Lu, J., Prater, C. (2005). Household decision making and evacuation in response to hurricane Lili. Natural Hazards Review, 6(4), 171–179. doi:10.1061/(ASCE)1527-6988(2005)6:4(171) Lindell, M.K. (2013). Disaster studies. Current Sociology, 61(5-6), 797–825. doi:10.1177/0011392113484456 Lindell, M. K., Kang, J. E., Prater, C. S. (2011). The logistics of household hurricane evacuation. Natural Hazards 58, 1093–1109. Lindell, M. K., Prater, C. S. (2007). Critical behavioral assumptions in evacuation analysis for private vehicles: Examples from hurricane research and planning. Journal of Urban Planning and Development, 133, 18–29. Liu, Y., Chang, G.L., Liu, Y., Lai, X. (2008). A corridor-based emergency evacuation system for Washington, D.C.: system development and case study. Transportation Research Record, 2041, 58-67 Lumbroso, D., Gaume, E., Logtmeijer, C., Mens, M., Van der Vat, M. (2008). Evacuation and traffic management, integrated flood risk analysis and management methodologies. Floodsite Consortium. Available at http://repository.tudelft.nl/assets/uuid:6f26c9bd-f7bf-4c75-a8fd- 56a7c7b89303/T17_07_02_Evacuation_and_traffic_management_D17_1_V4 _4_P01.pdf . Accessed 25 April 2015 Mastrogiannidou, C., Boile, M., Golias, M., Theofanis, S., Ziliaskopoulos, A. (2009). Using transit to evacuate facilities in urban areas: a micro-simulation based integrated tool. Paper presented at the 88th annual meeting of the Transportation Research Board, Washington DC, United States, 11-15 January 2009 Mesa-Arango, R., Hasan, S., Ukkusuri, S., Murray-Tuite, P. (2013). Household-level model for hurricane evacuation destination type choice using Hurricane Ivan data. Natural Hazards Review, 14(1), 11–20. doi:10.1061/(ASCE)NH.1527- 6996.0000083. Mileti, D. S., Sorensen, J. H., O’Brien, P. W. (1992). Toward an explanation of mass care shelter use in evacuations. International Journal of Mass Emergencies and Disasters, 10, 25–42. 83

Ref. code: 25605522300184DMX

Miskeen, M.A., Alhodairi, A.M., Rahmat, R.A. (2013). Modeling of intercity transport mode choice behavior in Libya: a binary logit model for business trips by private car and intercity bus. Australian Journal of Basic and Applied Sciences, 7(1), 302-311 Mitchell, S.W., Radwan, E. (2006). Heuristic prioritization of emergency evacuation staging to reduce clearance time. Paper presented at the 85th Annual Meeting Transportation Research Board, Washington DC, United States, 10-14 January 2006 Mitra, M., Buliung, R.N. (2015). Exploring differences in school travel mode choice behaviour between children and youth. Transport Policy, 42, 4–11 Modali, N. K. (2005). Modeling destination choice and measuring the transferability of hurricane evacuation patterns (M.Sc. thesis). Louisiana State University, Baton Rouge, LA. Murray-Tuite, P., Wolshon, B. (2013). Evacuation transportation modeling: An overview of research, development, and practice. Transportation Research Part C: Emerging Technologies, 27(0), 25–45. Na, L., Xueyen, S., Mingliang, Q. (2012) A bi-objective evacuation routing model with secondary evacuation expected costs. Systems Engineering Procedia Engineering, 5(0), 1–7. Naghawi, H., Wolshon, B. (2012). Performance of multimodal evacuation traffic networks: a simulation-based assessment. Natural Hazards Review 13(3), 196–204 Ng, M., Waller, S. (2010). Reliable evacuation planning via demand inflation and supply deflation. Transportation Research Part E, 46(6), 1086–1094. Noh, H., Chiu, Y.C., Zheng, H., Hickman, M., Mirchandani, P. (2009). An approach to modeling demand and supply for a short-notice evacuation. Transportation Research Record, 2091, 91–99 Paul, B.K. (2012). Factors affecting evacuation behavior: The case of 2007 Cyclone Sidr, Bangladesh. The Professional Geographer, 64(3), 401-414 Paul, B.K., Dutt, S. (2010). Hazard warnings and responses to evacuation orders: The case of Bangladesh's Cyclone Sidr. Geographical Review, 100(3), 336-355 84

