Perspectives on the Ridesourcing Revolution: Surveying Individual Attitudes Toward Uber and Lyft to Inform Urban Transportation Policymaking

Home , Turo (company), VentureBeat

Perspectives on the Ridesourcing Revolution: Surveying individual attitudes toward Uber and Lyft to inform urban transportation policymaking

Margo Dawes

Bachelor of Science in Civil Engineering and Planning Massachusetts Institute of Technology Cambridge, MA (2015)

Submitted to the Department of Urban Studies and Planning in partial fulfillment of the requirements for the degree of

Master in City Planning

at the

MASSACHUSETTS INSTITUTE OF TECHNOLOGY

June 2016

The author hereby grants to MIT the permission to reproduce and to distribute publicly paper and electronic copies of the thesis document in whole or in part in any medium now known or hereafter created.

Author______Department of Urban Studies and Planning

Certified by______Assistant Professor Jinhua Zhao, Ph.D. Department of Urban Studies and Planning Thesis Supervisor

Accepted by______Associate Professor P. Christopher Zegras Chair, MCP Committee Department of Urban Studies and Planning

Perspectives on the Ridesourcing Revolution: Surveying individual attitudes toward Uber and Lyft to inform urban transportation policymaking By Margo Dawes Submitted to the Department of Urban Studies and Planning on May 19, 2016, in partial fulfillment of the requirements for the degree of Master in City Planning

Abstract Media coverage of ridesourcing services such as Uber and Lyft has described a rivalry between new technology and the established taxi industry. Individual users and non-users of ridesourcing, however, may have more nuanced perspectives, but policymakers have little guidance on how to best represent these interests. This thesis uses a standardized questionnaire distributed across the United States by an online survey company to understand individual attitudes toward Uber, Lyft, and ridesourcing technology as a whole. It asks respondents if they identify as users or non-users of ridesourcing, why or why not, how they rank Uber and Lyft among their other travel modes, their attitudes toward the companies and toward the technology in general (on a Likert scale of 1, very negative, to 5, very positive), and their opinions on how their cities should respond, among other questions. 394 completed questionnaires from the most populous 15 metropolitan statistical areas in the U.S. reveal individuals’ use of and attitude toward ridesourcing technology along with variations across demographic groups, cities and regions, and primary travel mode. The survey returned a response rate of 27% and the spatial distribution of responses was roughly proportional to the population of each metropolitan area. The findings indicate that about 70% of respondents use some form of ridesourcing, mostly for special-purpose trips such as avoiding driving while intoxicated and getting to and from the airport. The vocal minority who don’t use Uber or Lyft for ethical reasons represent only a small subset of the sample (about 6%), but 1 in 5 respondents said the companies’ decision to treat drivers as independent contractors rather than employees made them want to use the services less. There are relationships between transportation characteristics (e.g. usage of Uber and Lyft, needing to travel for work, and having access to a car) and user identification and attitude, but demographics are the best predictor of user identification, which in turn predicts attitude, which predicts policy implications. The study suggests potential for policymakers to leverage constituent perspectives to change aspects of ridesourcing that have low public approval. Thesis supervisor: Jinhua Zhao Title: Edward H. and Joyce Linde Assistant Professor of Urban Planning

Acknowledgments

My first and biggest thanks must go to my advisor, Jinhua Zhao, whose enduring enthusiasm and expert guidance helped bring this thesis to life. Jinhua’s rigorous standards and constant encouragement made this truly a learning experience, and for that I am deeply grateful. Thank you also to my readers, Onesimo Flores and Craig Kelley, who provided the invaluable, if distinct, feedback that I used to round out the beginning and end of my thesis. Thank you to everyone at JTL who had a hand in the idea formation and evaluation along the way. Hongmou, Adam, Corinna, and Jeff all provided meaningful contributions, and Nick Allen’s outsize interest and availability never went unappreciated. A big thank you also to Mira Vale and Cali Warner, who provided the technical assistance I could not ask of anyone else. I would also like to acknowledge the project managers at Qualtrics, Nate Richard and Emily Davis, who helped me prepare my survey for distribution.

To my dad, Roy Dawes, who shared his expertise and aided in the transformation of my statistical analysis. And to my mom, Diane Iñiguez, for the reminder that the world still exists outside of thesis. To Carmen Castaños and Eric Benzschawel for the solidarity, and to Lindiwe Rennert for the same and for being my personal cheerleader. To Marisa Fryer for the incredulity. A heartfelt thank you to Anna Doty, for all of the love and support this year. Finally, a shout out to all of my MCP colleagues who brightened the thesis room and made my experience at DUSP what it was.

Contents

List of Figures ...... 7 List of Tables ...... 8 Terms and Definitions ...... 9 1. Background ...... 10 1.1 The Ridesourcing Phenomenon and its Roots in the So-Called Sharing Economy ...... 10 1.2 Ridesourcing Services and Regulatory Response ...... 13 1.3 Study Motivation and Research Questions ...... 16 2. Literature Review ...... 18 2.1 Regulatory History of Taxi Industry ...... 18 2.2 The Ridesourcing Debate ...... 19 3. Data and Methodology ...... 21 3.1 Research Methodology and Data Acquisition ...... 21 3.2 Analysis Methodology ...... 34 4. Findings and Analysis ...... 35 4.1 Survey Responses and Univariate Analysis ...... 35 4.2 Bivariate and Multivariate Analysis ...... 48 5. Discussion ...... 58 5.1 Key Findings ...... 58 5.2 Discussion of Research Questions ...... 59 5.3 Policymaker Impressions ...... 60 5.4 Limitations and Areas for Further Research ...... 62 References ...... 64 Appendices ...... 70 Appendix 1: Survey questionnaire and recode values ...... 70 Appendix 2: Correlations and regression model outputs ...... 77 Appendix 3: COUHES approval ...... 84

List of Figures

Figure 2. Question structure shown to survey respondents for the question “The following aspects of ride-hailing technology make me want to use Uber or Lyft [More, Less, or Neutral].” ...... 33

Figure 3. Comparison of preferences between Uber and Lyft based on seven criteria...... 42

Figure 4: Mode choice heat map showing how respondents ranked each mode...... 43

Figure 5. Favorability of certain attributes of the ridesourcing services, Uber and Lyft...... 47

List of Tables

Table 1. Summary of new and traditional mobility services that serve similar functions, thereby often acting as competitors. ……………………………………………………………………..10 Table 2. Targets used for recruitment. …………………………………………………………..22 Table 3. Potential sample sizes. …………………………………………………………………22 Table 4. Summary of survey questionnaire organized by section. ……………………………...28 Table 5. Demographics of survey respondents compared to adult population of the 15 aggregated metropolitan areas. ………………………………………………………………………………35 Table 6. U.S. Census region and metropolitan area of residence of respondents. ………………37 Table 7. Primary trip purpose when using Uber or Lyft. ………………………………………..38 Table 8. Reasons for and against choosing ridesourcing as opposed to other modes. ………….39 Table 9. Frequency of use of Uber and Lyft. ……………………………………………………41 Table 10. Preferences between Uber and Lyft. ………………………………………………….42 Table 11. Attitudes toward Uber, Lyft, and ridesourcing technology in general, irrespective of the two companies. ………………………………………………………………………………45 Table 12. Responses to the question “What should your city do about [Uber/Lyft]?” …………46 Table 13. Favorability of certain attributes of the ridesourcing services, Uber and Lyft. ………48 Table 14. Summary of independent and dependent variables. ………………………………….48 Table 15. Relationships between independent variables and dependent variable: user identification. ……………………………………………………………………………………50 Table 16. Relationships between independent variables and dependent variable: attitude toward Uber (Lyft). ……………………………………………………………………………………...51 Table 17. Relationships between independent variables and dependent variable: policy implications for Uber (Lyft). …………………………………………………………………….52 Table 18. Summary of logistic regression analysis of determinants of user identification. …….54 Table 19. Summary of ordered logit analysis of determinants of attitude toward ridesourcing services. ………………………………………………………………………………………….55 Table 20. Summary of multinomial logit analysis of determinants of policy implications for Uber (each is ‘Form Partnerships’ vs. ‘Regulate’). …………………………………………………...57

Terms and Definitions

There are a number of terms in use to describe Uber, Lyft, and similar app-based transportation services. The definitions below are meant to be a resource for the reader and represent a loose consensus among academics, regulators, and commentators. Transportation Network Company, or TNC, is a technical term used in a number of written regulations, while ‘ride-hailing’ and ‘ridesourcing’ are more colloquial terms, and ‘ride-sharing’ refers to another kind of service altogether. Following the lead of the AP Stylebook, the survey questions used in this study referred to Uber and Lyft as ‘ride-hailing’ services; however, this thesis will primarily use the term ‘ridesourcing’ as it is the most appropriate and interpretable term to describe the business model.

Transportation Network Companies (TNCs): Companies that use an online-enabled or mobile platform, commonly a smartphone app, to connect passengers with drivers who use their own personal vehicles. The California Public Utilities Commission (CPUC) coined the term in a 2013 rulemaking process to adopt new regulations for app-based ridesourcing services (California Public Utilities Commission, 2013). Ridesourcing: Ridesourcing services let people use smartphone apps to book and pay for car services (AP Stylebook, 2015). TNCs offer ridesourcing services and the terms may be used interchangeably. The Associated Press and transportation experts have promoted the use of terms such as ‘ride-hailing,’ ‘ride-sourcing,’ and ‘ride-booking’ in place of ‘ride-sharing’ to indicate that services like Uber and Lyft do not directly provide the vehicles or drivers, nor are the drivers sharing rides with the passengers. Ride-sharing: Ride-sharing as a practice is driven by a social mission to match drivers and passengers for the sake of increasing mobility and reducing costs and environmental impact. It often refers to carpooling and vanpooling, and implies that the driver and the passenger(s) share destinations, rather than the driver providing a service explicitly for the passenger as in ridesourcing.

1. Background

1.1 The Ridesourcing Phenomenon and its Roots in the So-Called Sharing Economy

Recent years have ushered in innovation in the mobility industry, both in technological advancements and with new business models. A bevy of transportation services from private cars to commuter buses have become available on-demand thanks to the expansion of smartphone apps and online-enabled platforms. Two frameworks provide context for this emerging shift: the urban mobility revolution, and the growing sharing economy. The urban mobility revolution consists of the confluence of the transportation component of the smart cities movement, innovation in autonomous vehicles, vehicle electrification and connectivity with other vehicles, and the advent of new mobility services (Bouton, Knupfer, Mihov, and Swartz, 2015; Quintal, 2015). These mobility services might more accurately be called new models for accessing traditional forms of mobility, and range from car-sharing to ridesourcing to a combination of carpooling and ridesourcing. The convenience (e.g. services available on-demand through smartphone apps) and cost-savings (e.g. not needing to rely on a private car for most trips) associated with the technology component of these alternatives has made many of them competitors to traditional modes and providers (see Table 1).

Table 1 Summary of new and traditional mobility services that serve similar functions, thereby often acting as competitors Traditional mobility New mobility services Leading examples solutions Individual- Private car Car sharing: A peer-to-peer platform Turo (formerly based ownership peer to peer where individuals can rent RelayRides) mobility out their private vehicles when they are not in use. Rental cars Car sharing: On-demand short-term car Zipcar fleet operator rentals with the vehicle owned and managed by a fleet operator. Taxi Ridesourcing Process of ordering a car or UberX, Lyft taxi via on-demand app, which matches rider with driver and handles payment automatically. Group- Car pooling Shared Allows riders going in the UberPool, Lyft based ridesourcing same direction to share the Line mobility car, thereby splitting the fare and lowering the cost.

Public transit On-demand App- and technology- Bridj private shuttles enabled shuttle service. Cheaper than a taxi, and more convenient than public transit. Private buses Shared and Wi-Fi-enabled Google buses commuter buses available to the public or to employees of select companies. Used to free riders from driving to work. Note. Reproduced from McKinsey & Company, with own addition of last column, “Leading examples” (Bouton, Knupfer, Mihov, and Swartz, 2015).

Occurring alongside this revolution—and encompassing many of the innovative models for accessing mobility described above—is a phenomenon known as the sharing economy, which refers to the market for sharing, trading, or renting goods and services from person to person rather than through traditional ownership or centralized institutions. Some argue that this framing describes only the consumption portion of a larger space that also includes collaborative production, finance, and learning, but the fundamental notion is that online platforms can help link excess supply to demand (Botsman, 2013). Put differently, as an alternative to traditional forms of ownership, consumers or firms can grant temporary access to under-utilized or idle assets (Meelen and Frenken, 2015). Popular examples include Airbnb, through which homeowners can rent out their rooms or homes when they would otherwise be unoccupied, and Turo (formerly RelayRides), which allows people to rent their cars when not in use. Some policymakers and commentators, however, are skeptical of users of Airbnb and similar services acquiring real estate for housing or office space with the express intent to advertise through online sharing platforms, which does not adhere to the spirit of sharing excess assets.1 Ridesourcing apps such as Uber and Lyft are often lumped into the transportation and mobility sector of the sharing economy based on their similar characteristics: they circumvent centralized institutions (i.e. the taxi industry, by using regular unlicensed drivers), and provide a supplemental source of income to their supplier bases (i.e. people who drive for Uber and Lyft). However, some argue that Uber, while innovative in its booking, payment, and ratings functionalities, should not be included in the purview of the sharing economy as it is providing a paid service with a contracted workforce, not helping people utilize idle capacity (Meelen and Frenken, 2015). Similar to the Airbnb example above, a driver who drives full-time for Uber or Lyft is not sharing her car, but running a service. The exceptions in this argument are UberPool and Lyft Line, carpooling extensions of their parent companies that facilitate ride-sharing among users of the apps. Arguments about which companies or services should or shouldn’t be included

1 See, e.g., Steinmetz (2014).

11 in discussions of the sharing economy legitimize a recent claim by the Harvard Business Review that the sharing economy really ought to be called an access economy:

Sharing is a form of social exchange that takes place among people known to each other, without any profit. Sharing is an established practice, and dominates particular aspects of our life, such as within the family. By sharing and collectively consuming the household space of the home, family members establish a communal identity. When “sharing” is market-mediated — when a company is an intermediary between consumers who don’t know each other — it is no longer sharing at all. Rather, consumers are paying to access someone else’s goods or services for a particular period of time. It is an economic exchange, and consumers are after utilitarian, rather than social, value. (Eckhardt and Bardhi, 2015) Reframing the discussion of the ‘sharing economy’ to one about an access economy or platform economy is a more honest portrayal that allows for meaningful critical thinking on the potential costs as well as the benefits (Kenney and Zysman, 2015; Lobel, 2015). The debate about the role of ridesourcing services in such a space has introduced some key concerns, such as the hazy distinction between a “sharing economy” and an informal economy that lacks the regulated benefits and protections of traditional employment (Giridharadas, 2012; Shapiro, 2012). In such a labor market, a person might peddle their services full-time through online platforms such as TaskRabbit or Skillshare without ever earning the safety nets of health care and sick leave afforded to full-time workers in the formal economy. Another admonition is the potential Jevons paradox of the ridesourcing model, wherein increasing the efficiency of hailing a ride may increase rather than decrease car usage and overall vehicle miles traveled (VMT), with the associated implication that a company that rents its physical assets from contractors is unaccountable for an environmental and social footprint it cannot technically leave (Correll, 2015). These concerns about current and forthcoming social costs associated with ridesourcing are joined by others, including that Uber and Lyft may compete unfairly with the taxi industry, may have monopolistic aspirations, may enable discrimination, and may foster an abuse of labor (Rogers, 2015). Still, ridesourcing may offer a more efficient service than taxis, may reduce drunk driving and other accidents, may fill gaps in public transit networks, may reduce discrimination toward potential passengers, and may diminish incentives for purchasing private vehicles, all of which represent potential benefits to consumer welfare that warrant experimental regulations (Posen, 2015; Rayle et al., 2016; Rogers, 2015). The arguments emerging in this debate are still largely untested, but the topic is explored in greater detail in the Literature Review. At the nexus of the conversation about the sharing economy and the revolution in urban mobility is this heated controversy around ridesourcing, with Uber at its center. While these contextual frameworks will not directly shape the analysis in this thesis, both were integral in motivating the topic of study. Uber is not the sole app-based ridesourcing service in operation in the United States—its primary competitor is Lyft—but it is by far the market leader and the more

12 controversial of the two.2 With nearly seven times the net revenue of Lyft, twelve times the number of average monthly rides, six to eight times the funding and fifteen to twenty-five times the market valuation, Uber dominates both the market for on-demand private transportation and cultural consciousness about the transportation sharing economy (Miller, 2015; Picker and Isaac, 2015). Public input on Uber and Lyft’s operation ranges from enthusiastic support (which is often manifested as rejection of the established taxi industry) to emphatic opposition. This opposition is often directed at Uber in particular, likely due to a combination of its market dominance and bold push for deregulation. This thesis aims to tap into those opinions and attitudes to reveal societal perceptions about ridesourcing apps in general and their role in the greater market of urban transportation.

1.2 Ridesourcing Services and Regulatory Response

As a business model, ridesourcing allows individuals to request a ride through a smartphone application and get connected to a nearby driver within minutes. Once the driver receives notice of the request, the passenger can observe the vehicle’s real-time progress as they await pick-up. The driver follows directions to the passenger’s destination through GPS-enabled navigation within the app; upon completion, the cost of the trip is automatically charged to the passenger’s credit card, and the driver and passenger rate one another on a scale of 1 to 5 stars. Fares change according to real-time shifts in demand, creating an incentive for more drivers to be available during periods of high demand. This business model relies largely on drivers who don’t have commercial vehicle licenses and who use their own vehicles to provide service on a part-time basis (Rayle et al., 2016). These aspects—on-demand booking, automatic payment, a driver-passenger rating system, and dynamic fares—distinguish ridesourcing from traditional taxi or livery services, which are otherwise the closest (and best-researched) analogous mode. When using taxis, rides must be hailed off the street or arranged in advance through a telephone dispatcher. Taxis also require payment in cash or by credit card upon completion of the trip; fares are regulated by the city; drivers either use their own navigation system or rely on their knowledge of the street grid; and passengers do not know before a trip begins whether they’re riding with a “good” or “bad” driver, and vice versa. Much has been written about the role of taxis in urban transportation, and literature is emerging on the benefits and costs of ridesourcing services in comparison (see “The Ridesourcing Debate” in the Literature Review), but Rayle et al. (2016) summarize the new business model succinctly: “What is new about recent ridesourcing by Uber, Lyft, and others is the combination of a model that leverages GPS-enabled smartphone technology and exemption from traditional taxi regulations, which allows more flexibility in supply and service

2 Described in the following section, but not included in this analysis, are ridesourcing precedents such as ridesharing apps (e.g. Carma, SideCar), which connect carpoolers, and taxi apps (e.g. Flywheel), which offer similar services to Uber and Lyft but through licensed taxi companies and drivers, as well as successful alternatives to Uber that operate exclusively outside of the United States.

