Abstract

Compliance, Congestion, and Social Equity:Tackling Critical Evacuation Challenges through the Sharing Economy, Joint Choice Modeling, and Regret Minimization

By

Stephen David Wong

Doctor of Philosophy in Civil and Environmental Engineering

Emphasis in Transportation Engineering

University of California, Berkeley

Professor Susan Shaheen, Co-Chair

Professor Joan Walker, Co-Chair

Evacuations are a primary transportation strategy to protect populations from natural and human-made disasters. Recent evacuations, particularly from hurricanes and wildfires, have exposed three critical evacuation challenges: 1) persistent evacuation non-compliance to mandatory evacuation orders; 2) poor transportation response, leading to heavy congestion, slow evacuation clearance times, and high evacuee risk; and 3) minimal attention in ensuring all populations, especially those most vulnerable, have transportation and shelter. With ongoing climate change and increasing land development and population growth in high-risk areas, these evacuation challenges will only grow in size, frequency, and complexity, further straining transportation response in disaster situations.

Research Objectives and Theoretical and Methodological Contributions: To tackle these three challenges and improve evacuation outcomes, I explored three research areas: the sharing economy, (joint) choice modeling, and regret minimization. 1) Sharing Economy: The sharing economy has grown rapidly in the past two decades, opening new mechanisms to share, sell, and buy goods and services via technology. Similar to other economic forms, the sharing economy must contend with and respond to external shocks, including disasters. Within this response, an opportunity arises: the sharing economy through private companies or residents could theoretically be a mechanism to increase available assets in evacuations and disasters. Due to the recent development of the sharing economy, research has yet to explore and assess this strategy fully. With limited evacuation literature in this area, an initial question arises: To date, what has been the role of the sharing economy in disasters? In addition, what are the benefits and limitations, particularly for vulnerable groups? On the private resident side, are people willing to share mobility and sheltering resources, and what influences this willingness? To address these questions and explore this new strategy, I tested the feasibility of the sharing economy by assessing the: • Current state of the sharing economy in evacuations, benefits and limitations of the sharing economy in disasters, and the willingness of individuals to provide shared resources through archival research, expert interviews, and post-disaster surveys; • Effect of different factors, including trust and compassion, on willingness to share transportation and sheltering through simple discrete choice models; • Extent to which sharing economy platforms and shared resources can benefit or limit social equity for vulnerable populations through focus groups and application of the STEPS (spatial, temporal, economic, physiological, social) equity framework; and • Behavioral nuances of different models – binary logit models, multi-choice latent class choice model, and portfolio choice model – for the willingness of individuals to share resources in multiple evacuation scenarios for transportation and sheltering.

2) (Joint) Choice Modeling: Disasters are stressful and complex events in which individuals must make rare choices related to evacuations and their safety. First, individuals must decide to evacuate or stay, after which evacuees must navigate through multiple complicated choices including departure day, departure time of day, transportation mode, destination, shelter type, route, and reentry time. Current evacuation behavior literature, while reflecting significant strides in recent years, contains several severe gaps. Much literature is focused on whether to evacuate or stay, with limited research on the complex decisions that must follow this initial choice. In addition, research has only minimally explored the different behavioral responses of unobserved classes of people or the influence of attributes of alternatives on choice. Choice modeling has also focused primarily on hurricanes, leaving a wide gap in the evacuation literature on wildfire behavior. What influences choice making in evacuations, particularly choices beyond the decision to evacuate or stay and especially for wildfire evacuations? Do attributes of alternatives or unobserved classes add behavioral understanding? Most importantly, literature has not considered the theoretical possibility that evacuation choices are inherently joint and multi-dimensional. What choices are correlated and dimensionally dependent, and how should this be modeled? I addressed these research gaps by applying a series of discrete choice models that conduct:• An attribute-based assessment of wildfire evacuation choices beyond the decision to evacuate or stay through simple multinomial logit models; • A latent classification of individuals for the decision to evacuate or stay via a latent class choice model for hurricanes; and • An assessment of decision-dimensional dependency of hurricane choices and wildfire choices (departure day, departure time of day, destination, shelter type, transportation mode, and route) using a portfolio choice model.