Ref. code: 25605522300184DMX

Pel, A., Bliemer, M., Hoogendoorn, S., (2012). A review on travel behaviour modelling in dynamic traffic simulation models for evacuations. Transportation, 39, 97–123. Pel, A., Bliemer, M., Hoogendoorn, S. (2011). Simulating travel behavior under evacuation conditions. European Journal for Transportation Infrastructure Research, 11(2), 166-193. Pel, A., Hoogendoorn, S., Bliemer, M. (2010). Evacuation modelling including traveler information and compliance behavior. Procedia Engineering, 3, 101- 111. Perry, R.W. (1979). Evacuation decision-making in natural disasters. Mass Emergencies, 4(1), 25–38. Renne, J.L., Sanchez, T.W., Jenkins, P., Peterson, R. (2009). Challenge of evacuating the carless in ﬁve major US cities: identifying the key issues. Transportation Research Record, 2119, 36–44 Renne, J.L., Sanchez, T.W., Litman, T. (2008). National study on carless and special needs evacuation planning: a literature review, 110p Revi, A., Singh, A.K. (2007). Cyclone and storm surge, pedestrian evacuation and emergency response in India. In Pedestrian and Evacuation Dynamics 2005 (pp. 119-130). Springer Berlin Heidelberg Ricchetti-Masterson, K., Horney, J. (2013). Social factors as modifiers of hurricane Irene evacuation behavior in Beaufort County, NC. PLOS Currents Disasters, 1. doi: 10.1371/currents.dis.620b6c2ec4408c217788bb1c091ef919. Sadri, A. M., Ukkusuri, S. V., Murray-Tuite, P., Gladwin, H. (2014a). Analysis of hurricane evacuee mode choice behavior. Transportation Research Part C: Emerging Technologies, 48, 37-46. doi:10.1016/j.trc.2014.08.008 Sadri, A., Ukkusuri, S., Murray-Tuite, P., Gladwin, H. (2014b). How to evacuate: model for understanding the routing strategies during Hurricane evacuation. Journal of Transportation Engineering, 140(1), 61–69. Sadri, A., Ukkusuri, S., Murray-Tuite, P., Gladwin, H. (2014c). Hurricane Evacuation Routing Strategy from Miami Beach: Choice of Major Bridges. Paper

Ref. code: 25605522300184DMX

submitted for presentation at Transportation Research Board Conference and publication in Transportation Research Record. Sayyady, F., Eksioglu, S. (2010). Optimizing the use of public transit system during no-notice evacuation of urban areas. Computers and Industrial Engineering, 59(4), 488–495. Scolobig, A., Marchi, B., Borga, M. (2012). The missing link between flood risk awareness and preparedness: findings from case studies in an Alpine Region. Natural Hazards, 63, 499–520. Sheffi, Y., Mahmassani, H., Powell, W. B. (1981). Evacuation studies for nuclear power plant sites: A new challenge for transportation engineers. Institute of Transportation Engineers Journal, 57, 25–28. Shiwakoti, N., Liu, Z., Hopkins, T., Young, W. (2013). An overview on multimodal emergency evacuation in an urban network. In: Proceedings of the Australasian transport research forum, Brisbane, Australia Siebeneck, L. K., Cova, T. J. (2008). An assessment of the return-entry process for Hurricane Rita 2005. International Journal of Mass Emergencies and Disasters, 26, 91–111. Siebeneck, L., Cova, T. (2012). Spatial and temporal variation in evacuee risk perception throughout the evacuation and return-entry process. Risk Analysis 32(9) Siebeneck, L., Lindell, M., Prater, C., Wu, H., Huang, S. (2013) Evacuees’ reentry concerns and experiences of Typhoon Ike. Natural Hazards, 65(3), 2267- 2286. Simonovic, S., Ahmad, S. (2005). Computer-based model for flood evacuation

emergency planning. Natural Hazards, 34, 25–51. Sinha, A.K., Avrani, S.U. (1984). The disaster warning process: A study of the 1981 Gujarat cyclone. Disasters, 8(1), 67-73 Sorensen, J.H. (2000). Hazard warning systems: Review of 20 years of progress. Natural Hazards Review, 1, 119-125