13 characteristics.” Combined with the success of this model, the exemption from regulation created a controversy with the taxi industry that has impacted public policy into the present day. This section introduces the US-based ridesourcing companies that are the focus of this thesis, Uber and Lyft, as well as precedents and non-domestic competitors that are left out of the analysis. It also details the variety of regulatory responses to Uber, Lyft, and early TNC competitors enacted by U.S. cities in recent years. 1.2.1 Uber and Lyft Uber was founded in March 2009, and began offering services in San Francisco between the summer of 2010 and early 2011. In the past six years it has expanded to offer service in over 400 cities globally, working with hundreds of thousands of drivers to provide rides for millions of passengers. Uber is widely considered to be the market leader in the U.S., but faces competition from Lyft domestically, as well as other transportation network companies in India, China, Southeast Asia, and Latin America (see “Precedents and alternatives” below). At its inception, Uber operated an on-demand black car service. Facing competition from Lyft and Sidecar—founded in 2012 and 2011, respectively—Uber expanded its model to include comparable ridesourcing service (Flores and Rayle, 2016). Lyft has a much smaller valuation than Uber—$5.5 billion compared to Uber’s $62.5 billion (Lien, 2016)—and operates only within the U.S., but domestically it is Uber’s most serious competitor. While Lyft is less widely used than Uber, it also presents itself as a friendlier alternative (Huston, 2016; and Shahani, 2016). Both companies were deemed Transportation Network Companies (TNCs) by a 2013 California Public Utilities Commission ruling, and cities nationwide followed suit. Amid claims that the companies were operating illegal taxis (i.e. lacked licenses and medallions), this designation held Uber and Lyft to minimum consumer protection requirements without imposing the same operational restrictions as those applied to taxis. Flores and Rayle (2016) argue that this mainstreaming resulted from a combination of the companies’ lobbying and organizing efforts and the favorable political climate of their home base, San Francisco. 1.2.2 Precedents and alternatives Uber and Lyft emerged dominant from a market that at one point held numerous competitors. Developing from the climate in which car-sharing services like Zipcar and car2go found success, numerous services such as Sidecar, Zimride (which evolved to become Lyft), Gett, Summon, Wingz, Carma, TaxiMagic, and Cabulous began offering e-hailing and ride-sharing services a little before or around the same time as Uber and Lyft. Most of these precedents have not survived to the present day, at least not in their original incarnation. Sidecar changed focus to become a parcel delivery service before selling to General Motors in December 2015; Wingz became an airport shuttle service; and Carma branched out to offer not only car-sharing, but carpooling and luxury shared commuting as well.

TaxiMagic and Cabulous—rebranded as Curb and Flywheel—are among the many taxi apps (including Way2Ride, mytaxi, and Arro) used by taxi industries in larger U.S. cities that employ similar matching technology to connect riders to licensed taxi drivers. These services are attempts by taxi companies and their supporters to capitalize on and compete with the convenience offered by the Uber and Lyft apps, and represent an “if you can’t beat them join them” attitude toward the competition posed by these industry disruptors. While less controversial than Uber or Lyft, these services have also been less successful, likely due to the decentralized, often city-specific nature of the apps, the associated lack of comparable brand recognition, and jurisdictional limitations on pick-up. Outside of the U.S., differently structured taxi or jitney industries and diverse regulatory climates have allowed for the emergence of regional competitors such as Didi Chuxing, which operates in China; Ola, in India; Grab, which operates in six countries across Southeast Asia; and Cabify, a new entry spreading across Spain and Latin America. These services are primarily equivalent to the taxi apps described above (though Didi Chuxing also uses ‘designated drivers’), and each is Uber’s primary competitor in its respective region. In a collaborative move against Uber’s market dominance, four of Uber’s major regionally-specific competitors—Lyft, Didi Chuxing, Ola, and Grab—announced in late 2015 that they would form an alliance to share riders across continents (Buhr, 2015; Punit, 2015). This alliance will allow users to book cabs from the local app when traveling; it went online in April 2016 (Carson). 1.2.3 Regulatory responses

As Uber and Lyft expanded across the U.S., regulators and the public responded to the competition they posed to the local established taxi industries. In the early years of their expansion, a number of cities enacted bans on Uber and Lyft’s operation while awaiting a long- term regulatory agreement.3 Other jurisdictions made regulatory attempts to constrain operations and limit growth.4 Most bans and strict regulations were overturned within a year as jurisdictions granted temporary allowances and updated their vehicle-for-hire regulations. Following the 2013 CPUC regulation that legalized ridesourcing operations as transportation network companies, dozens of cities and states have followed suit, making Uber and Lyft effectively legitimate, if still controversial. The spectrum of regulatory response spans from negative action (e.g. bans and severe limitations) to inaction to affirmative action (e.g. regulations that protect consumers but otherwise legitimize operations, and collaborative partnerships), and many cities have traveled

3 See, e.g., Los Angeles Department of Transportation (Novak, 2013); San Francisco Municipal Transportation Agency (Graves, 2010); Ann Arbor City Council (Reindl, 2014); South Carolina Public Services Commission (Waller, 2015); and Broward County, Fla. (Wallman, 2014). 4 The Seattle City Council, for example, voted in 2014 to limit the number of Uber and Lyft drivers on the road at any given time to 150 per company (Lawler, 2014), a move New York City almost repeated a year later before coming to an agreement with Uber (Flegenheimer, 2015). Also in 2014, the Virginia Department of Transportation levied a civil penalty against Uber and Lyft (Hazard, 2014), and the State of Illinois proposed a bill to strictly regulate commercial ridesharing services (HB 4075, 2014).

15 along the spectrum in the process of addressing this mobility phenomenon, rather than demonstrating a consistent and coherent strategy. Though regulatory approaches are converging, some cities are still waiting for guidance from the state level, and policymakers and academics cite lingering questions about how to regulate potential social and environmental costs (Rogers, 2015; A. Jette, interview, May 6, 2016). The disparate responses and changes in position demonstrate a fluidity still present in the discussion of this policy issue. While cities today can look to regulations enacted elsewhere in the U.S. to discern trends, decisions are largely made without a robust understanding of public opinion and with overrepresentation in the debate from the taxi industry. This vacuum of collective understanding among policymakers nationwide about how to move forward in the regulation of the access economy is a critical motivation for this thesis.

1.3 Study Motivation and Research Questions

1.3.1 Study motivation As described above, regulators’ responses to ridesourcing companies have been primarily without guidance. Popular media coverage portrays the question of how to proceed as one rooted in a controversy with the established taxi industry, and has only recently begun to shift toward questioning social and environmental impacts. While a popular focus of emerging literature, the topic of ridesourcing has not yet been fully debated or analyzed in peer-reviewed publications. The novelty of these companies and their proclivity for major operational changes have also thus far precluded any long-range studies of their societal impacts. Policymakers are left to shape regulation guided primarily by what they read in the news and their own personal experiences, leading to policy that responds more to emergencies or specific events than to long-range planning, and to an overall lack of supportive evidence. The gap in the literature and in the understanding of constituent perspectives motivated the design of a widely-distributed survey meant to capture information about public use of and attitude toward ridesourcing. I chose to use a survey instead of interviews as my primary methodology in an effort to quantify public perceptions and capture input from more than the most vocal or powerful stakeholders; this distinction is described in greater detail in the Methodology section. 1.3.2 Research questions

This study—a nationally-distributed survey and an analysis of the responses—serves to address the fundamental question of how individuals perceive of the ridesourcing revolution and how regulators can make policy with better information than currently available to them. This primary research question drives the exploration of the following topics: 1) When and why do people use Uber and Lyft? What makes an individual choose one over the other if the business models are so similar? 2) How do individuals perceive Uber compared to Lyft, and compared to ridesourcing technology in general? How can these attitudes potentially inform policymaking?

3) Do the ethical and ideological oppositions to Uber and Lyft featured in popular media coverage appear in individuals’ reasoning about whether or not to use the services? 4) Is there potential support for cities to make directed partnerships with Uber and Lyft to further their socially-driven goals? Data collected on respondents’ demographics, spatial characteristics (e.g. region, ‘urbanness’ of residence, etc.), usage characteristics, transportation needs, and stated preferences will act as independent variables to understand and predict dependent variables such as user identification (i.e. whether the respondent identifies as a user of Uber, Lyft, both, or neither), attitude, and preferred policy action on the part of their city. The hypotheses and results associated with these variables can be found in Chapter 4: Results Analysis, and a discussion of the research questions above occurs in Chapter 5: Discussion. 1.3.3 Thesis organization The thesis is organized as follows. Chapter 2 reviews relevant literature on taxi regulation and current literature on the emergence of ridesourcing. Chapter 3 outlines the data collection methodology and the survey design. Chapter 4 presents the results of the survey along with bivariate and multivariate analyses. Finally, Chapter 5 concludes with a discussion of the findings and impressions from local policymakers.

2. Literature Review

This review of related literature establishes a regulatory history of the taxi industry to provide context for the success of ridesourcing companies and the associated media-fueled controversy with taxi drivers. It then describes the limited but growing body of academic literature on ridesourcing, enumerating the potential costs and benefits of this new business model. This section concludes with a brief discussion of the gaps in literature and understanding that motivated the design of this study.

2.1 Regulatory History of Taxi Industry

From the hackney carriages of early 17th-century Europe, taxis evolved alongside private vehicle advancements until the automobile reached wide distribution in the early 20th century. By the early 1920s, few American cities had established any sort of regulation for private taxicabs (Gilbert and Samuels, 1982). However, as individuals gained enough wealth to start taxi companies, and then as the Great Depression motivated the search for alternative forms of employment, oversupply became the first problem cities tried to address with regulation (Dempsey, 1996; Gilbert and Samuels, 1982). In addition to exacerbating congestion, the surplus led new market entries to undercut standard fares, leading to rate wars in cities nationwide (Gilbert and Samuels, 1982; Hodges, 2007). As new drivers and vehicles continued to flood the market, newspapers and public officials demanded stricter regulation of the taxi industry (Dempsey, 1996; Rubinstein, 2014). In the decades following the Great Depression, cities established entry restrictions, medallion limitations, driver and vehicle licensing requirements, and fixed fare structures, nearly all of which went unchanged until the United States’ widespread industry deregulation in the 1970s and 1980s (Gilbert and Samuels, 1982; Rubinstein, 2014). After the taxi industry experienced its own wave of deregulation—which had ideological motivations and was intended to lower fares, improve services, and expand coverage to poorly served neighborhoods—almost all cities re-regulated their taxi industries due to the failure of deregulation to produce the expected results (Dempsey, 1996; Price Waterhouse, 1993; Teal and Berglund, 1987).

In the process of this regulatory way-finding, the taxi industry has become entrenched in the American economy, but in the last two decades it has been protected mainly by the regulations designed to keep competitors at bay. By arguing that taxi regulations are inapplicable because they do not own any cars or employ any drivers, Uber and Lyft were able to enter the market of private for-hire transportation without becoming subject to the operational restrictions imposed on taxis. This circumvention, combined with the companies’ widely cited efficiency improvements, enabled Uber and Lyft to pose serious competition to the established taxi industry, igniting a controversy between taxi drivers and these companies that came to be known as disruptors. The popular media has capitalized on this controversy in its coverage of the

18 evolution of Uber and Lyft,5 but some authors argue that this focus is too narrow, and leaves more pressing questions about vehicle use and ownership out of the discussion (Rayle et al., 2016).

2.2 The Ridesourcing Debate

As alluded to in the Background, ridesourcing as a model has numerous potential impacts, not just for the incumbent taxi industry, but for potential users of for-hire transportation, for the people contracted as drivers by these companies, and for society as a whole. These impacts may be positive or negative, and are only beginning to be understood as new studies are published on specific questions. For now, the literature largely frames the debate by enumerating the potential costs and benefits and identifying areas for future study.

Reflecting the fixation of ridesourcing as a direct competitor to the taxi industry, many authors cite the efficiency gains of ridesourcing companies over the services offered by taxis and livery vehicles, particularly in gains of reduced wait times and lowered search costs (Rayle et al., 2016; Rogers, 2015). Some emerging literature argues that potential benefits include potential earnings for drivers, potential savings for consumers, resource conservation for urban space previously reserved for parking, reduction of car ownership, and supplementation of public transit (Metcalf and Warburg, 2012; Rogers, 2015; Silver and Fischer-Baum, 2015). However, citing the dearth of conclusive literature on the measurable benefits of ridesourcing, many authors use literature on taxis, ridesharing, and paratransit as a stand-in, which allows for further claims that ridesourcing could constitute an integral component of a city’s public transit and mobility network if planners and regulators recognize its potential (Austin and Zegras, 2012; Gilbert and Samuels, 1982; King, Peters, and Daus, 2012). In contrast, other authors suggest ridesourcing may actually compete with transit, increase overall VMT and urban congestion, and enable companies to employ droves of contractors and part-time workers, thereby bypassing the need to offer full-time benefits (Austin and Zegras, 2012; Flegenheimer and Fitzsimmons, 2015; Kenney and Zysman, 2015; Rogers, 2015). Ridesourcing may also endanger consumer safety, enable implicit bias between drivers and passengers, and exclude those who don’t have access to smartphones (e.g. the low-income and elderly) (Rayle et al., 2016; Rogers, 2015). Echoing the sentiments of popular media, much recent literature refers to Uber and Lyft as industry disruptors, suggesting that they pose a threat to industries operating under existing regulations, which include not only the taxi industry and public transit, but the private car industry as well (Fritz, 2014; Kenney and Zysman, 2015; Lobel, 2015). Critics of this disruption point not to the economic decline of an industry in general, but to the disparate impacts felt by taxi drivers, who used to be able to make a livelihood out of driving (Gilbert and Samuels, 1982; Perez, 2016; Rubinstein, 2014). Supporters, however, argue that industry disruption is a form of innovation that may mitigate the inefficiencies of taxi

5 Examples are numerous. See, e.g., Rubinstein (2014), Flegenheimer and Fitzsimmons (2015), and Perez (2016).

19 licensing and sustain long-term economic growth (Cramer and Krueger, 2016; Rogers, 2015, citing Schumpeter, 1947). Weighing in on components of this larger debate, a number of local studies and case studies have been conducted by private consultants (commissioned by local officials or the companies themselves) and academics. These studies include investigations into congestion (City of New York, 2016); service to low-income neighborhoods in Los Angeles (BOTEC Analysis Corporation, 2015); motivations of and benefits for drivers (Anderson, 2014; Hall and Krueger, 2015); trip-specific comparisons of taxis, transit, and ridesourcing in San Francisco (Rayle et al., 2016); and annual travel surveys and references in public opinion polls (Corey, Canapary & Galanis Research, 2014; Morning Consult, 2015). While these studies have a role in informing research and policymaking, only the San Francisco case study contributes meaningfully to the body of academic literature, and only the national tracking poll offers a glimpse of trends observable outside individual cities. Still unrepresented or underrepresented in the literature is independent research on the use of ridesourcing and its costs and benefits, the constituent perspective, and mechanisms for direct translation of findings to policy. This thesis will not fill all of those gaps, but it will offer new information on public opinion at a national level, including connections to policy, while remaining unaffiliated with either Uber or Lyft.

3. Data and Methodology

3.1 Research Methodology and Data Acquisition

The primary data collection method was a standardized, self-administered questionnaire hosted and distributed by Qualtrics, a private research software and survey company. Qualtrics facilitates the creation of informational surveys for market research, customer preference studies, and a number of additional forms of online data collection. The survey used in this thesis included 25 questions about respondents’ use, behavior, and preferences, along with 7 questions capturing basic demographic information (see “Survey questionnaire design” below). The company allows users to distribute the completed questionnaire to potential respondents via email, or offers a service called Panels in which users can purchase responses collected from a pool of over 4 million potential respondents (Qualtrics, 2010). The survey was bookended by a series of short, informal interviews: diagnostic interviews preceded the survey design to help narrow the scope of salient research questions, and interviews with Cambridge-based policymakers followed the results analysis to discern impressions and potential impact. As a methodology, the nationally-distributed survey was selected in place of more detailed qualitative methods such as semi-structured interviews due to its ability to capture the perceptions of the general public (as opposed to, say, policymakers or opinionated commentators). Intercept interviews might have achieved a similar sample, but would have been difficult to conduct across multiple cities, leaving any trends in the responses potentially explainable by local factors. Information about the targeting mechanisms, the sample size, quota, data validation techniques, the response rate, and the survey design is detailed below. 3.1.1 Targeting, sampling, and quota Targeting This study targeted adults living in major metropolitan areas in the United States (see Table 2 below for further detail). The Panel was profiled such that Qualtrics sent links to the survey only to respondents who were a) living in the United States, and b) over the age of 18. To achieve the additional target for respondents living in major metropolitan areas, all respondents who consented to take the survey were asked to select their place of residence from a drop-down menu of the top 15 metropolitan statistical areas by population in the United States (U.S. Census Bureau, 2015). To discourage respondents from trying to guess the qualifying responses, the list included 20 metropolitan areas, as well as the options “I live more than 20 miles outside of one of these metropolitan areas” and “I live in or near a metropolitan area not on this list.” Respondents who selected one of the top 15 areas were allowed to continue the survey, while respondents who selected one of the disqualifying areas or one of the two alternative options described above were terminated from the survey and not included in the final sample.

Qualtrics quoted a price of $5.50 per response for the use of the three targets described above, but agreed to charge only $5.00 per response due to the study’s affiliation with MIT, a non-profit educational institution. This discount did not impact the amount paid to each respondent (see Recruitment and payment below under “Data validation”).

Table 2 Targets used for recruitment Target Justification United States residency This thesis has a domestic focus, and including international respondents would be too complex and expensive to justify collecting information of marginal relevance to American policymakers. Regional variations represent the extent of the comparison sought for this analysis. Adults Adults are the primary stakeholder in 18+ years of age transportation policy; ridesourcing users must be at least 18 to download the associated app, and members of the public must be at least 18 to vote. Major metropolitan areas Ridesourcing operations are primarily urban Top 15 in U.S. by population problems, so this target exists to eliminate suburban and rural respondents.

Sampling frame To achieve a sample of statistical significance for a population of adults living in U.S. metropolitan areas, I constructed a sensitivity table to show potential sample sizes contingent upon the desired confidence and margin of error (see Table 3).6

Table 3 Potential sample sizes Confidence 99% 95% 90% Margin of Error ± 3% 1849 1068 748 ± 5% 666 385 269 ± 7% 340 196 138 ± 10% 167 97 68 Note. Bold represents sample size chosen for study.

6 The population of adults living in U.S. metropolitan areas is anywhere between 82 and 198 million, depending on how it is estimated. The sample sizes calculated in Table 3 are correct for either value.

Balancing the desire for a statistically significant study against the cost per response for reaching a wide audience according to the target described above, I chose to seek a minimum sample size of 385 respondents, which produces results with 95% confidence and a 5% margin of error. The survey returned a total of 394 usable responses, called “good completes” by Qualtrics (see “Response rate” below).