3) Regret Minimization: Due to the risky and rare context of evacuations, people likely make decisions differently than under normal circumstances. Regret has been found to influence choices that are difficult and when individuals receive rapid feedback on whether their choices had positive or negative outcomes. Given the unique characteristics of disasters and evacuations, regret minimization (i.e., choice making by minimizing future anticipated regret) could theoretically present a more valid decision rule in evacuations than utility maximization, which has been assumed for most evacuation choice models. Literature in this area is limited, with few studies testing regret minimization in evacuations and only in a stated preference setting. Does random regret minimization (RRM) better describe evacuation behavior than traditional random utility maximization (RUM) in choice models? With no empirical testing of this theory in the literature using post-disaster data, what methodology should be used in a revealed preference setting to reconstruct complex evacuation choice sets and test regret minimization? To answer these research questions and test the theory of regret in evacuations, I analyzed:• Regret minimizing behavior of wildfire evacuees by developing a revealed preference (RP) methodology for challenging choice sets.

Empirical Contributions: One primary challenge in the evacuation field is the collection of post-disaster data, which can be difficult for a variety of reasons related to finding participants, securing funding, not interfering with recovery efforts, and deploying data-gathering instruments quickly. Finding enough participants for data collection is especially difficult for wildfire evacuations (compared to hurricane evacuations), due to their smaller size. To meet these challenges and contribute data to the broader evacuation field, I distributed online surveys, collecting responses from individuals impacted by three disasters:• 2017 Hurricane Irma in Florida: n=645 (collected Oct. - Dec. 2017); • 2017 December Southern California Wildfires: n=226 (collected Apr. - June 2019); and • 2018 Carr Wildfire: n= 284 (collected Feb. - Apr. 2019). One critical limitation of online (and disaster) surveys is the failure to represent vulnerable populations. Consequently, I supplemented the wildfire surveys with a series of four focus groups composed of individuals from four vulnerable groups – low-income individuals, older adult, individuals with disabilities, Spanish-speaking individuals – each impacted by a California wildfire between 2017 and 2018 (collected Aug. 2018 - Apr. 2019). To establish a foundation for my research on the sharing economy, I also interviewed 24 high-ranking experts on the benefits and limitations of this strategy in disasters (collected Feb. 2017 - Apr. 2017).

Sharing Economy Results: I find several key limitations of the sharing economy for both private companies and private citizens in hurricanes and wildfires including concerns related to safety, social equity, communication, and driver reliability (Chap. 3, Chap. 5). Yet, the sharing economy could provide benefits including augmenting resources, quickening transportation responsiveness, and improving compliance with evacuation orders Chap. 3). Results indicate that sharing economy companies (i.e., Airbnb, Lyft, Uber) have been acting in disasters since 2012, and their actions have become more consistent and structured in since 2016 (Chap. 3). Private citizens are moderately willing to share shelter and transportation in hurricanes and wildfires (Chap. 3, Chap. 4). The percentage of survey respondents extremely willing to share transportation before evacuating was 29% for hurricanes and 37% to 48% for wildfires. For transportation during an evacuation, 24% were extremely willing to share for hurricanes and 59% to 72% for wildfires. Individuals were more willing to share housing for free than for a cost (Chap. 3., Chap. 4). About 19% were extremely willing to share housing for free for hurricanes, with 24% to 30% for wildfires. I also find spare capacity in terms of beds/mattresses (ranging from 84% to 90%) exists widely (Chap. 3, Chap. 4). Approximately 77% of evacuating vehicles from Hurricane Irma had at least two empty seats with a seatbelt (Chap. 3), and 64% to 69% of evacuating vehicles from the California wildfires had at least two empty seats with seatbelts (Chap. 4).

Regarding social equity, I find that while the sharing economy would be a feasible strategy for some vulnerable groups (e.g., carless, asset poor, older adults, people of color, immigrants), many vulnerable groups would experience significant barriers (e.g., digital divide; communication issues; liability for providers; high expense; locating evacuees; citizenship status) to accessing and using shared resources (e.g., physically disabled, unbanked, non-English speaking, homeless, undocumented immigrants) (Chap. 5). I also find that high levels of trust and compassion, as well as a sense of urgency, are associated with increased willingness to share resources, suggesting that some limitations related to the sharing economy could be overcome (Chap. 4). While past volunteers and community organization members in the surveys were more willing to share, other demographic variables (e.g., age, gender, income, race/ethnicity) had weak effects on willingness, indicating the primacy of trust and compassion in sharing behavior. Assuming a high trust/compassion population versus a low trust/compassion population results in a change of likelihood to share between 30% to 55%, depending on the sharing scenario (Chap. 4). Finally, I find substantial joint preferences between different evacuation sharing scenarios through a portfolio choice model and three unique classes (adverse sharers, interested sharers, and transportation-only sharers) with different sharing preferences through a multi-choice latent class choice model (Chap. 6). I find that families are unlikely to share regardless of model type and spare capacity has a weak positive influence on willingness to share. Demographic variables had sporadic effects depending on the chosen model, suggesting that the selection of discrete choice model can heavily influence results (Chap. 6).