Ref. code: 25605522300184DMX

Stepanov, A., Smith, J. (2009). Multi-objective evacuation routing in transportation networks. European Journal of Operational Research, 198(2), 435–446. Steyerberg, E., Borsboom, G., van Houwelingen, H., Eijkemans, M., Habbema, D. (2004). Validation and updating of predictive logistic regression models: a study on sample size and shrinkage. Statistics Medicine, 12, 2567–2586. Stopher, P., Rose, J., Alsnih, R. (2004). Dynamic Travel Demand for emergency evacuation: the case of bushfires. Working Paper, Institute of Transport Studies. Thomas, S.J., Lieberman, E. (2010). Addressing uncertainty in travel modalities for an emergency evacuation. Paper presented at the National Evacuation Conference, New Orleans, Louisiana, United States Vanlandegen, D.L. (2010). Microsimulation of large-scale evacuations utilizing metrorail transit. Master of Science Thesis, Northern Illinois University Whitehead, J., Edwards, B., Van Willigen, M., Maiolo, J., Wilson, K., Smith, K. (2000). Heading for higher ground: Factors affecting real and hypothetical hurricane evacuation behavior. Environmental Hazards, 2(4), 133–142.Wang, H., Andrews, S., Daiheng, N., Collura, J. (2010). Scenario-based analysis of transportation impacts in case of dam failure flood evacuation in Franklin county, Massachusetts. Paper presented at the 89th Annual Meeting of the Transportation Research Board, Washington DC, United States, 10-14 January 2010 Williams, B., Tagliaferri, A., Meinhold, S., Hummer, J., Rouphail, N. (2007). Simulation and analysis of freeway lane reversal for coastal typhoon evacuation. Journal of Urban Planning and Development, 133(1), 61–72 Wolshon, B., Urbina, E., Wilmot, C.G., Levitan, M. (2005). Review of policies and practices for typhoon evacuation I: Transportation planning, preparedness, and response. Natural Hazards Review, 6(3), 129–142 Wu, C., Lindell, K., Prater, S. (2012). Logistics of hurricane evacuation in Hurricanes Katrina and Rita. Transportation Research Part F, 15 (5), 445–461. Wu, H.C., Lindell, M.K., Prater, C.S., Huang, S.K. (2013). Logistics of hurricane evacuation in Hurricane Ike. In J. Cheung and H. Song (Eds.), Logistics: 87

Ref. code: 25605522300184DMX

Perspectives, Approaches and Challenges (pp. 127-140). Hauppauge NY: Nova Science Publishers Yin, W., Murray-Tuite, P., Ukkusuri, S., Gladwin, H. (2014). An agent-based modeling system for travel demand simulation for hurricane evacuation. Transportation Research Part C, 42, 44–59.

Other Materials

EM-DAT, Université catholique de Louvain (2014). Philippines country profile on natural disasters. Retrieved from www.emdat.be National Disaster Risk Reduction Management Committee (NDRRMC) (2010) Republic Act No. 10121 Philippine Disaster Risk Reduction and Management Act of 2010. Available online at http://www.ndrrmc.gov.ph/attachments/article/95/Implementing_Rules_and_R egulartion_RA_10121.pdf Social Services Development Department (SSDD) (2013) Report on affected and evacuated families during flood in August 2013. Unpublished report. Quezon City Government and Earthquake and Megacities Initiative, QCG, EMI. (2013). Disaster risk reduction and management plan 2014-2020. Building a disaster resilient Quezon City project. Quezon City Planning and Development Office, QCPDO (2013). Quezon City actual and projected population by district and by barangay 2010-2020. Unpublished report. UNISDR. (2007). The Hyogo Framework for Action: 2005-2015: Building the resilience of nations and communities to disasters. Extract from the final report of the World Conference on Disaster Reduction (A/CONF.206/6). Retrieved from http://www.unisdr.org/files/1037_hyogoframeworkforactionenglish.pdf UNISDR (2012) Towards a post-2015 framework for disaster risk reduction. Available at

Ref. code: 25605522300184DMX

www.unisdr.org/files/25129_towardsapost2015frameworkfordisaste.pdf. Accessed February 2014 UNISDR (2015) The Sendai Framework for Disaster Risk Reduction 2015-2030. Available at http://www.wcdrr.org/uploads/Sendai_Framework_for_Disaster_Risk_Reducti on_2015-2030.pdf. Accessed 1 May 2015

Ref. code: 25605522300184DMX

Appendices

Ref. code: 25605522300184DMX

Appendix A Survey Questonnaire

Household Area Code/Location: ______

This survey form is intended for data collection for the research project entitled “Modeling Travel Behavior for Flood Evacuation” being conducted by the SIIT graduate student. Targeted to the head of the household (male/female), the questions here are based on information and travel-related decisions that you have made during floods at the period of typhoon “Maring” which happened during late of August this year. Floods during this period caused a total damage of and affected in various ways. The end goal of this study then is to be able to develop/recommend preparedness measures to minimize the loss of lives and property damage in case of future floods.

Your effort in providing your answers as accurate as you can, is highly appreciated in order to produce realistic and reliable data for development of evacuation models. Please be ensured that the personal information you will provide in this questionnaire will be kept confidential and only the data analysis results will be included in the report of the study. Your participation in this survey does not cost you anything. You just have to complete this questionnaire and hand over to the person in charge.