Quota mechanism The survey asks respondents to self-identify as a user or non-user of ridesourcing (response options include user of Uber only, user of Lyft only, user of both, and user of neither). To reduce the chance of overrepresentation of either users or non-users, a quota was instituted to eliminate any respondents who identified as a category that was already sufficiently represented. The quota, a tool operated by Qualtrics, was designed to prevent more than 75% of respondents identifying with one category, but to continue capturing the natural split between users and non- users in the event of overrepresentation of one category. In other words, if 75% of respondents (or 288 of the sought-after 385) had already identified as users, any future respondents who identified as users would be disqualified before completing the remainder of the study; however, the quota mechanism would still increment the user category to record the actual ratio of users to non-users. While a useful stopgap, the quota ended up being unnecessary. 124 respondents (31%) identified as non-users of ridesourcing, while 271 respondents (69%) identified as users, giving the study sufficient representation from both categories without having to implement the quota mechanism to suppress responses from one category. 3.1.2 Data validation Recruitment and payment Qualtrics partners with external organizations and programs (“Panels”) to recruit survey participants. Each panel partner has their own method of recruitment, though the methods are all relatively similar. Typically, respondents choose to join a panel through apps, online games, and rewards programs. Upon registration, they enter some basic personal data, including demographic information, hobbies, and interests, among others. When a survey is created for which an individual would qualify, that individual is notified via email and invited to participate in the survey for a given incentive. The email invitation is generic, and includes no identifying information on the topic of the survey. Individuals are simply told that they qualify for a survey, are given a link to the survey along with the anticipated duration, and are told to follow the link if they would like to participate in exchange for the given incentive. Though incentives may occasionally include direct cash deposits to the participant’s bank account, they are more often given on a point system. Points may be pooled from multiple survey completions and later redeemed in the form of gift cards, frequent flyer miles, credit for online games, and other

23 rewards. Exact amounts vary, and incentives are only earned by responses that meet Qualtrics’ quality control standards, described below. That respondents were targeted through online platforms rather than through traditional paper surveys sent in the mail suggests the potential for a sampling bias toward people who have access to and know how to use computers or smartphones. This bias is likely not to be particularly limiting as 84% of U.S. households reported owning a computer in 2013 (File and Ryan, 2014) and 64% of American adults are smartphone owners (Smith, 2015). Further, ridesourcing policy will likely have the most meaningful impact on people who are potential users, those who use computers and smartphones. Quality control measures

Of the respondents who initiate the survey, only “good completes” are included in the final sample size and compensated for completion. To qualify as “good completes,” respondents must make it to the end of the survey, and must adhere to the following quality control standards:7

 Attention checks: Depending on the length of the questionnaire, two or three “attention checks” are embedded within the questions to ensure respondents aren’t “straight-lining” through the survey. In this study, two attention checks were used: 1) “Please select neutral for this line,” and 2) “Please select somewhat negative for this question.” Respondents who did not respond as directed were disqualified.  Careful completion checks: Respondents who insert random strings of characters into text boxes (e.g. in questions for which “Other (please specify): _____” is an option) are disqualified. For example, a participant who responded “fnvkldn” in one question was excluded from the sample and not compensated for participating.  Speeding checks: Respondents who complete the survey in less than one-third of the median completion time (6.5 minutes, as measured through a “soft launch” described below) are disqualified for “not responding thoughtfully.” Due to a combination of the above measures, participants not qualifying or consenting, and participants dropping out due to survey fatigue, 673 people had to initiate the survey before 394 “good completes” were collected. Qualtrics quality control measures do not, however, disqualify more ambiguous, potentially unthoughtful responses, such as those that include selecting “neutral” for all the options listed in a single question. Consider the following question: “Between Uber and Lyft, if you prefer one over the other, how do the following factors influence your preference? For each row, please select the company you believe to be better, or select ‘Neutral’ if you believe them to be about equal.” Respondents were then shown a list of seven factors and asked to choose “Uber is better,” “Neutral,” or “Lyft is better” for each. 99 of 394 respondents (25.1%) chose the same

7 As explained in phone conversation with project managers Nathan Richard and Emily Davis, and supported by a guide published by Qualtrics (2014).

24 answer for each factor (the vast majority choosing all “neutral”), suggesting that they were potentially not paying close attention or being thoughtful with their responses. However, given that it would be impossible to effectively determine which respondents were being lazy and which legitimately found the two companies to be neutral in every aspect, none of the respondents were disqualified. Finally, though not a quality control measure, “forced-response” was implemented for all but one question in the survey. This feature eliminates the problem of participants skipping questions but still counting toward the final sample. Study “soft launch” Before the study was distributed for collection of all 385 requested responses, a pilot study, or “soft launch,” was conducted to collect a portion of the total sample (n=36 responses) to ensure proper manifestation of the survey logic and high-quality responses. The responses were collected over the course of a single day, and then sampling was paused while I reviewed the results for quality. The soft launch returned the median time for completion (394.5 seconds, or 6.5 minutes), and demonstrated that the survey logic worked as expected. Determining the median duration allowed for the use of a speeding check as described above. 3.1.3 Response rate The study had a response rate of 27%, which Qualtrics calculates by comparing the number of notifications presented within apps to the number of people who initiate the survey. The responses were collected over the course of three days (March 17, 2016, to March 19, 2016). Survey start times were clustered between 5:00pm and 9:00pm, and between 11:00pm and midnight, which may suggest that a majority of the respondents were employed. Of the 673 individuals who initiated the survey, 671 individuals answered the consent question, 657 consented to participate in the study, 639 gave the metropolitan area in which they live, 567 were eligible to continue, and 395 completed the survey without dropping out or being eliminated due to the quality control measures described in the previous section (i.e. an overall retention rate of 59%).

Though only 385 high-quality responses were requested, Qualtrics included 10 additional responses free of charge. Of the 395 “good completes” collected, n=394 were included in the final sample.8 Of those 394 responses, 45 demonstrated two or more instances of potential laziness as described in the above section (e.g. selecting “neutral” for each option in a multi-part matrix question) or gave unlikely responses (e.g. responding that they had taken more than 100 rides with Uber or Lyft, but had taken their first ride less than a month ago). Despite these

8 One response was disqualified despite passing Qualtrics’ quality checks due to inclusion of sexual innuendo in each optional text-box.

25 responses being candidates for disqualification, I elected to retain them in the final sample for two reasons: 1) they met the strict quality standards set by Qualtrics, and may just be thought of as responses on the lazy end of the qualifying spectrum; and 2) they included useful information in response to other questions in the survey.9 3.1.4 Limitations With a sample size of 394, this survey is representative of the adult population of U.S. metropolitan areas with 95% confidence and a margin of error of 5%. However, when responses are broken down by metropolitan area of residence, the resulting sub-samples do not retain this level of significance in representing their respective areas (e.g. in New York, the sub-sample of n=66 respondents has 90% confidence and a 10% margin of error). As such, the aggregate sample of n=394 respondents is representative of the U.S. adult population, but the sub-samples of particular metropolitan areas are not sufficiently representative for accurate cross-city comparisons. This limitation also implies that the findings may be meaningful for capturing national trends, but that policymakers might need to conduct similar studies locally to reach meaningful conclusions for specific metropolitan areas. Aggregating individual metropolitan areas into their corresponding census regions—Northeast, South, Midwest, and West—improves the statistical accuracy (e.g. the Northeast sub-sample of 110 respondents has 95% confidence with a roughly 8% margin of error). Targeting respondents from the top 15 metropolitan statistical areas by population also limits the study by failing to capture the personality of smaller cities where the respective market share of Uber and Lyft may differ significantly from that in larger cities, and by failing to capture the regulatory diversity of these metropolitan areas that might provide much-needed context for respondents’ attitudes and policy-related opinions. Further, by limiting the sample to those who have access to a computer or smartphone, this study underrepresented the elderly and the very low-income. While there is an argument for only attempting to capture the perspective of individuals who are potential users of ridesourcing (e.g. excluding minors and a majority of rural respondents), this underrepresentation excludes individuals who may be impacted by ridesourcing, even if they are not potential users.10

Finally, as is the case in most surveys, there is a tradeoff between asking respondents to give detailed information and keeping the questionnaire digestible enough to not fatigue potential respondents. Because the results of this survey are intended to be of use primarily to policymakers, the questionnaire does not include as many or as complex questions as might be

9 In some cases, it was reasonable for respondents to feel completely neutral about Uber and Lyft (e.g. in the comparison matrices) given that they were unfamiliar with the companies and relied instead on their own private vehicle for a majority of their transportation needs. 10 Low-income neighborhoods and elderly populations are invoked by Uber as vulnerable groups that may benefit from its accessible services (Smart et al., 2015; Uber Technologies, 2015), but limited independent research has been published in support of these suggestions.

26 useful to transportation engineers or to Uber and Lyft themselves. This study is meant to supplement trip-specific studies rather than offer a direct comparison.11 3.1.5 Survey questionnaire design This section describes the informal interviews conducted in advance of the iterative survey design process, the final questionnaire design, and the informal interviews conducted at the conclusion of the study to gather policymakers’ impressions on the results. Informal interviews Prior to the initial survey design, a series of informal interviews with transportation researchers established important topics to be included in the questionnaire. Apparent across multiple interviews, for example, was a high level of interest and concern regarding drivers’ status as contractors and their associated benefits. Interviewees were also curious about the age distribution of commonly-held attitudes and more than one interviewee emphasized the distinction between people who have used Uber or Lyft and people who have taken rides as guest but not initiated them. These interviews acted as a stand-in for a more open-ended pilot survey that similarly could have established the salient points of discussion in the ridesourcing conversation. They were moderately effective in discerning priority topics in that they resulted in the inclusion of questions about the driver-as-contractor issue, but did not result in the inclusion of questions regarding the surge pricing issue, which has been another popular point of controversy. No respondent mentioned surge pricing in their comments, but participants may have expressed definitive opinions on the subject if directed to. Survey design iterations Before the survey launch in March 2016, the questionnaire underwent several iterations through informal workshopping sessions with limited audiences. Unlike a soft launch, these workshopping sessions did not reach the anticipated audience and return a small sample of interpretable results for evaluation. Rather, small groups of transportation researchers evaluated the questionnaire in terms of content and logic, and gave their feedback on the types of responses they expected and lines of inquiry they believed to be absent or underdeveloped. As with the informal interviews described above, this iterative process resulted in a more thoughtful final survey design without the time or financial costs associated with conducting multiple pilot studies. Final survey design This section describes the questionnaire structure and the logic behind each question. Table 4 below summarizes the questions asked by section: screeners, use and behavior, preference, mode

11 See, e.g., Rayle, Dai, Chan, Cervero, and Shaheen (2016).

27 choice, attitude, implications, and demographics. Note that nearly every question was asked of every respondent, regardless of whether they identified as users or non-users of ridesourcing, based on the idea that an individual does not need to be a user of Uber or Lyft to have opinions on or be impacted by their operations. Following the table, each question and the motivation for its inclusion are described in greater detail.

Table 4 Summary of survey questionnaire organized by section Section Questions Screeners Do you consent to participate in this study? Please select the metropolitan area in which you live.

Use and Behavior – Pt. 1 Do you consider yourself to be a user of Uber only, Lyft only, both Uber and Lyft, or neither Uber nor Lyft? When do you use Uber or Lyft? [Contingent display: Not shown if respondent selects ‘Neither Uber nor Lyft’ above.]

Preference – Pt. 1 When you decide to use Uber or Lyft as opposed to other modes, what is your primary reason for doing so? As a non-user of Uber, Lyft, or both, what is your primary reason for not using Uber or Lyft as opposed to other modes? [Contingent display: Not shown if respondent selects ‘Both Uber and Lyft’ above.] What is the source of your ideological or ethical opposition to Uber or Lyft? [Contingent display: Only shown if respondent selects ‘I have an ideological or ethical opposition to it’ in response to previous question.]

Use and Behavior – Pt. 2 Have you ever taken an Uber/Lyft? Note. The questions in this How many times have you taken an Uber/Lyft? section are asked twice: About how long ago was your first Uber/Lyft ride? once about Uber and a Do you have the Uber/Lyft app installed on your smartphone? second time about Lyft.

Preference – Pt. 2 Between Uber and Lyft, if you prefer one over the other, how do the following factors influence your preference?

Mode Choice Please rank the travel modes listed below according to the frequency with which you use them. What is your priority when choosing between modes of transportation for your daily activities?

Attitude How would you describe your attitude toward Uber?

How would you describe your attitude toward Lyft? How would you describe your attitude toward ride-hailing technology in general?

Implications How should your city respond to Uber? How should your city respond to Lyft? If you think your city should form partnerships with Uber or Lyft, what kind of partnerships would you support? The following aspects of ride-hailing technology make me want to use Uber or Lyft…[More, Less, or Neutral]

Demographics What is your gender? What is your age? Which best describes your race and/or ethnicity? What is the highest level of education that you have completed? What is your approximate average household income? In what zip code do you live? In what zip code do you work?

The questionnaire begins by asking respondents to consent to the study, and then to select the metropolitan area in which they live. These two questions are meant to screen out individuals who do not consent to the use of their responses in academic research as well as those who do not live in a major metropolitan area as defined in the above section, Targeting. The residence question also serves to elucidate regional trends, though direct city-to-city comparisons are not possible due to insufficient sample sizes on a disaggregate level. It is supplemented by two questions in the demographics section at the end of the survey that ask respondents to give the zip codes in which they live and work. The first substantive question in the survey is “Do you consider yourself to be a user of Uber only, Lyft only, both Uber and Lyft, or neither Uber nor Lyft?”12 The ordering of this self- identification question is intentional as it forces respondents to share how they identify before they are influenced by the follow-up questions about their use of the services and their preferences. It is a stated-preference question that does not prime the respondent beforehand. This question serves to reveal the division between users and non-users of ridesourcing services, and to compare individuals’ self-identification against their stated use, preferences, and demographic traits. The questionnaire next asks respondents when (i.e. for what purposes) they use Uber or Lyft, if they’ve initiated rides or ridden only as tagalongs, how many times they’ve used either service,

12 Asking respondents to identify as users or non-users establishes a specific dichotomy. The survey could have asked respondents to identify as drivers or non-drivers, or as users or drivers (and could have subsequently disqualified unaffiliated participants). The choice to establish a user/non-user dichotomy was to represent the public as best and simply as possible, and to emphasize that non-users may also share insights helpful to policymakers.

29 for how long, and whether or not they have the respective app installed on their smartphones. These use and behavior questions offer self-reported information about respondents’ actual usage profiles for comparison against their self-identification as users or non-users, as well as for predicting the responses given in the attitude and implication sections near the end of the survey. Study participants who respond that they use Uber, Lyft, or both are asked, “When you decide to use Uber or Lyft as opposed to other modes, what is your primary reason for doing so?” Respondents may then select their top three reasons from a list including “It is cheaper,” “It is faster,” “It is more convenient,” “It is safer,” “It makes me feel modern,” “My friends do it,” and “Other (please specify).”13 The first four options represent primary mode choice considerations in choosing among diverse transportation options (Ben-Akiva and Lerman, 1985; Olsson, 2003; Rodrigue, Comtois, and Slack, 2013). The other two reasons were included as options due to conclusions reached during the informal interviews described above, namely that people may be influenced to use ridesourcing services because of the feeling of being high-tech or because of peer pressure and desire to conform. Similarly, participants who respond that they don’t use Uber, don’t use Lyft, or don’t use both are asked to give their primary reasoning. They may again choose their top three reasons from a list including “It’s expensive,” “It’s slow,” “It’s inconvenient,” “It’s unsafe,” and “I have an ideological or ethical opposition to it.” The first four options are conversely analogous to the primary mode choice considerations described above. The latter option was included in response to significant media coverage demonstrating ethical concerns with Uber’s leadership and corporate attitude, and with the societal impacts of ridesourcing services generally.14 If respondents indicated an ethical or ideological opposition, they were then asked to select the source of their opposition from a list including “I am opposed to the corporate attitude,” “I disagree with the choice to treat drivers as contractors and not employees,” “I am opposed to the use of private cars as a mode of transportation in general,” and “I think one or both companies enjoy a lack of regulatory oversight that is unfair to consumers and/or to the taxi industry.” For all of the questions described above, respondents could also select “Other” and provide their own response. Gathering data on the presence and nature of ethical opposition to ridesourcing services allows for comparison against the degree suggested by popular media coverage. Finally, before transitioning to more explicit mode choice questions, the survey asks respondents to describe their preferences between Uber and Lyft. The question reads, “Between Uber and Lyft, if you prefer one over the other, how do the following factors influence your preference? For each row, please select the company you believe to be better, or select ‘Neutral’ if you believe them to be about equal.” Respondents are then shown a list of factors in a matrix-style

13 Nine questions throughout the survey make use of the answer option “Other (please specify).” This answer option allows respondents to provide an alternative if the multiple choice options given are not adequate. It counts as a valid answer, but only if the respondent inputs intelligible text into the subsequent text field. 14 Examples are numerous. See, e.g., Bilton (2014), Cherry (2015), and Weinstein (2014).

30 format in which they can select the prevailing company for each factor (see Fig. 1 below). Unlike previous questions, which were contingently displayed based on whether respondents considered themselves to be users or non-users, this question was displayed to all participants based on the assumption that people didn’t have to be a user of either Uber or Lyft to have opinions about which was ‘better’ in certain respects. Though company-specific, this question remains of use to policymakers, who may benefit from knowing if and how people perceive these two companies differently. There is limited data on this topic, and the primary source of information is popular media, which paints Uber as simultaneously more popular and more morally questionable than Lyft.15

Figure 1. Question structure shown to survey respondents for the question “Between Uber and Lyft, if you prefer one over the other, how do the following factors influence your preference?” Note. The last option, “Please select neutral for this line,” is an attention check as described in the above section, Quality control measures. Respondents are then asked to rank their transportation modes according to frequency of use. They can choose between Uber, Lyft, their own car or one that they rent or borrow, carpooling, public transit or commuter rail, taxi, walking, bicycle, or specify a different mode. In the first round of survey design, this question was supplemented by a ranking of preferred mode choice for the purpose of comparing the modes people wanted to be using against the modes they were actually using. However, deliberations during the survey design iterations described above revealed that such a question would not produce reliable responses unless the survey also

15 Sometimes, this disparity is present in individual articles. See, e.g., Huston (2016) and Shahani (2016).

31 provided respondents with specific trip scenarios to help frame their thinking, which would be too complicated for this intentionally short and digestible questionnaire. This question shows where people rank Uber and Lyft relative to their other travel modes. Next, respondents are asked, “What is your priority when choosing between modes of transportation for your daily activities?” They are allowed to choose from a list including “travel time,” “cost,” “availability,” “level of productivity while traveling,” or “Other/ some combination,” which is then followed by a medium-sized text box to allow the respondent to elaborate. The medium-sized text box accepts a longer comment than the small-sized text box used elsewhere in the questionnaire and also implicitly prompts the respondent to leave a longer explanation. This design decision originates from disagreement during deliberations over whether a multiple choice or open text box answer structure would be more appropriate. The combined answer format allows respondents to select one standard option if it is most appropriate, or explain the combination of options or alternative options that best describe their mode choice considerations. Comparing the results of this question to the corresponding mode ranking and reasons provided for using or not using Uber and Lyft should reveal inconsistencies in mode choice reasoning. Nearing the end of the survey, respondents are asked to state their attitude toward Uber, toward Lyft, and toward ridesourcing technology in general (anticipating that respondents are more familiar with the terms Uber and Lyft, and may conflate their attitude toward a company with that of the technology itself), each on a five-point scale of very positive to very negative. The responses to these questions are useful to compare across demographic indicators, and against respondents’ statements about their frequency of use and identification as a user or non-user of ridesourcing services. Respondents also answer several questions about the policy implications of their attitude, including “How should your city respond to Uber?” and “How should your city respond to Lyft?” with options including “My city should ban [Uber/Lyft],” “My city should regulate [Uber/Lyft] so that it has the same operational restrictions as taxis,” “My city shouldn’t do anything about [Uber/Lyft],” “My city should consider forming partnerships with [Uber/Lyft] (e.g. to supplement public transit, or to discourage private car ownership),” and “Other.” Respondents who answer that their cities should consider forming partnerships are asked what kind of partnerships they would support among the following: “To reduce instances of drunk driving,” “To reduce private car ownership or usage,” “To supplement access to public transportation,” and “Other.” These questions are meant to be of direct use to policymakers, who may benefit from having a gauge of public attitudes towards ridesourcing services and technology independent from popular media portrayals. Similarly, the implications questions provide explicit information on the salience of specific policy responses. Finally, respondents complete another matrix-style question titled “The following aspects of ride-hailing technology make me want to use Uber or Lyft [More, Less, or Neutral].” The survey displays a list of ten aspects associated with ridesourcing technology such as “automatic payment” and “rating system,” and respondents are directed to choose “More,” “Neutral,” or

“Less” for each aspect (see Fig. 2 below). This question helps capture the nuance in the consumer behavior/ policy response relationship, and should help policymakers understand what aspects of ridesourcing technology have the approval and disapproval of the public. Aspects that have significant approval demonstrate features that cities may want to promote about ridesourcing services (e.g. the option to save money by sharing a ride with another passenger), and potentially replicate in public transit services (e.g. automatic payment). Aspects with significant disapproval might be useful for policymakers to use as leverage or contingencies during negotiations with Uber and Lyft regarding continued operation (e.g. regulating the drivers-as-contractors issue).