(Joint) Choice Modeling Results: Through the development of portfolio choice models for hurricane and wildfire evacuations, I find that evacuation choices should be modeled jointly to account for correlation among choices and develop more nuanced transportation strategies for evacuations (Chap. 7, Chap. 8). For hurricanes (Chap. 7), joint preferences were especially strong between departure day and other choices (i.e., departure time of day, route) and between destination and other choices (i.e., transportation mode, route, shelter type). For wildfires (Chap. 8), strong joint preferences were found for departure day and other choices (i.e., departure time of day, destination, shelter type, transportation mode, route) and destination and other choices (i.e., departure time of day, shelter type, route). However, joint preferences are not always the same between the two wildfire cases (2017 December Southern California Wildfires and 2018 Carr Wildfire), suggesting that joint choice making is contextually, geographically, and/or culturally dependent.

I also find, via a latent class choice model, two classes of individuals for the decision to evacuate or stay in a hurricane (Chap. 7). A class of “keen evacuees” – composed of families, individuals living near the hurricane landfall area, and those with risk perceptions who were more likely to evacuate but could not be influenced by mandatory evacuation orders. A class of “reluctant evacuees” – comprised of previous evacuees, long-time residents, and those with concerns over evacuation logistic barriers – was much less likely to evacuate but could be influenced to leave through mandatory evacuation orders (Chap. 7). The decision to evacuate or stay/defend in a wildfire is influenced by mandatory evacuation orders and risk perceptions but with uneven influence of household and individual characteristics (Chap 8.). Finally, I developed a series of wildfire models, finding that attributes of departure times (e.g., immediate fire danger, pressure from neighbors to leave, uncertainty of escape route, visibility, visual fire level) and routes (e.g., distance, fire danger along route) influence choice making. (Chap. 9). However, attributes related to shelter type, transportation mode, and reentry timing were not influential, suggesting that demographics, risk perceptions, and/or resource availability may better explain those choices.

Regret Minimization Results: Finally, through a series of random utility maximization (RUM) and random regret minimization (RRM) models for wildfires (Chap. 9), I find regret minimizing behavior to be relatively weak for all considered choices (i.e., departure timing, route, transportation mode, shelter type, and reentry timing). Given my findings of weak attribute-level regret for departure timing, route, and reentry timing as well as weak class-oriented regret for route and transportation mode, I conclude that regret minimization does not explain behavior in evacuations better than utility maximization. However, results indicate that the survey construction and methodology could be significantly improved to better test the presence of regret minimizing behavior, and regret minimization should continue to be explored in evacuee choice making.

Policy Recommendations: Employing these results, I provide a series of recommendations to local and regional agencies to improve compliance, reduce congestion, and increase social equity. For example, a sharing economy strategy, regardless of hazard (based on Chap. 3 to 6), should: 1) develop low-tech communication and matching methods; 2) leverage neighborhood networks and community-based organizations to distribute resources to vulnerable groups; 3) connect with public transit plans; 4) incorporate significant input from vulnerable populations; and 5) combine both transportation and sheltering resources across all temporal points of the disaster. Based on hurricane choice modeling results (Chap. 7), agencies should be prepared to deploy transportation resources, responses, and services significantly before landfall, at night, and along highways. Agencies should also target mandatory evacuation orders in certain neighborhoods (previously evacuated zones, long-time residents) and leverage orders as an instrument to reduce concerns over evacuation logistic barriers to increase compliance. Agencies are recommended to also target mandatory evacuation orders for wildfires (Chap. 8), but orders need to be distributed more rapidly and through low-tech communication methods. Results also suggest that agencies should be prepared to rapidly deploy transportation responses at night, close to the evacuation zones (i.e., highly localized), and along arterial and local streets (Chap. 8, Chap. 9) Finally, agencies in wildfires should encourage people to leave before they can visually see the fire, increase evacuation information at the neighborhood level, and provide clear routing information (Chap. 9).

Summary: In this dissertation, I present several new pathways and research areas to better tackle three critical evacuation challenges related to compliance, congestion, and social equity. Through theoretical, methodological, and empirical contributions, I reinforce well-known and offer new evacuation strategies that can be implemented by governments faced with the complicated task of moving thousands and even millions of people. Ultimately, the research presented in this dissertation offers an academic building block and launching point for future work in the evacuation field, while also remaining grounded in the need for stronger practical applications of research to improve evacuation plans, strategies, and policies.