Part I. Socio-demographic & Household-relevant Information:

Please tick appropriate box of your answer and provide your answer to items that require details. 1. Age: ______years 2. Gender: ☐ Male ☐ Female 3. Marital Status: ☐ Single ☐ Married ☐Widower ☐Others (specify): ______4. Number Household Members: ______

Member #: Relationship (wife/husband/child/grandchild, Age etc.) 1 2 3 4 5 6 7 91

Ref. code: 25605522300184DMX

8 9 10 11

5. Level of Education: ☐ Primary ☐ Elementary ☐ Highschool ☐ Diploma ☐ Undergraduate ☐ Graduate ☐ Others (specify): ______6. Household Income per month: ______7. How far is your home located from the source of flood? ______meters/kilometer 8. Does your family have personal vehicles? ☐ Yes ☐ No If yes, how many and what type of vehicles do the family own? ☐ Jeep: how many? ______☐ Motorcycle: how many? ____ ☐ Others (specify): ______how many? _____ 9. Do you have any boat/life vest or any equipment prepared for flood? ☐ Yes ☐ No 10. a. Did you experience flood in the past even before “Maring” period which happened August this year? ☐ Yes ☐ No b. If your answer above (10.a) is no, do you have knowledge about flood before flood “Maring” happened? ☐ Yes ☐ No c. If your answer above (10.a) is yes, how many times have you experienced flood? ☐ once ☐ twice ☐ more than twice ☐ Others (specify): ______d. During most of your experience, did you evacuate your place? ☐ Yes ☐ No

Part II. Information Related to Flood Experience During “Maring” 11. Did you experience flood during the typhoon “Maring” period which happened August this year? ☐ Yes ☐ No If your answer above is no, do not answer the following questions. If the answer is yes, continue answering from question number 12. 12. Before the flood, did you hear an advice to evacuate your place due to possibility of flooding? ☐ Yes ☐ No 12.a. Where did you get most of your news about the flood/evacuation advise? ☐TV ☐Radio ☐Friend/family ☐head of village/barangay ☐Others (specify): ______12.b. According to the warning, what time of the day was flood expected to reach your home? ______☐ AM (0:01-6:00) ☐ Mid-day (6:01-12:00) ☐PM (12:01-6:00) ☐ NIGHT (6:01-12:00) 13. Did you evacuate from your house according to the evacuation advise? ☐ Yes ☐ No

Ref. code: 25605522300184DMX

13.a. If your answer in 23 is no, do not continue answering the following questions. 13.b. If your answer is yes, please answer the following questions: i). When did you evacuate in relation to when the flood reached your home? ☐ hours before ☐ when flood is flowing to place ☐ when my home is flooded ☐ others (specify): ______Please indicate the time of the day when you evacuated based on your answer above. (e.g.: 1:00pm, 9:00am, 6:00pm, etc.) ______☐ AM (0:01-6:00) ☐ Mid-day (6:01-12:00) ☐PM (12:01-6:00) ☐Night (6:01-12:00) ii) Did you evacuate with the whole members of the family, or some members were left at home and evacuated later? ☐ Evacuated the whole family ☐ Some evacuated and some were left home ☐ Others (specify): ______iii) Where did you go when you evacuated? ☐ evacuation center provided by government ☐ friends’/relatives house ☐ workplace ☐others (specify): ______Please indicate where the place is located? (e.g. QC High School, Holiday Inn, etc.) ______

iv) When you evacuated, which type of vehicle did you use? ☐ own car ☐ vehicle provided by government/administration ☐ public taxi ☐ rented van with friends/relatives ☐ others (specify): ______v) Around how many meters/kilometers did your travel? (for example: 20 km, 100 km). Specify here: ______

14. When you evacuated, which route did you take? ☐ nearest route ☐ familiar route ☐the route with no floodwaters ☐ others (specify): ______Please indicate the specific route or name of the road you took. ______

Ref. code: 25605522300184DMX

Bagong Silangan, Quezon, City, Philippines Source: Google Maps (2013)

Ref. code: 25605522300184DMX

Bahay Toro, Quezon, City, Philippines Source: Google Maps (2013)

Ref. code: 25605522300184DMX

Sto. Domingo, Quezon, City, Philippines Source: Google Maps (2013)

Part III. Suggestions/comments on problems faced during floods or on how to improve conditions in future flood/evacuation: ______

Thank you very much for your participation!

Research Team 96

Ref. code: 25605522300184DMX