Figure 2. Question structure shown to survey respondents for the question “The following aspects of ride-hailing technology make me want to use Uber or Lyft [More, Less, or Neutral].”

The last section of the questionnaire asks respondents to share demographic information. This section is saved for the end to avoid priming respondents. Instead of respondents potentially thinking about their identifying characteristics or education level, for example, while answering the survey, respondents are asked to describe their activity and opinions first, then asked to share demographic information once it will no longer potentially impact their responses. The survey asks for gender, age, race and/or ethnicity, highest level of education completed, approximate average household income, zip code of home address, and zip code of work address. In each question, respondents are given the option to respond “Prefer not to answer” to protect their privacy.

3.1.6 Interviews with Cambridge policymakers

Following the analysis of the results, I shared my findings with four local policymakers: the City of Cambridge licensing commission chair and executive director, a member of the Cambridge City Council, and a policy analyst at the Volpe National Transportation Systems Center. The survey described above is the primary research methodology employed in this study, but these interviews served as a supplement that allowed me to evaluate the utility of my findings to my intended audience. Of the policymakers interviewed directly, each was first given the opportunity to describe their understanding of regulation in Cambridge that might concern Uber and Lyft. I then asked the interviewee to describe if and how they felt their work included responsibility for regulating Uber and Lyft. I concluded by describing my study and the key findings, and asking the interviewee to comment on the utility of the results and on how they thought such a study might fit into the body of research that informs their perspectives as policymakers. Most interviewees also volunteered their thoughts on the most pressing questions related to the topic. These interviews helped ground the study in planning practice, and the results can be found in the Discussion section.

3.2 Analysis Methodology

The results analysis in the next section occurs in two parts: a basic univariate analysis of the survey responses, followed by bivariate and multivariate statistical analyses to understand relationships between variables and factors that determine variation in response to the survey questions. The univariate analysis consists primarily of distributions and cross-tabulations, while the bivariate analysis includes correlation tables showing relationships between independent and dependent variables. The multivariate analysis includes three types of logit models: a basic logistical regression, an ordered logit, and a multinomial logit (MNL).

4. Findings and Analysis

4.1 Survey Responses and Univariate Analysis

4.1.1 Respondent demographics Survey respondents were generally younger, better educated, and more represented by women than the average population of urban adults in the United States (see Table 5). Individuals between the ages of 18 and 24 were overrepresented (as were the 25-34 and 35-44 age groups, but to a much lesser degree) Respondents were also slightly wealthier than the average population, but about as white. These differences may be influenced by the sampling method: individuals who completed the survey may be more computer-savvy than the average urban adult, which could impact age, education, and income data.

Table 5 Demographics of survey respondents compared to adult population of the 15 aggregated metropolitan areas. Sources: a U.S. Census Bureau, 2010 Census. b 2010-2014 ACS 5-Year Estimates. Survey Results US Census, 2010 Responses Percent Gender a Female 237 60% 51% Male 155 39% 49% Prefer not to answer 2 1% n 394 100% 100% Age a 18-24 91 23% 10% 25-34 112 28% 14% 35-44 84 21% 14% 45-54 60 15% 15% 55+ 45 11% 23% Prefer not to answer 2 1% n 394 100% Race and/or Ethnicity a Native American or Alaska Native 0 0% 1% Asian or Pacific Islander 23 6% 9% Hispanic or Latino/a 51 13% 26% Black or African American 57 15% 16% White 242 61% 62% Multiple races and/or ethnicities 14 4% 3% Prefer not to answer 7 2% n 394 100% Education b High school (incl. GED) or less 77 20% 42%

Some college 97 25% 21% 2-year degree 52 13% 8% 4-year degree 97 25% 18% Professional degree 63 16% 11% Doctorate 6 2% n/a Prefer not to answer 2 1% n 394 100% 100% Household Income b Less than $25 K 49 12% 23% $25-49 K 85 22% 24% $50-74 K 93 24% 18% $75-99 K 58 15% 12% $100-124 K 31 8% 13% $125-149 K 23 6% $150-174 K 13 3% 5% $175-199 K 9 2% $200 K + 12 3% 5% Prefer not to answer 21 5% n 394 100% 100% Note. Education and income census data represent the U.S. population as a whole. Different collection techniques for U.S. Census data and survey data account for apparent discrepancy in the Hispanic or Latino/a category. Age percentages reported by U.S. Census do not add to 100 because categorization excludes minors.

Based on the population densities associated with home zip code, 56% of respondents live in urban areas, while 26% live in suburban areas, and 18% live in areas classified as rural, despite identifying as residents of a nearby metropolitan area. Respondents live between 3.3 and 156.8 miles from the center of the nearest metropolitan area, with an average distance from center of 32.1 miles and a standard deviation of 24.7 miles. Despite being instructed to not select a metropolitan area of residence in the survey if their home address was more than 20 miles outside of the metropolitan area, respondents with high distances from center were not disqualified due to variability in metropolitan area land coverage. 183 respondents (46%) reported living and working in the same zip code. Given that 7 respondents made explicit references to not working or working from home, it is reasonable to assume that some of those 183 respondents may also have given a work zip code without holding a traditional form of employment. Remaining in the sample are 204 respondents (52%) who reported different zip codes for their home and work addresses, implying at least some form of a regular commute. Correspondingly, while the survey did not explicitly direct participants to describe their level of access to a private vehicle, respondents made numerous implications on the topic. 60% of respondents made negligible or no reference to having access to a private vehicle, 29% of respondents implied that they have access to a private vehicle (i.e. by ranking car use highly among their modes or referring to avoiding drunk driving), 4% of respondents explicitly referenced having access to a friend or relative’s private vehicle, and 7% of respondents

36 explicitly referenced having access to their own private vehicle. The survey also did not ask respondents to identify their level of access to transit, but the respondents’ ranking of transit in the mode choice question could feasibly act as a proxy. Representation of the metropolitan areas included in the study closely tracked the corresponding populations, with the New York City, Chicago and Los Angeles metropolitan areas being the most highly represented (at 16.8%, 13.2% and 12.7%, respectively). Table 6 below shows the distribution with the metropolitan areas grouped according to U.S. Census region. Note that these areas of residence are self-reported and that home and work zip code are used in the statistical analyses for greater accuracy.

Table 6 US Census region and metropolitan area of residence of respondents Responses Percent Northeast New York City 66 16.8% Philadelphia 26 6.6% Boston-Cambridge 18 4.6% South Washington, DC 29 7.4% Houston 24 6.1% Dallas-Fort Worth 22 5.6% Atlanta 22 5.6% Miami-Fort Lauderdale 12 3.0% Midwest Chicago 52 13.2% Detroit 27 6.9% West Los Angeles 50 12.7% San Francisco-Oakland 22 5.6% Phoenix 18 4.6% Riverside-San Bernardino 6 1.5% Seattle-Tacoma 0 0% n 394 100%

4.1.2 Use and behavior User identification In response to the first substantive question in the survey, “Do you consider yourself to be a user of Uber only, Lyft only, both Uber and Lyft, or neither Uber nor Lyft?” 32% of respondents identified as a user of both services, 35% identified as a user of Uber only, only 1% identified as a user of Lyft only, and 31% identified as a user of neither service. Aggregated, these responses

37 show that 69% of respondents identify as ridesourcing users, while 31% identify as non-users. No other studies reviewed for this thesis offer an analogous finding for comparison. Trip purpose Table 7 presents primary trip purposes when using ridesourcing services. This question was only displayed to respondents who identified as users of ridesourcing, and respondents were allowed to choose up to three primary purposes from a predetermined list or provide their own. Of 270 responses, 51% were avoiding driving while intoxicated, 46% were for social/leisure purposes (e.g. bar, restaurant, concert, visiting friends and family, etc.), and 40% were getting to or from the airport. Smaller percentages of respondents reported using ridesourcing for times when transit is unavailable, for locations where transit is unavailable, for bad weather, for work or school, or for errands or shopping. The ‘work or school’ option was meant to capture people using ridesourcing for regular commutes, but it may have also captured people using ridesourcing in the middle of the business day (e.g. for client meetings), trips that can be expensed to the individual’s company. Only 3% of respondents provided some other purpose for using Uber or Lyft, including “getting somewhere faster than public transit,” “coming home from the mechanic,” “when taxi is unavailable,” or for appointments (e.g. doctor).

Table 7 Primary trip purpose when using Uber or Lyft Responses Percent Avoiding driving while intoxicated 138 51% For social/leisure purposes 125 46% Getting to or from the airport 108 40% Getting somewhere at times when public transit is unavailable 88 33% Getting to or from locations not accessible by public transit 67 25% Getting somewhere in bad weather 52 19% Getting to work or school 38 14% For running errands and shopping 34 13% Other 9 3% n 270 Note. Percentages do not add to 100 because respondents were allowed to choose their top three purposes. This question was not displayed to respondents who said they were users of neither Uber nor Lyft.

Reasons for and against choosing ridesourcing

Participants who responded that they considered themselves to be users of Uber, Lyft, or both were asked to give their primary reasons for choosing ridesourcing over other modes for the trips they described previously. Similarly, participants who responded that they did not consider themselves to be users of Uber, Lyft, or both were asked to give their primary reasons for not choosing ridesourcing. Reasons of cost, speed, and convenience dominated the arguments for

38 using Uber and Lyft (see Table 8), but 15% of respondents said one of their reasons was simply that “my friends do it.” Reasoning against choosing one or both of the ridesourcing services was more distributed: more than 30% of respondents said it was unsafe, and 35% said it was too expensive. 25% of respondents gave another reason (primarily having access to a car, using Uber so not having any need for Lyft, and not having heard of or thought of it) and 16% said it was inconvenient, some portion of which may have been in comparison to using a private vehicle. Only 12% said it was slow, but those respondents were nearly all (90%) Uber users commenting on why they don’t use Lyft, whereas most responses included a number of people (33-65%) who use neither.

Finally, though media coverage suggests that this number would be higher, only 6% of respondents gave an ethical or ideological opposition as a reason to not use one or both of the services. Interestingly, half of those respondents were Uber users (the other half were users of neither). Users of Lyft only or of both Uber and Lyft did not express an ideological opposition. When prompted to describe the source of the ethical or ideological dilemma, 12% said they are opposed to the corporate attitude, 41% said they disagree with the choice to treat drivers as contractors instead of employees, 29% said they are opposed to the use of private cars in general, and 35% said they think one or both companies enjoy a lack of regulatory oversight that is unfair to consumers and/or to the taxi industry—these responses represent 0.51%, 1.78%, 1.02%, and 1.52% of the total 394 responses, respectively. One respondent offered, “All of the above. I dislike the societal implications of this business model for several reasons.” Even if the media does over-represent these more sensational opinions, Nike’s experience with consumer disapproval due to perceived ethical violations in the 1990s suggests that this 6% figure may yet be significant (Nisen, 2013; Russell, 2014). In sum, the limited follow-up questions make it difficult to determine if this figure is low compared to representation in the media, or high compared to ethical opposition to other transport modes and companies.

Table 8 Reasons for and against choosing ridesourcing as opposed to other modes Responses Percent Reasons for It’s more convenient 202 75% It’s faster 127 47% It’s cheaper 111 41% It’s safer 57 21% It makes me feel modern 27 10% My friends do it 41 15% Other 4 1% n 270 Reasons against It’s expensive 93 35%

It’s unsafe 82 31% Other 66 25% It’s inconvenient 43 16% It’s slow 30 11% I have an ideological or ethical opposition to it 16 6% n 266 Note. Percentages do not add to 100 because respondents were allowed to choose their top three reasons for and against. This question was not displayed to respondents who said they were users of neither Uber nor Lyft.

This section has several limitations associated with the nature and structure of the questions asked. Some reasons for choosing or not choosing ridesourcing were actually included in the trip purpose question (e.g. avoiding driving while intoxicated, getting somewhere in bad weather, and at times when transit is unavailable), where they can still be analyzed, but not against their complements. The survey design also didn’t explicitly account for individuals’ car ownership, which people referenced heavily in their reasoning for not using ridesourcing. Finally, the options given for reasoning were very basic, and didn’t include more nuanced possibilities such as avoiding looking for parking, or specific convenient attributes (e.g. ease of payment, ease of booking, short wait time, reliability, etc.).

Frequency of use

After sharing what services they use and why, respondents answered a series of questions about their specific usage characteristics of both Uber and Lyft (see Table 9). The results show that 72% of respondents have taken an Uber ride at least once, while only 41% of respondents have taken at least one Lyft ride, and the people who have taken at least one Lyft ride have done so only a few times (55% of people who have used Lyft have done so between 1 and 3 times, compared to only 36% of people who have used Uber). Of each category, between 21 and 29% of people who have taken a ride have only done so as someone else’s guest (i.e. did not book or pay for the ride themselves). Further, 64% and 38% of respondents said that they currently have the Uber and Lyft apps, respectively, installed on their smartphone or did at one time (5% of respondents indicated that they do not have a smartphone). Compared to the user identification data, which demonstrated that 69% of respondents identify as users of at least one ridesourcing service, these findings suggest that some people who identify as users may do so even if they don’t currently have the app on their phone or tag along with other people who do use the app. Respondents who said they have taken at least one ride with Uber or Lyft have mostly done so between one and ten times over the course of the last three years. The 5% of respondents who indicated that they do not have a smartphone represent 12% of overall non-users of ridesourcing, suggesting that not owning a smartphone may be a barrier to ridesourcing (they also represent 4% of the respondents who identify as users of Uber only).

Table 9 Frequency of use of Uber and Lyft Uber Lyft Responses Percent Responses Percent Ever used Yes, and I have initiated at least one ride myself 223 57% 115 29% Yes, but someone else booked the ride for me 60 15% 47 12% No 111 28% 232 59% n 394 100% 394 100% Number of times Between 1 and 3 102 36% 89 55% Between 4 and 10 117 41% 46 28% Between 11 and 100 60 21% 25 15% More than 100 4 1% 2 1% n 283 100% 162 100% Length of time since first ride Less than one month 83 29% 32 20% Between a month and a year 136 48% 103 63% Between one and three years 60 21% 23 14% Between three and six years 4 1% 4 2% n 283 100% 162 100% App installed on smartphone Yes, currently 193 49% 90 23% No, never 125 32% 225 57% No, but once 58 15% 58 15% No smartphone 18 5% 21 5% n 394 100% 394 100% Note. Participants who responded that they have never taken an Uber were not asked Uber-related follow-up questions, and those who responded that they have never taken a Lyft were not asked Lyft- related follow-up questions.

4.1.3 Preference between Uber and Lyft To understand how people choose between Uber and Lyft, the survey asked all respondents to select the company they believe to be better on individual criteria in a list of factors (Fig. 3 and Table 10). Respondents who identified as non-users of ridesourcing were largely responsible for the “neutral” responses, but half of them weighed in on at least one criterion. Respondents were most neutral about safety, cost, technical and aesthetic elements of the app, and promos and coupons. Though more people said Uber was better than Lyft on all fronts, those four criteria represented the least dramatic distributions. Uber was a clear winner in people’s minds on two fronts: “Number of drivers available, and thereby wait time” (46% of respondents said Uber was better compared to 4% for Lyft), and “It’s what my friends use, or it’s the only one I’ve ever considered using” (54% Uber compared to 3% Lyft). These two factors provide evidence for potential network effects, which might help explain some of Uber’s market dominance. One

41 front on which Lyft still lost, but had the highest support (10% compared to 35% for Uber), was the “company’s attitude or identity,” which aligns with the media- and self-promoted notion that Lyft is a friendlier company (i.e. Lyft’s slogan is “Your friend with a car,” while Uber’s is “Everyone’s private driver”).

Between Uber and Lyft, if you prefer one over the other, how do the following factors influence your preference? (%) 0 10 20 30 40 50 60 70 80 90 100 It's what my friends use, or it's the only one I've ever considered using Number of drivers available and thereby wait time

The company's attitude or identity Technical or aesthetic elements of the app or technology Promotions and coupons

Cost

Safety

Uber is better Neutral (the companies are about equal) Lyft is better

Figure 3. Comparison of preferences between Uber and Lyft based on seven criteria.

Table 10 Preferences between Uber and Lyft Uber is better Neutral Lyft is better It’s what my friends use, or it’s the only one I’ve ever 54% 43% 3% considered using Number of drivers available, and thereby wait time 46% 50% 4% The company’s attitude or identity 35% 55% 10% Technical or aesthetic elements of the app or 29% 66% 5% technology Promotions and coupons 27% 64% 8% Cost 23% 68% 9% Safety 20% 72% 9% Note. The “Promotions and coupons” and “Safety” categories do not add to 100 due to rounding. All other categories (rows) add to 100.

4.1.4 Mode choice and justification Respondents were told to consider the trips they make on a regular or semi-regular basis (e.g. to work or school, to run errands, to socialize, to access entertainment, etc.), and were then asked to

42 rank eight common travel modes according to the frequency with which they use them. The average ranking is shown below in Fig. 4, ranked according to the average value attained by each mode (shown in parentheses). People who chose “Other” provided alternative modes to those listed, including rides from friends or relatives, running or jogging, and motorcycle.

Mode Choice (Avg.) 1 - Highest 2 3 4 5 6 7 8 9 - Lowest Car (2.23) 67% 8% 6% 3% 4% 3% 4% 4% 1% Walking (4.26) 6% 20% 16% 16% 12% 11% 10% 9% 1% Uber (4.28) 8% 18% 15% 14% 13% 11% 13% 7% 1% Transit (4.33) 10% 12% 21% 14% 13% 9% 11% 10% 1% Carpooling (4.37) 4% 24% 13% 14% 11% 13% 10% 9% 1% Taxi (5.32) 2% 5% 12% 14% 18% 20% 12% 14% 1% Lyft (5.77) 2% 6% 8% 12% 13% 14% 20% 21% 4% Bicycle (5.78) 3% 5% 8% 11% 14% 17% 17% 22% 3% Other (8.66) 0% 1% 1% 0% 2% 1% 3% 5% 88%

Figure 4: Mode choice heat map showing how respondents ranked each mode. Figure 4 shows that responses are generally clustered around the diagonal (as expected, e.g. the highest ranked mode should have many respondents labeling it ‘1’ and the second-highest ranked mode should have many respondents labeling it ‘2’ and so-on). Standing out from this trend is carpooling, which 24% of respondents ranked second. I predicted that most people would not have or use cars due to being city-dwellers, but including respondents from metropolitan areas rather than just urban areas resulted in capturing a high proportion of individuals who own cars. As such, 83% of respondents ranked “car” among their top four modes. Though public transit falls just below Uber in terms of average ranking, slightly more than half of respondents (52%) ranked public transit higher than Uber. Metropolitan areas in which public transit is doing better than Uber (according to percentage of respondents who rank transit above Uber) include Boston-Cambridge (72%), San Francisco-Oakland (68%), Riverside-San Bernardino (67%), Washington, DC-Arlington-Alexandria (66%), and Chicago (65%). New York City, where one might expect this percentage to be the highest, is not far behind at 58%. The remaining metro areas all have percentages below 60%, with Houston, Miami, Phoenix, and Detroit all bottoming out at 33%. While compelling, note that these values are representative only of the sample as the disaggregated samples are not large enough to represent the populations of their associated metropolitan areas. Respondents were also asked to give their priority when choosing between modes of transportation. 36% said their priority was availability, 30% said cost, and 27% said travel time. Only 4% of respondents said their level of productivity while traveling was a priority, and 5% gave another priority or some combination of the above: four respondents said a combination of cost and availability, three referenced a combination of cost and travel time depending on

43 urgency, four said they had a car and therefore used it, and the remainder referenced parking, safety, traveling green, and being disabled. Relating the mode choice priority question back to the reasons provided for or against using ridesourcing, respondents were largely consistent in their stated reasoning. For each priority (availability, cost, travel time), a plurality of respondents chose its equivalence when asked to give their reasons for or against choosing a ridesourcing service. For example, of the 111 respondents who chose “It’s cheap” as one of their reasons for using Uber or Lyft, 42 also said that cost was their priority when choosing between modes, more than associated with any other priority. Similarly, of the 202 respondents who chose “It’s convenient” as one of their reasons, 69 also said availability was their priority when choosing between modes, also more than associated with any other priority. The same is true for “It’s expensive,” “It’s slow,” and “It’s inconvenient” as reasons against choosing a ridesourcing service aligning with cost, travel time, and availability, respectively, as mode choice priorities. The only exception was in the “It’s fast” reason, in which 30 of 127 respondents said travel time was their priority, but 45 and 41 respondents chose cost and availability, respectively. This finding supports the hypothesis that individuals’ mode choice priorities should align with their reasoning for or against using certain modes. 4.1.5 Attitude and policy implications Reported attitudes toward Uber, Lyft, and ridesourcing technology Despite hypothesizing that respondents would have a generally negative attitude toward Uber and a more positive attitude toward Lyft and ridesourcing technology in general, respondents actually expressed an overall positive attitude toward Uber and a more neutral attitude toward Lyft, with unaffiliated ridesourcing technology landing in between (see Table 11). In sum, 67% of respondents indicated a “very positive” or “somewhat positive” attitude toward Uber, with 23% indicating “neutral/ambivalent;” while 41% of respondents indicated a “very positive” or “somewhat positive” attitude toward Lyft, with 53% indicating “neutral/ambivalent.” Toward ridesourcing technology irrespective of the two companies, 63% indicated a “very positive” or “somewhat positive” attitude, and 28% selected “neutral/ambivalent,” closely tracking the response rates toward Uber, which suggests potential conflation of the company and the technology. Uber’s average attitude rating is slightly more positive than Lyft’s average (2.1 and 2.5, respectively, on a scale in which 1 = “Very Positive” and 5 = “Very Negative”). The aggregate data don’t show much divergence in negative attitudes between categories, but the breakdown by user identification in Table 10 shows users of Lyft only expressing neutral or negative attitudes toward Uber but not Lyft or ridesourcing technology in general. Despite Uber’s more positive average rating, Lyft had fewer total negative (“somewhat” or “very”) attitude ratings.

Table 11 Attitudes toward Uber, Lyft, and ridesourcing technology in general, irrespective of the two companies Total Very Somewhat Neutral/ Somewhat Very Count Positive Positive Ambivalent Negative Negative

Attitude toward Uber Neither Uber nor Lyft 124 5% 22% 52% 18% 4% Lyft only 5 20% 0% 40% 40% 0% Uber only 137 50% 40% 8% 1% 1% Both Uber and Lyft 128 47% 37% 12% 4% 1% n 394 135 129 92 31 7 Attitude toward Lyft Neither Uber nor Lyft 124 2% 10% 73% 10% 5% Lyft only 5 80% 20% 0% 0% 0% Uber only 137 9% 14% 71% 4% 1% Both Uber and Lyft 128 36% 47% 16% 1% 0% n 394 66 92 209 19 8 Attitude toward ridesourcing technology Neither Uber nor Lyft 124 6% 23% 55% 12% 4% Lyft only 5 80% 20% 0% 0% 0% Uber only 137 35% 39% 20% 4% 1% Both Uber and Lyft 128 41% 40% 13% 5% 1% n 394 113 134 112 27 8 Note. Values shown are percentages of total respondents in user identification categories. Some rows do not add to 100 due to rounding. The “Lyft only” respondents constitute too small a sample to be reliable.

Corresponding with the hypothesis that respondents would have a more negative attitude toward Uber, I also predicted that the attitude expressed toward ridesourcing technology would be generally more positive than the attitude expressed toward Uber, and potentially toward Lyft as well. With 39% of respondents expressing a more positive attitude toward ridesourcing technology than toward Lyft and 15% vice versa, this prediction held true for Lyft. It did not, however, hold true for Uber: only 19% of respondents expressed a more positive attitude toward ridesourcing technology than toward Uber, while 26% of respondents expressed a more positive attitude toward Uber than toward ridesourcing technology. Notable outliers include individuals who reported feeling very positive or neutral toward Uber and/or Lyft, but very negative toward ridesourcing technology in general. This result could be due to people actually holding a higher opinion of Uber than the technology in general, or it could be due to conflations or assumptions that Uber is the technology in question. In either case, Uber’s negative exposure in popular media so far does not appear to have deleterious effects on its public approval. Policy implications When asked what their city should do about Uber and Lyft, respondents largely said regulate, consider forming partnerships, or “nothing” (see Table 12). Only 2% of respondents thought

45 their cities should ban the ridesourcing companies from operation, representing a departure from calls for action from years prior. Respondents were relatively evenly distributed across the remaining categories, though more people thought Lyft should be met with inaction compared to Uber. About one-third of respondents suggested that their city should regulate Uber and Lyft to have the same operational restrictions as taxis, while about two-thirds suggested potentially more affirmative policies such as inaction and forming partnerships to further social goals. People who selected “Other” were largely unsure of what would be a better alternative (40-45%), suggesting the option prompts fell short in helping respondents imagine policy responses. Others said their city should impose ethics/societal impact regulations, background checks, and other safety regulations (e.g. require cars to have cameras and breathalyzers, use background checks to determine criminal history and sex offender status, and conduct regular drug screenings of drivers).

Table 12 Responses to the question “What should your city do about [Uber/Lyft]?” Uber Lyft Responses Percent Responses Percent My city should ban [Uber/Lyft] 6 1.5% 8 2.0% My city should regulate [Uber/Lyft] so that it has 135 34.3% 127 32.2% the same operational restrictions as taxis My city shouldn’t do anything about [Uber/Lyft] 124 31.5% 143 36.3% My city should form partnerships with [Uber/Lyft] 120 30.5% 106 26.9% Other 9 2.3% 10 2.5% n 394 100% 394 100%

Respondents who indicated that they thought their city should form partnerships with Uber or Lyft were asked what kind of partnerships they would support. In line with the high proportion of respondents indicating that they use Uber or Lyft to avoid driving while intoxicated (35%), 76% of the respondents who answered this question said they would support partnerships to reduce instances of drunk driving. 50% said they would support partnerships to supplement access to public transit, and 24% said they were interested in reducing private car ownership or usage. Predictably, respondents who said they used Uber or Lyft to avoid driving while intoxicated were particularly likely to support city partnerships to curb drunk driving (V=0.2009, p=0.0001). 4.1.6 Favorability of aspects of ridesourcing technology Finally, respondents were asked whether certain aspects of ridesourcing technology made them want to use Uber or Lyft more or less. Included in the list were ten attributes associated with the ridesourcing services, such as “automatic payment,” “no automatic tips,” “drivers are independent contractors, not employees,” “rating system,” and “company identity or CEO attitude,” among others (see Fig. 5 and Table 13). Nearly three-quarters of respondents (74%)

46 indicated that ridesourcing being “accessible where transit is sometimes not” made them want to use Uber or Lyft more. In contrast, one in five respondents said that the decision by both Uber and Lyft to label their drivers independent contractors instead of employees made them want to use the services less. This latter observation is in contrast to the 6% of respondents who reported that they don’t use Uber or Lyft due to an ethical or ideological opposition. It appears that while only 6% of respondents feel strongly enough to cite an ethical dilemma as reason to not use the service, as much as 20% of respondents express that one particular dilemma makes them uncomfortable enough to want to use the service less. In line with media coverage about Uber’s acerbic leadership, only 25% of respondents indicated that the company identity or CEO attitude made them want to use ridesourcing more, and 11% said it made them want to use the services less. Less controversial aspects that enjoy wide approval include automatic payment (54% “more”), the rating system (48% “more”), brand trust (52% “more”), and the high-tech nature (48% “more”). The brand trust approval appears high given that customers interact primarily with independent contractors (drivers), suggesting there may be a linkage between this aspect and the rating system, which adds an element of accountability to each interaction with a contractor. The option to save money by sharing a ride with other passengers also has high approval (52% “more”), but has the second-highest disapproval of all the options listed (14% “less”), demonstrating heterogeneity of opinions and suggesting that respondents may have had negative experiences with the carpooling services UberPool and Lyft Line (e.g. not understanding the models or not actually wanting to share).

The following aspects of ride-hailing technology make me want to use Uber or Lyft... [More, Less, or Neutral] (%)

0 20 40 60 80 100

Accessible where transit is sometimes not Automatic payment The option to save money by sharing a ride with other… Brand trust Rating system High-tech Personable driver-passenger relationship No automatic tips Drivers are independent contractors, not employees Company identity or CEO attitude

More Neutral Less

Figure 5. Favorability of certain attributes of the ridesourcing services, Uber and Lyft.

Table 13 Favorability of certain attributes of the ridesourcing services, Uber and Lyft More Neutral Less Accessible where transit is sometimes not 74% 23% 3% Automatic payment 54% 41% 6% The option to save money by sharing a ride with other passengers 53% 34% 14% Brand trust 52% 43% 5% Rating system 48% 46% 6% High-tech 48% 46% 6% Personable driver-passenger relationship 42% 47% 11% No automatic tips 42% 48% 11% Drivers are independent contractors, not employees 34% 47% 20% Company identity or CEO attitude 25% 64% 11% Note. Some categories (rows) do not add to 100 due to rounding.

4.2 Bivariate and Multivariate Analysis

To better understand the relationships between responses to multiple survey questions, I identified a list of 24 key independent and dependent variables. I constructed a large correlation matrix of all the variables and identified 112 statistically significant correlations of potential interest (see Appendix 2). Table 14 below summarizes and describes the variables included in this analysis.

Table 14 Summary of independent and dependent variables Variable name Variable type Range Description Independent Variables Demographics Gender Dichotomous 0 to 1 FemaleMale Age Ordinal 1 to 5 YoungOld Race Categorical/ Nominal 1 to 6 Whiteness Dichotomous 0 to 1 Non-WhiteWhite Education Ordinal 1 to 7 LessMore Income Ordinal 1 to 9 LowHigh Spatial Characteristics Metropolitan area Categorical/ Ordinal 1 to 14 Top 15 in U.S. by pop. (by population) High population Low population Home address type Ordinal 1 to 3 Based on population density of zip code RuralUrban Work address type Ordinal 1 to 3 Based on population density of zip code

RuralUrban Distance from city center Continuous [miles] 3.3 to 156.8 Based on home zip code Transportation Characteristics Travel for work Dichotomous 0 to 1 Different home and work zip codes NoYes Access to car Ordinal 0 to 3 Implicit interpretation Frequency of Uber use Continuous 0.25 to 4 No. of trips/ length of use LowHigh Frequency of Lyft use Continuous No. of trips/ length of use LowHigh Have Uber app Ordinal 0 to 3 NoYes Have Lyft app Ordinal 0 to 3 NoYes Mode choice: Uber Ordinal 1 to 9 LowHigh Mode choice: Lyft Ordinal 1 to 9 LowHigh Dependent Variables User identification Ordinal 0 to 3 NeitherBoth User/ Non-user Dichotomous 0 to 1 Non-userUser Attitude toward Uber Ordinal 1 to 5 NegativePositive Attitude toward Lyft Ordinal 1 to 5 NegativePositive Attitude toward ridesourcing Ordinal 1 to 5 NegativePositive technology in general Policy implications: Uber Ordinal/ Categorical 1 to 4 BanPartnerships Policy implications: Lyft Ordinal/ Categorical 1 to 4 BanPartnerships Note. When used as independent variables in regression analyses, ordinal variables are coded as dummy variables.

4.2.1 Bivariate analysis

Key findings from the correlation analysis are summarized in the following section, organized by dependent variable. For multivariate analysis including discussion of the theoretical framework and logit models, see the next section.

User identification

The ordinal variable “User identification” refers to respondents’ personal identification as a user of neither Uber nor Lyft (given a value of 0), of Lyft only (given a value of 1), of Uber only (given a value of 2), or of both Uber and Lyft (given a value of 3). The variable increases in value as the respondent’s identity as a ridesourcing user strengthens. I tested the associations between this variable and the independent variables listed in Table 12 with the following hypotheses:

1) Spatial and transportation characteristics will correlate with user identification more than demographic characteristics;

2) Young, well-educated, white men with high incomes are more likely to identify as users of ridesourcing (Rayle et al., 2016); and 3) People who live in an urban area and need to travel for work are more likely to identify as users of ridesourcing.

A variable-by-variable analysis using Spearman’s Rho, Cramer’s V, and Pearson’s r showed the strongest relationships between reported usage characteristics and user identification (e.g. unsurprisingly, having the Uber or Lyft app was strongly correlated with identifying as a user) (see Table 15). Several transportation characteristics are related to user identification, but two demographic characteristics—gender and age—are weakly related as well. Specifically, knowing gender and age allows for a reduction in errors in predicting the dependent variable, user identification, by 12% and 17%, respectively. While youth and “maleness” correlate positively with user identification, there is no evidence for a relationship between user identification and race, income, education, or home or work address type. No spatial characteristics are significantly related to user identification.

Needing to travel for work and having access to a car both weakly positively correlate with user identification, though it was expected that having access to a car might negatively correlate with user identification. Knowing whether the respondent’s home and work zip codes differ and the likelihood of their access to a private vehicle allows for a reduction in error in predicting user identification by 14% and 16%, respectively. As expected, there are strong correlations among using Uber and/or Lyft frequently, having the Uber and/or Lyft apps, ranking Uber and/or Lyft highly in mode choice, and identifying as a user of Uber or Lyft.

Table 15 Relationships between independent variables and dependent variable: user identification Independent Variable Coefficient of p-value Expected Actual Association a Direction Direction Gender 0.1219 0.0034 Positive Positive Age -0.1699 0.0007 Negative Negative Travel for work 0.1407 0.0500 Positive Positive Access to car 0.1664 0.0009 Negative Positive Frequency of Uber use 0.2296 0.0001 Positive Positive Frequency of Lyft use 0.1978 0.0117 Positive Positive Have Uber app 0.6469 0.0000 Positive Positive Have Lyft app 0.6073 0.0000 Positive Positive Mode choice: Uber 0.5298 0.0000 Positive Positive Mode choice: Lyft 0.4910 0.0000 Positive Positive Note. a All coefficients are Spearman’s ρ except for “Gender” and “Travel for work,” which are Cramer’s V, and “Frequency” variables, which are Pearson’s r.

Related to the observation that gender and race are correlated with user identification, the following relationships were also observed: gender is correlated with frequency of uber use (rpb = 0.1882, p = 0.0015), and attitude toward Uber (rpb = 0.1542, p = 0.0022); and age is weakly correlated with attitude toward Uber (ρ = -0.0935, p = 0.0637), attitude toward Lyft (ρ = -0.1454, p = 0.0038), and attitude toward ridesourcing technology (ρ = 0.1158, p = 0.0215).

Attitude toward Uber, Lyft, and ridesourcing technology

The ordinal variable “Attitude” is a Likert scale wherein the value ‘1’ corresponds to “Very Negative” attitudes and the value ‘5’ corresponds to “Very Positive.” I made the following hypotheses:

1) User identification, frequency of use, and having the app will correlate positively with attitude (with weaker relationships between independent Uber variables and dependent Lyft variables and vice versa) 2) Mode choice will correlate positively with attitude (i.e. high ranking of Uber and Lyft should correlate with positive attitudes of Uber and Lyft)

As expected, identifying as a user, using Uber and/or Lyft often, having the apps, and ranking the services highly all correlated with attitude (see Table 16). There was no evidence of a relationship between frequency of Uber use and attitude toward Lyft, nor between frequency of Lyft use and attitude toward Uber. Further, some demographic variables demonstrated a relationship with attitude, but as they have been shown to have a relationship with user identification (employed here as an independent variable) in Table 15, I elected to not include them in Table 16. Finally, attitude toward ridesourcing technology in general (i.e. irrespective of Uber and Lyft) was not included in the table due to its having very similar results to attitude toward Uber (as alluded to in the univariate analysis section above).

Table 16 Relationships between independent variables and dependent variable: attitude toward Uber (Lyft) Independent Variable Coefficient of p-value Expected Actual Association a Direction Direction User identification 0.4818 0.0000 Positive Positive (0.5616) (0.0000) Frequency of Uber use 0.2254 0.0001 Positive Positive 0.0634 (0.2880) (No relationship) Frequency of Lyft use 0.0044 0.9556 Positive No relationship (0.2268) (0.0037) (Positive) Have the Uber app 0.4595 0.0000 Positive Positive 0.3552 (0.0000) Have the Lyft app 0.2436 0.0000 Positive Positive (0.5156) (0.0000)

Mode choice: Uber 0.5578 0.0000 Positive Positive (0.2661) (0.0000) Mode choice: Lyft 0.2493 0.0000 Positive Positive (0.5441) (0.0000) Note. a All coefficients are Spearman’s ρ except for “Frequency” variables, which are Pearson’s r.

Policy implications

Responses to the questions “What should your city do about Uber?” and “What should your city do about Lyft?” are referred to as policy implications. These variables are categorical and semi- ordinal: the associated values range from negative implications (i.e. ban) to positive implications (i.e. form partnerships), but the in-between categories (i.e. “regulate” and “do nothing”) do not have an explicit hierarchy. Nonetheless, I tested the association between this dependent variable and the two other dependent variables, user identification and attitude, which are standing in as independent variables for the purpose of this analysis.16 I hypothesized that identifying as a user of ridesourcing and having a positive attitude toward ridesourcing would correlate with “positive” policy implications, and the results corroborated this hypothesis (see Table 17). Knowing respondents’ user identification and attitudes allowed for a reduction in errors in predicting policy implication responses by between 12% and 24%.

Relationships observed in the correlation analysis but not included in the table are related to demographics, having the app, and mode choice, all of which have been shown to have relationships with user identification and attitude, which were included in this table.

Table 17 Relationships between independent variables and dependent variable: policy implications for Uber (Lyft) Coefficient of Expected Actual Independent Variable p-value Association a Direction Direction User identification 0.1509 0.0027 Positive Positive (0.1514) (0.0026) Attitude toward Uber 0.2427 0.0000 Positive Positive (0.2068) (0.0000) Attitude toward Lyft 0.1245 0.0134 Positive Positive (0.1958) (0.0001) Attitude toward ridesourcing 0.2335 0.0000 Positive Positive tech. (0.2440) (0.0000) Note. a All coefficients are Spearman’s ρ

16 For a more rigorous analysis of this dependent variable, see the multinomial logit model detailed in the following section.

Additional observations Beyond the relationships discussed in the first half of this section, a number of other inter- variable relationships were observed in the correlation analysis. They are largely predictable, and therefore do not require detailed description. Demographic observations, for example, include relationships between age and education, age and income, gender and income, race and education, race and income, education and income, and race and access to a car. There are also statistically significant relationships between home and work address type; distance from city center and home address type; frequency of Uber use and Lyft use; attitudes toward Uber, Lyft, and ridesourcing technology; and policy implications for Uber and Lyft. No statistically significant relationship was observed between population density (“urbanness”) of home address and having access to a car, nor between needing to travel for work and having access to a car. A weak positive correlation, however, was observed between taking Uber or Lyft to avoid drunk driving and supporting city partnerships with Uber or Lyft to curb drunk driving (V=0.2009, p=0.0001). 4.2.2 Multivariate analysis Bivariate analyses allow for the conclusion that demographic variables and transportation characteristics are related to whether an individual identifies as a user of Uber, Lyft, both, or neither. They also show that people who use Uber and Lyft tend to have more positive attitudes towards them and toward ridesourcing technology in general, and that these attitudes are closely related to the types of policy interventions that individuals support in their home cities. Building on the hypotheses discussed in the previous section, three key hypotheses dictate the models employed in this logistic regression analysis. Each hypothesis corresponds to a specific logit model, summarized below:

 Model 1: Logit o User identification as a function of demographics, spatial characteristics, and transportation characteristics o Hypothesis: Spatial and transportation characteristics will determine user identification better than demographic information  Model 2: Ordered Logit o Attitude as a function of frequency of use, user identification, having the app, and mode ranking o Hypothesis: Regularity of use will determine attitude  Model 3: Multinomial Logit (MNL) o Policy implications as a function of attitude, frequency of use, and mode choice o Hypothesis: Attitude and use will determine policy implications

Model 1 In the first model, the dichotomous User/Non-User designation serves as the dependent variable, while demographic information and spatial and transportation-related characteristics are included as independent variables. Regression analyses of each independent variable category alone (e.g. demographic predictors gender, age, race, education, and income only) showed that only the gender, age (at three levels), and education (at three levels) coefficients were significant among the demographic variables. Similarly, only the travel for work coefficient was significant among the transportation characteristic variables, and none of the spatial characteristic variables were significant.17 As such, the final model includes only gender, age, education, and travel for work as independent variables. The results of the logistic regression show that the model as a whole fits significantly better than the null model (likelihood ratio chi-square of 60.52 with a p-value of <0.0001), and a small amount of variation is explained by the model, which has a Pseudo R2 value of 0.1233. Table 18 below summarizes the coefficients found for each independent variable in the logistic regression.

Table 18 Summary of logistic regression analysis of determinants of user identification Variable Log odds (SE) Odds ratio (SE) Gender (Base: Female) Male 0.74 (0.25)*** 2.11* Age (Base: 18 to 24) 25 to 34 0.23 (0.37) 1.26 35 to 44 -0.71 (0.38)* 0.49* 45 to 54 -0.80 (0.41)* 0.45* 55 or older -1.16 (0.42)*** 0.31*** Education (Base: HS or less) Some college 1.32 (0.36)*** 3.73*** 2-year degree 0.69 (0.40)* 1.99* 4-year degree 1.38 (0.38)*** 3.99*** Prof. degree 0.10 (0.38) 1.10 Doctorate 1.09 (0.93) 3.00 Travel for work Yes 0.47 (0.25)* 1.60* LL -215.13 LR chi2 (14) 60.52 Pseudo R2 0.1233 p <0.0001 Note. *p<0.10, **p<0.05, ***p<0.01

17 Some independent variables (e.g. app and mode choice) are statistically significant when serving as the only predictor or in combination with one other predictor, but are not statistically significant when included with multiple other predictors.

The model shows that being male increases the odds of identifying as a ridesourcing user by a factor of 2.11, and that each incremental level of education completed between some college and a bachelor’s degree increases the odds of identifying as a ridesourcing user by a factor of between 2 and 4. Being between 35 and 44 years of age as opposed to 18 to 24 years of age decreases the odds of identifying as a ridesourcing user by a factor of 0.49 (being 45 to 54 decreases the odds by a factor of 0.45, and being 55 or older decreases the odds by a factor of 0.31). Finally, needing to travel for work (i.e. listing different home and work zip codes) increases the odds of identifying as a ridesourcing user by a factor of 1.6. In sum, though the bivariate analysis shows that respondents’ transportation usage characteristics have the most significant relationships with the ordinal user identification variable, demographics are a much better predictor of user identification in this model than spatial or transportation characteristics. Model 2 The second model employs ordered logit analysis to understand factors that determine variation in reported attitude, an ordinal variable on a Likert scale (from 1, very negative, to 5, very positive). The three attitude responses—toward Uber, toward Lyft, and toward ridesourcing technology in general—serve as the dependent variables. The independent variable, user identification, is joined by usage controls such as mode ranking and having the app, as well as by demographic controls including gender and education to establish any direct effects. A first- round iteration demonstrated that other demographic controls were not significant, and that frequency of use was only significant when replacing mode choice as a predictor. Including variables that previous iterations showed to be not significant (e.g. ‘Mode choice: Lyft,’ having the Uber app, and frequency of Uber use in the Attitude toward Uber model, and the metropolitan area of residence, access to car, and population density in all three models) negatively affects the statistical significance of the explanatory variables; therefore, variables that are not statistically significant are not included in the final models presented in Table 19 below.

Table 19 Summary of ordered logit analysis of determinants of attitude toward ridesourcing services Attitude toward Attitude toward Attitude toward Variable ridesourcing Uber (SE) Lyft (SE) technology (SE) User (Base: Neither Uber nor Lyft) Lyft only -0.69 (0.86) 4.33 (1.20)*** 3.86 (1.15)*** Uber only 2.19 (0.31)*** 0.93 (0.30)*** 0.92 (0.32)*** Both Uber and Lyft 1.68 (0.32)*** 2.71 (0.35)*** 1.07 (0.36)*** Mode choice: Uber 0.40 (0.06)*** 0.19 (0.06)*** Mode choice: Lyft 0.39 (0.07)*** Have the Uber app 0.39 (0.14)*** Gender (Base: Female)

Male 0.35 (0.20)* -0.29 (0.22) -0.02 (0.20) Education (Base: HS or less) Some college -0.62 (0.31)** -0.55 (0.33)* -0.21 (0.29) 2-year degree -0.34 (0.36) -0.21 (0.39) -0.26 (0.35) 4-year degree -0.54 (0.31)* -0.29 (0.33) -0.01 (0.29) Prof. degree -0.97 (0.33)*** -0.41 (0.37) -0.26 (0.32) Doctorate 0.24 (0.93) -0.23 (0.88) 0.37 (0.81) LL -421.47 -363.20 -465.54 LR chi2 216.00 219.66 129.09 Pseudo R2 0.2040 0.2322 0.1218 p <0.0001 <0.0001 <0.0001 Note. *p<0.10, **p<0.05, ***p<0.01

The likelihood ratio chi-square ranges from 129.09 to 219.66, and each model is statistically significant compared to the null models. The Pseudo R2 ranges from 0.1218 to 0.2322, demonstrating that a small to moderate amount of variation is explained by each model. Direct effects of demographics (gender and education) are observed for the attitude toward Uber dependent variable, but not for attitude toward Lyft or ridesourcing technology in general.

The models show that identifying as a user of both ridesourcing services as opposed to neither results in a 1.68 unit increase in the ordered log-odds of having a more positive attitude toward Uber, while holding the other variables in the model constant. It also results in a 2.71 unit increase in the ordered log-odds of having a more positive attitude toward Lyft, and a 1.07 unit increase in the ordered log-odds of having a more positive attitude toward ridesourcing technology in general (when the other variables in the model are held constant). Being male as opposed to female results in a 0.35 unit increase in the ordered log-odds of having a more positive attitude toward Uber, and being more educated generally results in minor decreases. Regarding the transportation characteristic controls, a one-unit increase in mode ranking of Uber results in a 0.40 unit increase in the ordered log-odds of having a more positive attitude toward Uber. A one-unit difference in the ranking of Uber or Lyft in a respondent’s mode choice is associated with a 0.19 and 0.39 unit difference in attitude toward ridesourcing technology and toward Lyft, respectively. While the usage characteristic variables did act as predictors of attitude, the user identification variable allowed for log odds predictions of a greater magnitude.

Note that the ordinal variables user, attitude toward Uber, and attitude toward Lyft, are also significant predictors of attitude toward ridesourcing technology. Model 3 In the last model, policy implications for Uber serve as the dependent variable, while attitude toward Uber, frequency of Uber use, and mode ranking of Uber serve as independent variables. The survey also asked respondents to weigh in on how their city should respond to Lyft, but the only statistically significant predictor of proposed policy intervention toward Lyft is attitude

56 toward Uber, so the model focuses on policy implications for Uber alone. Table 20 below summarizes three different models considered for this analysis.

Table 20 Summary of multinomial logit analysis of determinants of policy implications for Uber (each model shows the coefficients associated with ‘Form Partnerships’ against the base case ‘Regulate’) Variable Model A (SE) Model B (SE) Model C (SE) Attitude toward Uber (Base: Very negative) Somewhat negative 0.79 (0.31)** 0.75 (0.34)** 0.64 (0.32)** Neutral/ ambivalent 2.08 (.41)*** 2.03 (0.62)*** 1.70 (0.45)*** Somewhat positive 1.80 (0.53)*** 1.48 (0.76)** 1.34 (0.58)** Very positive 18.67 (3468.93) 16.38 (1406.58) 18.19 (3368.08) Frequency of Uber use 0.04 (0.21) Mode choice: Uber 0.15 (0.08)* LL -457.40 -328.21 -451.99 LR chi2 64.67 36.33 75.49 Pseudo R2 0.0660 0.0524 0.0771 p <0.0001 0.0140 <0.0001 Note. *p<0.10, **p<0.05, ***p<0.01

Model A included only attitude toward Uber as a predictor, while Models B and C included frequency of Uber use and mode ranking of Uber, respectively. Demographic variables, user identification, spatial and transportation characteristics are not significant predictors of policy implications. All three models were statistically significant, but Model B included a warning that standard errors may be questionable. Each model explains a very small amount of variation: the Pseudo R2 value ranges from 0.0524 for Model B to 0.0771 to Model C. Model B also shows that frequency of Uber use is not a significant predictor of the policy interventions a respondent will support. Attitude toward Uber and mode ranking of Uber, however, are significant predictors of policy implications. Model A, for example, shows that the relative log odds of a respondent thinking their city should form partnerships instead of regulating ridesourcing services like taxis will increase by 2.08 if moving from a very negative attitude toward Uber to a neutral attitude toward Uber.

This series of models shows that demographic information can predict user identification, which can in turn predict attitude, which can in turn predict policy implications. This modeling emphasizes the importance of evaluating who is using ridesourcing and who it is impacting. There are some direct effects from demographic characteristics on attitude toward Uber, but direct effects are otherwise not observed in downstream models. A more comprehensive analysis of latent influences would involve a structural equation model and/or factor analysis, which are beyond the scope of this thesis.

5. Discussion

Uber and Lyft represent innovation in the way people access mobility. The platform used to connect riders and drivers distinguishes these companies from taxi providers—the transportation mode they most directly replace—and thereby called regulatory requirements into question in their first years of operation. State and local regulatory agencies have since responded to the growing mobility shift engendered by Uber and Lyft, but through exploratory and dynamic processes that have not yet coalesced into a typical or national-level strategy. Policymaking is influenced by media coverage, personal experiences, emerging literature, and occasional local studies, but ridesourcing companies still maintain a significant asymmetry of information that allows them to control much of what we know about how and why people use their services. Given the potential social impacts—both positive and negative—of ridesourcing on urban mobility patterns, it is imperative to determine the legitimate role of government in managing this new model and the evolving economy of which it is a part. This section highlights some key findings from the study and relates them back to the research questions posed at the beginning of the thesis. It recounts the impressions and reactions of a handful of local policymakers who reviewed the results, and concludes by offering avenues for future study.

5.1 Key Findings

Respondents tended to be younger, better educated, and higher earning than the average population. This finding corroborates a previous study that suggested that ridesourcing users were younger and of higher socio-economic status (Rayle et al., 2016). Respondents were about as white as the population (of US adults in the top 15 metropolitan statistical areas by population), and they were also more female than average, though the study also showed that maleness correlates positively with user identification. This association might be related to the typical profile of a tech-savvy millennial (well-educated white male).

69% of respondents identified as users of ridesourcing and 31% identified as non-users. Previous studies have not sought to capture the breakdown between users and non-users, though some have attempted to relative quantify market share. Uber’s valuation compared to Lyft’s suggested that it would capture a greater proportion of riders, and the results showed this was true: 32% of respondents said they used both, but 35% said they used Uber only while just 1% said they used Lyft only.

Ridesourcing is used primarily for special purpose trips, like avoiding driving while intoxicated and getting to or from the airport. Ridesourcing does not appear to offer a functional alternative to commuting to work or school, a market served heavily by private cars and public transit. Rather, it offers a supplement to these regular trips and an alternative to unsafe driving behaviors. This finding aligns with suggestions in the literature that ridesourcing may

58 reduce individual reliance on private vehicles and have public safety benefits, but further research will need to be done on whether the convenience induces people to make car trips and on whether Uber and Lyft are practicing due diligence in ensuring their drivers are safe, sober, and well-rested.

Only a small percentage of respondents identified an ethical opposition that prevented them from using ridesourcing, but many more said the driver-as-contractor issue made them want to use Uber or Lyft less. 6% of respondents said they didn’t use Uber or Lyft (or both) due to an ethical or ideological opposition, which included the companies’ choice to treat drivers as contractors rather than employees, the lack of regulatory oversight, the use of private cars in general, the corporate attitude, and the overarching societal implications. Of the 16 respondents that constitute this 6% figure, 9 identified as non-users of ridesourcing, and the other 7 identified as users of Uber only. Respondents were evenly distributed across gender, age, race and income categories, and were clustered on the lower end of the education spectrum. 6% may be low compared to what media coverage has suggested, but it still may be high compared to average ethical opposition to other transport modes or other companies. At the end of the survey, 20% of respondents said the driver-as-contractor issue made them want to use Uber or Lyft less, the highest disapproval of the ten aspects of ridesourcing technology presented to participants. These findings suggest that ethical dilemmas are only strong enough to prevent a small portion of people from using Uber and Lyft, but are significant enough to make 1 in 5 people uncomfortable.

According to public perception, Uber is ridesourcing technology. The results analysis showed that people responded almost identically to the question about their attitude toward ridesourcing technology as they did to the question about their attitude toward Uber. The bivariate analysis corroborated this observation by demonstrating that many of the relationships observed between independent variables and attitude toward Uber could also be observed with attitude toward ridesourcing technology in general. These findings suggest either a greater approval of the Uber brand than of the generic technology, or a conflation of the two, wherein Uber is the common term for ridesourcing much in the same way that Americans refer to adhesive bandages as Band-Aids and facial tissues as Kleenex. Uber might even offer a stand-in term for the sharing or access economy, as suggested by a 2015 article about queueing posted on Salon.com entitled “The Uber-ization of Everything” (Grabar).

5.2 Discussion of Research Questions

Most of the key findings enumerated above touch upon or directly answer the research questions posed at the beginning of this thesis, but I will return to those questions now to offer some concluding remarks. 1) When and why do people use Uber and Lyft? What makes an individual choose one over the other if the business models are so similar?

People use Uber and Lyft for special purpose trips, and to fill gaps in transit accessibility. They choose these services because they are more convenient, cheaper, and faster than alternate modes. Those who don’t use them say that they’re still more expensive than alternate modes, that they’re skeptical of the level of safety, and that they have no need because they can use their own car. In some cases, the company identity and can sway people toward Lyft because of its friendlier marketing narrative, but many more people use Uber; it’s what their friends use, or it’s the only ridesourcing service they’ve ever considered using. This network effect seems to attract a greater number of drivers, decreasing wait times for passengers and creating a positive feedback loop. 2) How do individuals perceive Uber compared to Lyft, and compared to ridesourcing technology in general? How can these attitudes potentially inform policymaking?

Despite negative media coverage that might suggest otherwise, people have positive attitudes toward Uber and relatively neutral attitudes toward Lyft. Instead of reserving their best judgment for the technology on its own, people seem unable or unwilling to consider the technology as something distinct from Uber. 3) Do the ethical and ideological oppositions to Uber and Lyft featured in popular media coverage appear in individuals’ reasoning about whether or not to use the services? As discussed above, 6% of respondents said their ethical or ideological opposition to Uber or Lyft was strong enough to keep them from using the service. However, 20% of respondents later said that the driver-as-contractor issue—one of the more controversial ethical questions up for discussion at present—made them want to use Uber or Lyft less. 4) Is there potential support for cities to make directed partnerships with Uber and Lyft to further their socially-driven goals? More than a quarter of respondents explicitly support their city forming partnerships with Uber or Lyft to reduce instances of drunk driving, to supplement access to public transit, and to reduce private car ownership or usage. To develop and benefit from directed partnerships, jurisdictions may leverage the aspects of ridesourcing that have strong public support (e.g. supplementing transit accessibility) to make meaningful progress on the aspects that have high disapproval (e.g. the issue of drivers working as contractors rather than employees with benefits). For regulations to be legitimate and defensible, policymakers need more information on ridesourcing operations as well as constituent perspectives. While this study reports only national-level trends, the methodology may prove useful to policymakers as they attempt to fill these information gaps.

5.3 Policymaker Impressions

This section recounts impressions and reactions collected from short interviews and conversations with policymakers in Cambridge, MA, including Craig Kelley of the Cambridge City Council, and Aaron Jette, a policy analyst at the Volpe National Transportation Systems Center. The Cambridge License Commission declined to comment, citing the political sensitivity

60 of the present regulatory climate. Representatives from the Cambridge Department of Traffic, Parking, and Transportation could not be reached for comment. The sources of the perspectives discussed below have been abstracted by request of interviewees. The interviewees all began our conversations by emphasizing their tangential role to the policymaking process regarding ridesourcing in Cambridge. They expressed diverse insights, and one described their role as “supporting policy decision-making,” but agreed that the city is currently awaiting guidance from the state level, and that a regulatory decision regarding the operations of Uber, Lyft, and any future ridesourcing company should be made by the conclusion of the 2016 legislative session (ends July 31). Though the Cambridge License Commission declined to comment on nearly every question in the interview, they did suggest agreement with this framing and said they were looking for ways to regulate transportation network companies.

Policymakers framed the policy issue as a question about the continuation of the taxi industry in the face of industry disruption, as a question of consumer protection (including, perhaps, Uber and Lyft drivers), and as a question about a new mobility solution. There was consensus on the improvements offered by Uber and Lyft (e.g. performance and affordability), and on the importance of striking the right balance in regulation between supporting these improvements for the sake of consumer welfare and addressing concerns of safety, accessibility, and social equity. Expanding on these concerns, interviewees described the potential loss of street-hail service and accessibility of for-hire transportation for people who don’t have smartphones and who have specific vehicle or app accessibility needs; they described Uber’s market dominance and potential to achieve a monopoly; and they described apprehension about implicit racial and ethnic bias in the user and driver profiles, rating system, and tipping functions. Policymakers have made explicit and implicit references to basing impressions (if not necessarily decisions) on a combination of media coverage and personal experiences, suggesting a need for more rigorous studies. They want to understand why cities respond the way they have and what makes Uber and Lyft decide to leave a city (e.g. as was the case with Austin in early May 2016). They cite the most pressing questions to be about whether ridesourcing will increase or decrease overall VMT and congestion, whether it will act as a substitute or a complement to public transit, and whether it will engender mode shifts between private cars, public transit, and ridesourcing. More generally, they also cite questions about who is using and being impacted by ridesourcing, and how cities should respond to changing technology and to the evolving access economy. These needs and questions point to a role for social science research in making ridesourcing policy, a conclusion the interviewees all shared. Finally, multiple interviewees cited the lack of funding or data affiliation with Uber or Lyft as a contributing factor to the study’s legitimacy. Despite its broad focus and limitation to reporting national trends, this study and others like it should help policymakers address the information asymmetry in both volume and quality of input from Uber and Lyft.

5.4 Limitations and Areas for Further Research

Since the start of this study, media coverage and popular discussion have taken a turn toward some of the social impact questions alluded to earlier: about collective bargaining rights, which Seattle won for its drivers in December 2015; about implicit racial bias between riders and drivers, which Uber used as an argument against including a tipping function in their app in late April 2016; and about the appropriate extent of driver background checks, which led to Uber and Lyft suspending operations in Austin in early May 2016. And on May 13, 2016, I took my very first ride with Fasten, a Boston-based startup that has been quietly edging into the market since Sept. 2015 and that accelerated their expansion to Austin in the vacuum left by Uber and Lyft the previous weekend. Conducting relevant research necessarily means staying on top of the current discussions in popular media, legislative proceedings, and marketing done by both Uber and Lyft. That said, remaining relevant exposes research to criticisms of transience and over-ambition in focus. This thesis suffered from the latter; in attempting to answer a bevy of questions, the survey achieved surface-level answers without conducting the necessary follow-up to determine the conclusiveness of the findings. As a structured questionnaire on usage and perceptions of ridesourcing, it is informative, but future studies would benefit from first developing a narrower topic of focus (e.g. actual use of ridesourcing and justification, costs and benefits to consumer safety, potential labor force implications, etc.). Below are some specific and general recommendations for future research. Specific suggestions

 Explicitly incorporate private car as a transportation mode into the questionnaire (e.g. ask respondents about their access to a private car and their usage of it).

 Ask respondents to specify their level of transit accessibility (e.g. number of minutes to the nearest public transit station on foot, if they need to use another mode to access transit).

 If not conducting local intercept surveys, develop a trip scenario or several to ask respondents about, which should allow for direct questions about modes used for specific purposes, alternative modes, and potential mode shifts.

 Regarding policy implications, ask respondents about their perception of potential impacts of TNCs on public transit, and about surge pricing.

 Complement questionnaire with formal semi-structured interviews of policymakers before and after the survey is distributed.

General recommendations

 Design a methodology that includes factor analysis or a structural equation model to elucidate latent factors in user identification and attitude, and/or an organizational decision model to capture citywide attitudes and decision-making processes.

 Develop a survey that allows for a more direct translation from reported usage and attitudes to policy (e.g. an academic study that addresses implicit bias and driver rights in the question of whether Uber should or should not include a tipping function in its app).

 Explore the principles that people draw upon when identifying as users or non-users of ridesourcing. Do people base this identification on amount or frequency of use, brand loyalty, or some combination?

 Finally, ask how Uber and Lyft might fill the paratransit role that has been discussed in the literature as an actual and potential function of taxis for at least three decades (Cervero, 1997; Gilbert and Samuels, 1982). Do policymakers have an opportunity to claim a part in a changing urban mobility paradigm? Can Uber and Lyft and their successors provide transport options to the public in suburbs and transit deserts? That Uber is ranked comparably to transit in the mode choice question in a way taxis are not, and that the most highly-approved aspect of ridesourcing is its accessibility where transit is sometimes not, suggest that these might be some of the most compelling questions moving forward.

References

Anderson, D. N. (2014). “Not just a taxi”? For-profit ridesharing, driver strategies, and VMT. Transportation, 41(5), 1099-1117. doi:10.1007/s11116-014-9531-8

AP Stylebook. (2015). AP Style tip: Ride-hailing or ride-booking services let people use smartphone apps to book and pay for car service. Do not use ride-sharing. Retrieved December 15, 2015, from https://twitter.com/apstylebook/status/565515953430364163

Austin, D., & Zegras, P. (2012). Taxicabs as Public Transportation in Boston, Massachusetts. Transportation Research Record: Journal of the Transportation Research Board, 2277, 65-74. doi:10.3141/2277-08

Ben-Akiva, M. E., & Lerman, S. R. (1985). Discrete choice analysis: Theory and application to travel demand. Cambridge, MA: MIT Press.

Bilton, N. (2014, November 25). The Slippery Slope of Silicon Valley. Retrieved from http://www.nytimes.com/2014/11/27/style/uber-facebook-and-others-bedeviled-by- moral-issues.html?_r=1

Botsman, R. (2013, November 21). The Sharing Economy Lacks A Shared Definition. Retrieved from http://www.fastcoexist.com/3022028/the-sharing-economy-lacks-a-shared- definition#3

Bouton, S., Knupfer, S. M., Mihov, I., & Swartz, S. (2015, September). Urban mobility at a tipping point. Retrieved from http://www.mckinsey.com/business- functions/sustainability-and-resource-productivity/our-insights/urban-mobility-at-a- tipping-point

Buhr, S. (2015, September 16). China’s Didi Kuaidi Put $100M Into Lyft, Inks Ridesharing Alliance To Rival Uber. Retrieved from http://techcrunch.com/2015/09/16/ubers-rivals- didi-kuadi-and-lyft-form-international-ridesharing-partnership/

California Public Utilities Commission. (2013, July 30). Proposed Decision of Commissioner Peevey, Agenda ID #12291. Retrieved from http://docs.cpuc.ca.gov/PublishedDocs/Published/G000/M077/K112/77112285.PDF

Carson, B. (2016, April 11). The global alliance against Uber just came online. Retrieved from http://www.businessinsider.com/lyft-didi-ola-grab-anti-uber-alliance-launches-2016-4

Cherry, J. (2015, Spring). The Missing Link At Uber: Ethics, Innovation, & Conscious Capitalism. Retrieved from http://www.consciouscompanymagazine.com/blogs/press/22766337-the-missing-link-at- uber-ethics-innovation-conscious-capitalism

City of New York, Office of the Mayor. (2016, January). For-Hire Vehicle Transportation Study (Rep.). Retrieved http://www1.nyc.gov/assets/operations/downloads/pdf/For-Hire- Vehicle-Transportation-Study.pdf

Corey, Canapary & Galanis Research. (2014). Travel Decisions Survey 2014 Summary Report (Rep.). CA: San Francisco Municipal Transportation Authority.

Correll, D. (2015, February 23). Three Jevons Paradoxes for the Future of Sustainable Supply Chain Management – and One Way to Resolve Them All. Retrieved from http://www.csrwire.com/blog/posts/1524-three-jevons-paradoxes-for-the-future-of- sustainable-supply-chain-management-and-one-way-to-resolve-them-all

Cramer, J., & Krueger, A. B. (2016). Disruptive Change in the Taxi Business: The Case of Uber. American Economic Review, 106(5), 177-182. doi:10.1257/aer.p20161002

Dempsey, P. S. (1996) Taxi Industry Regulation, Deregulation, and Reregulation: The Paradox of Market Failure. University of Denver College of Law, Transportation Law Journal, 24(1), 73-120.

Eckhardt, G. M., & Bardhi, F. (2015, January 28). The Sharing Economy Isn't About Sharing at All. Retrieved from https://hbr.org/2015/01/the-sharing-economy-isnt-about-sharing-at- all

File, T., & Ryan, C. (2014, November). Computer and Internet Use in the United States: 2013 (Rep.). Retrieved from http://www.census.gov/content/dam/Census/library/publications/2014/acs/acs-28.pdf

Flegenheimer, M., & Fitzsimmons, E. G. (2015, July 16). City Hall and Uber Clash in Struggle Over New York Streets. Retrieved from http://www.nytimes.com/2015/07/17/nyregion/city-hall-and-uber-clash-in-struggle-over- new-york-streets.html

Flegenheimer, M. (2015, July 22). De Blasio Administration Dropping Plan for Uber Cap, for Now. Retrieved from http://www.nytimes.com/2015/07/23/nyregion/de-blasio- administration-dropping-plan-for-uber-cap-for-now.html?_r=0

Flores, O. & Rayle, L. (2016). How Ridesourcing went from ‘Rogue’ to Mainstream in San Francisco. Manuscript in preparation.

Fritz, C. (2014, December 13). Mobility-as-a-service: Turning transportation into a software industry. Retrieved from http://venturebeat.com/2014/12/13/mobility-as-a- service-turning-transportation-into-a-software-industry/

Gilbert, G., & Samuels, R. E. (1982). The taxicab: An urban transportation survivor. Chapel Hill: University of North Carolina Press.

Giridharadas, A. (2012, June 1). Is Technology Fostering a Race to the Bottom? Retrieved from http://www.nytimes.com/2012/06/02/us/02iht-currents02.html?_r=1

Grabar, H. (2015, January 10). The Uber-ization of everything. Retrieved from http://www.salon.com/2015/01/10/the_uber_ization_of_everything/

Graves, R. (2010, October 25). Uber’s Cease & Desist. Retrieved from https://newsroom.uber.com/ubers-cease-desist/

Hall, J. V. & Krueger, A. B. (2015, January). An Analysis of the Labor Market for Uber’s Driver-Partners in the United States. Working Papers (Princeton University. Industrial Relations Section); 587. Retrieved from http://arks.princeton.edu/ark:/88435/dsp010z708z67d

Hazard, C. (2014, August 6). Uber and Lyft win temporary state approval. Retrieved from http://www.richmond.com/business/uber-and-lyft-win-temporary-state- approval/article_fd28f360-1d82-11e4-8847-001a4bcf6878.html

HB 4075. Gen. Assemb. (Ill., 2014). Retrieved from http://www.ilga.gov/legislation/fulltext.asp?DocName=09800HB4075ham001&GA=98& SessionId=85&DocTypeId=HB&LegID=77989&DocNum=4075&GAID=12&Session

Hodges, G. R. (2007). Taxi!: A social history of the New York City cabdriver. Baltimore: Johns Hopkins University Press.

Huston, C. (2016, January 28). Lyft takes a friendlier road than Uber. Retrieved from http://www.marketwatch.com/story/lyft-takes-a-friendlier-road-than-uber-2016-01-27

Kenney, M. & Zysman, J. (2015). Choosing a Future in the Platform Economy: The Implications and Consequences of Digital Platforms. Manuscript in preparation.

King, D. A., Peters, J. R., & Daus, M. W. (2012, November 14). Taxicabs for Improved Urban Mobility: Are We Missing an Opportunity? TRB 2012 Annual Meeting. Retrieved from http://amonline.trb.org/12-2097-1.2377303?qr=1

Lawler, R. (2014, March 17). Seattle Deals A Blow To Uber And Lyft By Limiting The Number Of Ride-Sharing Drivers On The Road. Retrieved from http://techcrunch.com/2014/03/17/seattle-hates-ride-sharing/

Lien, T. (2016, January 5). Lyft defies predictions by continuing to grow as a rival to Uber. Retrieved from http://www.latimes.com/business/technology/la-fi-0105-lyft-growth- 20160105-story.html

Lobel, O. (2015). The Law of the Platform. Manuscript in preparation.

Meelen, T., & Frenken, K. (2015, January 14). Stop Saying Uber Is Part Of The Sharing Economy. Retrieved from http://www.fastcoexist.com/3040863/stop-saying-uber-is-part- of-the-sharing-economy

Miller, D. (2015, May 24). Lyft vs. Uber: Just How Dominant Is Uber in the Ridesharing Business? -- The Motley Fool. Retrieved from http://www.fool.com/investing/general/2015/05/24/lyft-vs-uber-just-how-dominant-is- uber-ridesharing.aspx

Morning Consult. (2015, June 31). National Tracking Poll #150505: Crosstabulation Results (Rep.). Retrieved https://morningconsult.com/wp- content/uploads/2015/06/150505_crosstabs_mc_v2_AD.pdf

Nisen, M. (2013, May 09). How Nike Solved Its Sweatshop Problem. Retrieved from http://www.businessinsider.com/how-nike-solved-its-sweatshop-problem-2013-5

Novak, M. (2013, June 25). Sorry Uber, Los Angeles Has Been Banning Ride-Shares For a Century. Retrieved from http://paleofuture.gizmodo.com/sorry-uber-los-angeles-has- been-banning-ride-shares-fo-574851806

Perez, L. (2016, January 5). The Uber controversy reveals the rottenness of the taxi industry. Retrieved from http://www.marxist.com/uber-controversy-reveals-rottenness-of-taxi- industry.htm

Picker, L., & Isaac, M. (2015, November 17). Lyft Said to Seek $500 Million in Funding. Retrieved from http://www.nytimes.com/2015/11/18/business/dealbook/lyft-said-to-seek- dollar500-million-in-funding.html?_r=2

Price Waterhouse. (1993). Analysis of Taxicab Deregulation and Re-Regulation I,(6). Retrieved from https://www.colorado.gov/pacific/sites/default/files/0708TransAttachI.pdf

Punit, I. S. (2015, December 03). The new global anti-Uber alliance: Ola, Lyft, Didi Kuaidi and GrabTaxi agree to ride together. Retrieved from http://qz.com/564795/a-new-anti-uber- alliance-strengthens-ola-lyft-didi-kuaidi-and-grabtaxi-agree-to-ride-together/

Qualtrics. (2010, July 23). Finding Respondents. Retrieved March 15, 2016, from https://www.qualtrics.com/blog/finding-respondents/

Qualtrics. (2014, June 20). 28 Questions to Help Research Buyers of Online Samples. Thesis survey launch through Qualtrics - purchasing respondents [E-mail to N. Richard]. (2016, March 16). Quintal, B. (2015, November 25). Smart Moves for Cities: The Urban Mobility Revolution Will Start With These 3 Projects. Retrieved from http://www.archdaily.com/777791/smart- moves-for-cities-the-urban-mobility-revolution-will-start-with-these-3-projects

Rayle, L., Dai, D., Chan, N., Cervero, R., & Shaheen, S. (2016). Just a better taxi? A survey- based comparison of taxis, transit, and ridesourcing services in San Francisco. Transport Policy, 45, 168-178. doi:10.1016/j.tranpol.2015.10.004

Reindl, J. C. (2014, May 16). Uber and Lyft are deemed legal in Detroit but not in Ann Arbor. Retrieved from http://paddcertified.org/uber-and-lyft-are-deemed-legal-in-detroit-but- not-in-ann-arbor/

Rodrigue, J., Comtois, C., & Slack, B. (2013). The geography of transport systems (3rd ed.). New York: Routledge.

Rogers, B. (2015). The Social Costs of Uber. The University of Chicago Law Review. doi:10.2139/ssrn.2608017

Rubinstein, D. (2014, October 30). Uber, Lyft, and the end of taxi history. Retrieved from http://www.capitalnewyork.com/article/city-hall/2014/10/8555191/uber-lyft-and-end- taxi-history

Russell, J. (2014, November 19). Uber's Moral Compass Needs Recalibration. Retrieved from http://techcrunch.com/2014/11/19/uber-off/

Shahani, A. (2016, January 18). In The Battle Between Lyft And Uber, The Focus Is On Drivers. Retrieved from http://www.npr.org/sections/alltechconsidered/2016/01/18/463473462/is- uber-good-to-drivers-it-s-relative

Shapiro, C. (2012, December 4). What's The Future Of The Sharing Economy? Retrieved from http://www.fastcoexist.com/1681009/whats-the-future-of-the-sharing-economy

Smart, R., Rowe, B., Hawken, A., Kleiman, M., Mladenovic, N., Gehred, P., & Manning, C. (2015, July). Faster and Cheaper: How Ride-Sourcing Fills a Gap in Low-Income Los Angeles Neighborhoods (Rep.).

Smith, A. (2015, April 01). U.S. Smartphone Use in 2015. Retrieved from http://www.pewinternet.org/2015/04/01/us-smartphone-use-in-2015/

Teal, R. F., & Berglund, M. (1987). The impacts of taxicab deregulation in the USA. Journal of Transport Economics and Policy, 21(1), 37-56. Retrieved from http://www.bath.ac.uk/e- journals/jtep/pdf/Volume_XX1_No_1_37-56.pdf

Uber Technologies. (2015, July 14). Introducing uberASSIST. Retrieved from https://newsroom.uber.com/us-california/introducing-uberassist-la/

United States Census Bureau. (2015). Annual Estimates of the Resident Population Estimates. Retrieved from http://factfinder.census.gov/faces/tableservices/jsf/pages/productview.xhtml?src=bkmk

United States Census Bureau. (2010). 2010 Census. Retrieved from http://factfinder.census.gov/faces/tableservices/jsf/pages/productview.xhtml?pid=ACS_1 4_5YR_DP05&src=pt

Waller, T. (2015, January 29). Uber now legal in South Carolina, PSC decides. Retrieved from http://www.wyff4.com/news/uber-now-legal-in-south-carolina-psc-decides/30992456

Wallman, B. (2014, November 15). Broward goes after Uber underground taxis. Retrieved from http://www.sun-sentinel.com/local/broward/fl-uber-broward-conflict-20141115- story.html

Weinstein, B. (2014, November 21). Opinion: Four other ways Uber is ethically challenged. Retrieved from http://money.cnn.com/2014/11/21/technology/uber-ethics-oped/

Appendices

Appendix 1: Survey questionnaire and recode values

This appendix lists the questions shown to survey participants. Changes to the recode values are shown in bold.

Individual Perceptions of Ride-Hailing Apps

Q1 Informed Consent. Thank you for participating in the “Individual Perceptions of Ride-Hailing Apps” survey. The survey is being conducted by researchers at MIT, and the information you provide will only be used for academic purposes. Participation in this survey is voluntary, and it should take less than 10 minutes to complete. You may decline further participation in the survey at any time without adverse consequences. If you choose to participate in the survey you understand that your responses to the survey questions will be stored and accessed by the researcher. However, any responses you give will remain anonymous and confidential, and all data will only be reported in an aggregate format (by reporting only combined results and never reporting original ones). All questionnaires will be concealed, and no one other than then primary investigator and assistant researches listed below will have access to them. The data collected will be stored in the HIPPA-compliant, Qualtrics-secure database until it has been deleted by the primary investigator.

Q2 I have read and understood the above consent form and desire of my own free will to participate in this study.  Yes (1)  No (2) If No Is Selected, Then Skip To End of Block

Q34 Please select the metropolitan area in which you live. New York-Newark-Jersey City (1) Los Angeles-Long Beach-Anaheim (2) Chicago-Naperville-Elgin (3) Dallas-Fort Worth-Arlington (4) Houston-The Woodlands-Sugar Land (5) Philadelphia-Camden-Wilmington (6) Washington, DC-Arlington-Alexandria (7) Miami-Fort Lauderdale-West Palm Beach (8) Atlanta-Sandy Springs-Roswell (9) Boston-Cambridge-Newton (10) San Francisco-Oakland-Hayward (11) Phoenix-Mesa-Scottsdale (12) Riverside-San Bernardino-Ontario (13) Detroit-Warren-Dearborn (14) Seattle-Tacoma-Bellevue (15) San Diego-Carlsbad (16) New Orleans-Metairie (17) Baltimore-Columbia-Towson (18) Pittsburgh (19) Austin-Round Rock (20) I live more than 20 miles outside of one of these metropolitan areas (21) I live in or near a metropolitan area not on this list (22)

If San Diego-Carlsbad Is Selected, Then Skip To End of Block If New Orleans-Metairie Is Selected, Then Skip To End of Block If Baltimore-Columbia-Towson Is Selected, Then Skip To End of Block If Pittsburgh Is Selected, Then Skip To End of Block If Austin-Round Rock Is Selected, Then Skip To End of Block If I live more than 20 miles o... Is Selected, Then Skip To End of Block If I live in or near a metropo... Is Selected, Then Skip To End of Block

Q3 Transportation Network Companies (TNCs), also called ride-hailing apps, use mobile platforms (i.e. smartphone apps) to connect passengers seeking rides with drivers who use their own personal vehicles. The most common examples of TNCs are Uber and Lyft. This survey will ask you to distinguish between your feelings about Uber, Lyft, and TNC/ride-hailing technology irrespective of those two companies.

Q4 Do you consider yourself to be a user of:  Uber only (1) (2)  Lyft only (2) (1)  Both Uber and Lyft (3) (3)  Neither Uber nor Lyft (4) (0) If Neither Uber nor Lyft Is Selected, Then Skip To As a non-user of Uber, Lyft, or both,...

Q5 When do you use Uber or Lyft? (You may select up to three primary purposes)  Getting to or from the airport (1)  Getting somewhere in bad weather (2)  Getting somewhere at times when public transit is unavailable (e.g. not running or too crowded) (3)  Getting to or from locations not accessible by public transit (4)  Avoiding driving while intoxicated (5)  For running errands or shopping (6)  Getting to work or school (7)  For social/leisure purposes (e.g. bar, restaurant, concert, visiting friends and family, etc.) (8)  Other (please specify) (9) ______

Q6 When you decide to use Uber or Lyft as opposed to other modes (e.g. in the situations you selected above), what is your primary reason for doing so? (You may select up to three reasons)  It is cheaper (1)  It is faster (2)  It is more convenient (3)  It is safer (4)  It makes me feel modern (5)  My friends do it (6)  Other (please specify) (7) ______

Answer If Do you consider yourself to be a user of: Both Uber and Lyft Is Not Selected Q7 As a non-user of Uber, Lyft, or both, what is your primary reason for not using Uber or Lyft as opposed to other modes? (You may select up to three reasons)  It's expensive (1)  It's slow (2)  It's inconvenient (3)  It's unsafe (4)  I have an ideological or ethical opposition to it (5)  Other (please specify) (6) ______

Answer If As a non-user of Uber, Lyft, or both, what is your primary reason for not using Uber or Lyft as opposed to other modes? (You may select up to three reasons) I have an ideological or ethical opposition to it Is Selected Q8 What is the source of your ideological or ethical opposition to Uber or Lyft?  I am opposed to the corporate attitude (1)  I disagree with the choice to treat drivers as contractors and not employees (2)  I am opposed to the use of private cars as a mode of transportation in general (3)  I think one or both companies enjoy a lack of regulatory oversight that is unfair to consumers and/or to the taxi industry (4)  Other (please specify) (5) ______

Q9 Have you ever taken an Uber?  Yes, and I have initiated at least one ride myself. (1)  Yes, but someone else booked or paid for the ride for me. (2)  No. (3) If No. Is Selected, Then Skip To Do you have the Uber app installed on...

Q11 How many times have you taken an Uber?  Between 1 and 3 times (1)  Between 4 and 10 times (2)  Between 10 and 100 times (3)  More than 100 times (4)

Q12 About how long ago was your first Uber ride?  Less than one month (1)  Between a month and a year (2)  Between one and three years (3)  Between three and six years (4)

Q13 Do you have the Uber app installed on your smartphone?  Yes, I currently have the app installed on my phone. (1) (3)  No, I do not have the app installed on my phone and I never have. (2) (1)  No, I do not have the app installed on my phone, but I did once. (3) (2)  No, I don't have a smartphone. (4) (0)

Q14 Have you ever taken a Lyft?  Yes, and I have initiated at least one ride myself. (1)  Yes, but someone else booked or paid for the ride for me. (2)  No. (3) If No. Is Selected, Then Skip To Do you have the Lyft app installed on...

Q15 How many times have you taken a Lyft? Between 1 and 3 times (1) Between 4 and 10 times (2) Between 10 and 100 times (3) More than 100 times (4)

Q16 About how long ago was your first Lyft ride? Less than one month (1) Between a month and a year (2)

Between one and three years (3) Between three and six years (4)

Q17 Do you have the Lyft app installed on your smartphone? Yes, I currently have the app installed on my phone. (1) (3) No, I do not have the app installed on my phone and I never have. (2) (1) No, I do not have the app installed on my phone, but I did once. (3) (2) No, I don't have a smartphone. (4) (0)

Q18 Between Uber and Lyft, if you prefer one over the other, how do the following factors influence your preference? For each row, please select the company you believe to be better, or select 'Neutral' if you believe them to be about equal. Neutral (the companies Uber is better (1) Lyft is better (3) are about equal) (2) Cost (1)    Number of drivers available, and thereby    wait time (2) Safety (3)    The company's attitude or    identity (4) It's what my friends use, or it's the only one I've    ever considered using (5) Technical or aesthetic elements of the app or    technology (6) Promotions and coupons    (7) Please select neutral for    this line (8) If Please select neutral for t... Is Not Selected, Then Skip To End of Block

Q19 Consider the trips you make on a regular or semi-regular basis (e.g. to work or school, to run errands, to socialize, to access entertainment, etc.). Please rank the travel modes listed below according to the frequency with which you use them ('1' for the mode you use for most trips, '2' for the mode you use when your primary mode is not available or appropriate, and so on). ______Uber (1) ______Lyft (2) ______Your own car or one that you rent/borrow (3) ______Carpooling (4) ______Public transit or commuter rail (5) ______Taxi (6) ______Walking (7) ______Bicycle (8) ______Other (please specify) (9)

Q20 What is your priority when choosing between modes of transportation for your daily activities?  Travel time (1)  Cost (2)  What is available to you (3)  Level of productivity while traveling (4)  Other/ some combination (please specify) (5) ______

Q21 How would you describe your attitude toward Uber?  Very positive (1) (5)  Somewhat positive (2) (4)  Neutral/ambivalent (3) (3)  Somewhat negative (4) (2)  Very negative (5) (1)

Q22 How would you describe your attitude toward Lyft?  Very positive (1) (5)  Somewhat positive (2) (4)  Neutral/ambivalent (3) (3)  Somewhat negative (4) (2)  Very negative (5) (1)

Q23 How would you describe your attitude toward ride-hailing technology in general?  Very positive (1) (5)  Somewhat positive (2) (4)  Neutral/ambivalent (3) (3)  Somewhat negative (4) (2)  Very negative (5) (1)

Q35 Please select somewhat negative for this question.  Very positive (1)  Somewhat positive (2)  Neutral/ambivalent (3)  Somewhat negative (4)  Very negative (5) If Somewhat negative Is Not Selected, Then Skip To End of Block

Q24 How should your city respond to Uber?  My city should ban Uber (1)  My city should regulate Uber so that it has the same operational restrictions as taxis (2)  My city shouldn't do anything about Uber (3)  My city should consider forming partnerships with Uber (e.g. to supplement public transit, or to discourage private car ownership) (4)  Other (please specify) (5) ______

Q25 How should your city respond to Lyft?  My city should ban Lyft (1)  My city should regulate Lyft so that it has the same operational restrictions as taxis (2)  My city shouldn't do anything about Lyft (3)  My city should form partnerships with Lyft (e.g. to supplement public transit, or to discourage car ownership) (4)  Other (please specify) (5) ______

Q26 If you think your city should form partnerships with Uber or Lyft, what kind of partnerships would you support? (Check all that apply)  To reduce instances of drunk driving (1)  To reduce private car ownership or usage (2)  To supplement access to public transportation (3)  Other (please specify) (4) ______

Q27 The following aspects of ride-hailing technology make me want to use Uber or Lyft... More (1) Neutral (2) Less (3) Automatic payment (1)    No automatic tips (2)    Drivers are independent contractors, not    employees (3) Personable driver-    passenger relationship (4) Rating system (5)    The option to save money by sharing a ride with    other passengers (6) Accessible where transit    is sometimes not (7) Company identity or CEO    attitude (8) Brand trust (9)    High-tech (10)   

Q28 What is your gender?  Female (1) (0)  Male (2) (1)  Prefer not to answer (3) (2)

Q29 What is your age?  18 to 24 (1) (1)  25 to 34 (2) (2)  35 to 44 (3) (3)  45 to 54 (4) (4)  55 to 64 (5) (5)  65 or older (6) (5)  Prefer not to answer (7) (6)

Q30 Which best describes your race and/or ethnicity?  Native American or Alaska Native (1)  Asian or Pacific Islander (2)  Hispanic or Latino/a (3)  Black or African American (4)  White (5)  Multiple races and/or ethnicities (6)  Prefer not to answer (7)

Q31 What is the highest level of education that you have completed?  Less than high school (1) (1)  High school graduate or GED (2) (1)  Some college (3) (2)  2 year degree (4) (3)  4 year degree (5) (4)  Professional degree (6) (5)  Doctorate (7) (6)  Prefer not to answer (8) (7)

Q32 What is your approximate average household income?  $0-$24,999 (1)  $25,000-$49,999 (2)  $50,000-$74,999 (3)  $75,000-$99,999 (4)  $100,000-$124,999 (5)  $125,000-$149,999 (6)  $150,000-$174,999 (7)  $175,000-$199,999 (8)  $200,000 or higher (9)  Prefer not to answer (10)

Q34 In what zip code do you live?

Q35 In what zip code do you work?

Appendix 2: Correlations and regression model outputs

Bivariate analysis correlation matrix (significant associations are highlighted)

Model 1: Logistic regression logit user_dich i.DEMO_GEN i.DEMO_AGE i.DEMO_ED i.travel_work

Iteration 0: log likelihood = -245.39342 Iteration 1: log likelihood = -215.92599 Iteration 2: log likelihood = -215.13472 Iteration 3: log likelihood = -215.13375 Iteration 4: log likelihood = -215.13375

Logistic regression Number of obs = 394 LR chi2(14) = 60.52 Prob > chi2 = 0.0000 Log likelihood = -215.13375 Pseudo R2 = 0.1233

. logit, or

Logistic regression Number of obs = 394 LR chi2(14) = 60.52 Prob > chi2 = 0.0000 Log likelihood = -215.13375 Pseudo R2 = 0.1233

| travel_work | Travel for work | 1.598701 .393161 1.91 0.056 .9872681 2.588804 _cons | .9431294 .3059293 -0.18 0.857 .4994146 1.781071 ------

Model 2: Ordered logit

______**ATTITUDE TOWARD UBER** ______ologit ATT_UBER i.user i.DEMO_GEN i.DEMO_ED MOD_UBER

Iteration 0: log likelihood = -529.47484 Iteration 1: log likelihood = -427.58374 Iteration 2: log likelihood = -421.50983 Iteration 3: log likelihood = -421.47324 Iteration 4: log likelihood = -421.47323

Ordered logistic regression Number of obs = 394 LR chi2(12) = 216.00 Prob > chi2 = 0.0000 Log likelihood = -421.47323 Pseudo R2 = 0.2040

**ATTITUDE TOWARD LYFT** ______ologit ATT_LYFT i.user MOD_LYFT i.DEMO_GEN i.DEMO_ED

Iteration 0: log likelihood = -473.03276 Iteration 1: log likelihood = -370.32345 Iteration 2: log likelihood = -363.30187 Iteration 3: log likelihood = -363.2032 Iteration 4: log likelihood = -363.20298 Iteration 5: log likelihood = -363.20298

Ordered logistic regression Number of obs = 394 LR chi2(12) = 219.66 Prob > chi2 = 0.0000 Log likelihood = -363.20298 Pseudo R2 = 0.2322

------ATT_LYFT | Coef. Std. Err. z P>|z| [95% Conf. Interval] ------+------user | Lyft only | 4.325124 1.19808 3.61 0.000 1.97693 6.673318 Uber only | .9292802 .2988581 3.11 0.002 .3435292 1.515031 Both Uber and Lyft | 2.711624 .3497812 7.75 0.000 2.026065 3.397182 | MOD_LYFT | .3863994 .0669609 5.77 0.000 .2551584 .5176404 | DEMO_GEN | Male | -.2864883 .2197433 -1.30 0.192 -.7171772 .1442006 Prefer not to answer | .300677 1.949424 0.15 0.877 -3.520124 4.121478 | DEMO_ED | Some college | -.5519205 .3345713 -1.65 0.099 -1.207668 .1038271 2yr degree | -.2071818 .3864162 -0.54 0.592 -.9645436 .55018 4yr degree | -.2934638 .3317525 -0.88 0.376 -.9436866 .3567591 Prof. degree | -.4119704 .3720327 -1.11 0.268 -1.141141 .3172004 Doctorate | -.2334081 .8764682 -0.27 0.790 -1.951254 1.484438 Prefer not to answer | -3.046999 1.695109 -1.80 0.072 -6.369351 .2753537 ------+------/cut1 | -2.416379 .4862434 -3.369398 -1.463359 /cut2 | -1.075766 .3835065 -1.827425 -.3241066 /cut3 | 3.048985 .4176936 2.230321 3.867649 /cut4 | 4.990131 .4694993 4.069929 5.910332 ------______**ATTITUDE TOWARD RIDESOURCING TECHNOLOGY** ______ologit ATT_RHT i.user uber_app MOD_UBER i.DEMO_GEN i.DEMO_ED

Iteration 0: log likelihood = -530.0819 Iteration 1: log likelihood = -467.22901 Iteration 2: log likelihood = -465.53672 Iteration 3: log likelihood = -465.53579 Iteration 4: log likelihood = -465.53579

Ordered logistic regression Number of obs = 394 LR chi2(13) = 129.09 Prob > chi2 = 0.0000 Log likelihood = -465.53579 Pseudo R2 = 0.1218

/cut4 | 3.642336 .4033268 2.85183 4.432842 ------

Model 3: Multinomial logit

______**MODEL A** ______mlogit city_uber i.att_uber

Iteration 0: log likelihood = -489.73066 Iteration 1: log likelihood = -467.42729 Iteration 2: log likelihood = -458.44328 Iteration 3: log likelihood = -457.72489 Iteration 4: log likelihood = -457.46195 Iteration 5: log likelihood = -457.41186 Iteration 6: log likelihood = -457.39999 Iteration 7: log likelihood = -457.39771 Iteration 8: log likelihood = -457.3972 Iteration 9: log likelihood = -457.39709 Iteration 10: log likelihood = -457.39706

Multinomial logistic regression Number of obs = 394 LR chi2(16) = 64.67 Prob > chi2 = 0.0000 Log likelihood = -457.39706 Pseudo R2 = 0.0660

______**MODEL B** ______mlogit city_uber i.att_uber freq_uber, base(2)

Iteration 0: log likelihood = -346.3806 Iteration 1: log likelihood = -330.50887 Iteration 2: log likelihood = -328.68338 Iteration 3: log likelihood = -328.31229 Iteration 4: log likelihood = -328.23463 Iteration 5: log likelihood = -328.21869 Iteration 6: log likelihood = -328.215 Iteration 7: log likelihood = -328.21411 Iteration 8: log likelihood = -328.21393 Iteration 9: log likelihood = -328.2139

Multinomial logistic regression Number of obs = 283 LR chi2(20) = 36.33 Prob > chi2 = 0.0140 Log likelihood = -328.2139 Pseudo R2 = 0.0524

------city_uber | Coef. Std. Err. z P>|z| [95% Conf. Interval] ------+------Ban | att_uber | Somewhat positive | -14.34237 553.4611 -0.03 0.979 -1099.106 1070.421 Neutral/ambivalent | -.8431871 1.184784 -0.71 0.477 -3.165321 1.478947 Somewhat negative | -14.89865 1645.839 -0.01 0.993 -3240.683 3210.886 Very negative | -16.86488 6283.185 -0.00 0.998 -12331.68 12297.95 | freq_uber | -.3706635 .7331245 -0.51 0.613 -1.807561 1.066234 _cons | -1.46023 .989583 -1.48 0.140 -3.399777 .4793167 ------+------Regulate | (base outcome) ------+------Nothing | att_uber | Somewhat positive | -.4475801 .351882 -1.27 0.203 -1.137256 .2420959 Neutral/ambivalent | -.6524616 .4780018 -1.36 0.172 -1.589328 .2844047 Somewhat negative | -.7894796 .70452 -1.12 0.262 -2.170313 .5913542 Very negative | -16.08919 1482.563 -0.01 0.991 -2921.86 2889.682 | freq_uber | .0878154 .2092921 0.42 0.675 -.3223896 .4980204 _cons | .323929 .3729102 0.87 0.385 -.4069614 1.054819 ------+------Partnerships | att_uber | Somewhat positive | -.7451836 .3399205 -2.19 0.028 -1.411416 -.0789516 Neutral/ambivalent | -2.029096 .617139 -3.29 0.001 -3.238666 -.8195258 Somewhat negative | -1.480277 .7563612 -1.96 0.050 -2.962718 .0021634 Very negative | -16.38189 1406.584 -0.01 0.991 -2773.236 2740.472 | freq_uber | -.035952 .2119105 -0.17 0.865 -.451289 .379385 _cons | .8113768 .3616684 2.24 0.025 .1025198 1.520234 ------+------Other | att_uber | Somewhat positive | 13.58655 563.5871 0.02 0.981 -1091.024 1118.197 Neutral/ambivalent | 13.38981 563.5878 0.02 0.981 -1091.222 1118.002 Somewhat negative | -.7444491 1928.177 -0.00 1.000 -3779.902 3778.413 Very negative | -2.73506 7053.343 -0.00 1.000 -13827.03 13821.56 | freq_uber | .1781658 .6624752 0.27 0.788 -1.120262 1.476593 _cons | -16.20902 563.5876 -0.03 0.977 -1120.82 1088.402 ------Note: 3 observations completely determined. Standard errors questionable. ______**MODEL C** ______

mlogit city_uber i.att_uber mod_uber

Iteration 0: log likelihood = -489.73066 Iteration 1: log likelihood = -462.7514 Iteration 2: log likelihood = -453.16125 Iteration 3: log likelihood = -452.33976 Iteration 4: log likelihood = -452.05421 Iteration 5: log likelihood = -452.00372 Iteration 6: log likelihood = -451.99048 Iteration 7: log likelihood = -451.98815 Iteration 8: log likelihood = -451.98762 Iteration 9: log likelihood = -451.98749 Iteration 10: log likelihood = -451.98746

Multinomial logistic regression Number of obs = 394 LR chi2(20) = 75.49 Prob > chi2 = 0.0000 Log likelihood = -451.98746 Pseudo R2 = 0.0771

------city_uber | Coef. Std. Err. z P>|z| [95% Conf. Interval] ------+------Ban | att_uber | Somewhat positive | -15.08006 717.9434 -0.02 0.983 -1422.223 1392.063 Neutral/ambivalent | -2.009635 1.269842 -1.58 0.114 -4.49848 .4792109 Somewhat negative | -1.183559 1.340395 -0.88 0.377 -3.810684 1.443567 Very negative | -19.44257 17122.08 -0.00 0.999 -33578.11 33539.22 | mod_uber | .0970324 .2240345 0.43 0.665 -.3420672 .536132 _cons | -2.292379 .9929775 -2.31 0.021 -4.238579 -.3461786 ------+------Regulate | (base outcome) ------+------Nothing | att_uber | Somewhat positive | -.3604575 .3377968 -1.07 0.286 -1.022527 .3016121 Neutral/ambivalent | -.1380265 .3862587 -0.36 0.721 -.8950797 .6190266 Somewhat negative | -.7287561 .5609387 -1.30 0.194 -1.828176 .3706636 Very negative | -1.509915 1.150784 -1.31 0.189 -3.765412 .7455807 | mod_uber | -.1862947 .0728701 -2.56 0.011 -.3291175 -.0434719 _cons | 1.000576 .3329317 3.01 0.003 .3480417 1.65311 ------+------Partnerships | att_uber | Somewhat positive | -.6362031 .3235331 -1.97 0.049 -1.270316 -.0020898 Neutral/ambivalent | -1.697995 .4510239 -3.76 0.000 -2.581986 -.8140046 Somewhat negative | -1.337469 .577647 -2.32 0.021 -2.469636 -.2053018 Very negative | -18.19188 3368.084 -0.01 0.996 -6619.515 6583.131 | mod_uber | -.1477786 .0759394 -1.95 0.052 -.296617 .0010598 _cons | 1.224449 .333327 3.67 0.000 .5711403 1.877758 ------+------Other | att_uber | Somewhat positive | 13.11221 564.2671 0.02 0.981 -1092.831 1119.055 Neutral/ambivalent | 13.31086 564.2671 0.02 0.981 -1092.632 1119.254 Somewhat negative | -1.813983 1371.076 -0.00 0.999 -2689.074 2685.446 Very negative | 13.69512 564.2681 0.02 0.981 -1092.25 1119.64 | mod_uber | .2486536 .2246022 1.11 0.268 -.1915586 .6888657 _cons | -16.97044 564.2676 -0.03 0.976 -1122.915 1088.974 ------

Appendix 3: COUHES approval