USING CROWDSOURCED AND OPEN DATA TO EXPLORE EMERGING MOBILITY TRENDS: A NEW YORK METROPOLITAN AREA CASE STUDY By LESLIE BROWN A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF URBAN AND REGIONAL PLANNING UNIVERSITY OF FLORIDA 2018
2018 Leslie Brown
To my m om
4 ACKNOWLEDGMENTS I would like to thank my mom, who taught me the importance of working for a greater good, t he value of human and natural environments and to ride a bicycle and drive a car when I was barely tall enough to reach the pedals of either All of these have been formative in my life and career, and I cannot thank her enough I cherish my memories of her and know she would be proud to see me come this far. I would also like to thank my dad for his steadfast support through the years, and my wife Jill, who persuaded me to move to Florida with her and pursue the Master of Urban and Regional Planning degree I had been considering for years. I owe her a big debt of gratitude for her support, her shining example as a working University of Florida law student, and her tireless love and encouragement through the ups and downs of graduate school and beyond. I would also like to thank Dr. Ruth St einer, who mentored me and supported me greatly during my time at the University of F lorida. This has been a very rewarding experience and I have grown much under her guidance. I would also like to thank her for chairing my thesis committee and for helping me develop this research. Additionally, I would like to thank Dr. Siva Srinivasan fo r serving as a member on my thesis committee and for helping bring focus to my research, and Dr. Kari Watkins of Georgia Tech for her input and for kindly agreeing to serve on my committee as a special member Finally, I would like to acknowledge Dr. Paul Zwick for his valuable input during the conceptual stages of this research.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS .................................................................................................. 4 LIST OF TABLES ............................................................................................................ 8 LIST OF FIGURES .......................................................................................................... 9 LIST OF ABBREVIATIONS ........................................................................................... 10 ABSTRACT ................................................................................................................... 11 CHAPTER 1 INTRODUCTI ON .................................................................................................... 13 Research Question ................................................................................................. 16 Study Area .............................................................................................................. 18 2 LITERATURE REVIEW .......................................................................................... 19 Transportation and Housing Context for the New York Metropolitan Area .............. 19 Overview of Transportation Network Companies .................................................... 22 New York City Market Penetration ................................................................... 23 Composition of Trip Types ................................................................................ 23 Trip Volume by Time of Day and Time of Week ............................................... 24 User Demographics .......................................................................................... 24 Impact of Transportation Network Companies ........................................................ 26 Travel Time Savings and Increased Mobility .................................................... 27 Reductions in Drunk Driving and Parking Need ............................................... 27 Other Purported Social Benefits ....................................................................... 28 Increased Congestion and Vehicle Miles Traveled ........................................... 29 Induced travel demand .............................................................................. 30 Low efficiency and miles traveled without passengers ............................... 31 Critique of Business Practices .......................................................................... 31 Exemption from supply restrictions ............................................................ 32 Data reciprocity and transparency ............................................................. 32 Longterm viability of price structures ........................................................ 34 Equity concerns ......................................................................................... 35 Impact of Transportation Network Companies on Other Modes ............................. 36 Driving and Ownership of Personal Vehicles .................................................... 36 Public Transit Service ....................................................................................... 37 Increased transit accessibility .................................................................... 37 Increased heavy rail rider ship .................................................................... 38 Public private partnerships ......................................................................... 38 Decreased bus and light rail ridership ........................................................ 38 Transit quality as a moderating variable .................................................... 39
6 Promise and Shortcomings of Dynamic Ridesharing .............................................. 40 Overview of Shared Autonomous Vehicles ............................................................. 41 Terminology and Assumptions ......................................................................... 41 Deployment Trajectories ................................................................................... 42 Potential Impacts of Shared Autonomous Vehicles ................................................ 43 Congestion and Vehicle Miles Traveled ........................................................... 44 First/Last Mile Transit Connection .................................................................... 45 Infill Development ............................................................................................. 46 Changes to the Street Hierarchy ...................................................................... 47 Transit Competition and Ridership Attrition ...................................................... 47 Shared Autonomous Vehicle Sentiment ................................................................. 48 Economic Factors and Willingness to Pay ........................................................ 48 Influence of Cognitive Factors and Modality Styles .......................................... 49 Demographic Profile of Likely Users ................................................................ 52 Policy Responses ................................................................................................... 52 Congestion Pricing and RideHailing Surcharge .............................................. 52 Transit Improvements ....................................................................................... 54 Structural Issues ............................................................................................... 56 Evolution of Public Sector Transportation......................................................... 57 Significance of Shared Autonomous Vehicle Deployment in New York City ........... 60 Preparing for the Impact of Shared Autonomous Vehicles ..................................... 62 3 METHODOLOGY ................................................................................................... 65 Summary of Methodology ....................................................................................... 66 Amazon Mechanical Turk ....................................................................................... 67 Pilot Study ........................................................................................................ 69 Screening Criteria Considerations .................................................................... 71 Geographic Information System Analysis ............................................................... 73 Delineation of Study Subareas ......................................................................... 73 Crowdsource Reporter Application ................................................................... 76 Crowdsource Reporter Pilot Study ................................................................... 78 Qualitative Analysis of Open Data Sets ............................................................ 79 Distribution of trip origins and destinations in New York State ................... 79 Distribution of Uber pickups and Airbnb hotspots in New York City ........... 80 Design of the Survey Instrument ............................................................................. 82 Survey Branching Method ................................................................................ 82 Survey Design and Display Logic ..................................................................... 82 Survey Distribution .................................................................................................. 84 Target Sample Size .......................................................................................... 84 Distribution of Survey Instrument Using TurkPrime .......................................... 85 4 RESULTS ............................................................................................................... 95 Respondent Demographics .................................................................................... 95 Time Since Last use of RideHailing and Other Sharing Economy Platforms ......... 95 Impact of RideHailing on Other Modes and Reasons for Transit Substitution ....... 96
7 Willingness to Use Autonomous RideHailing ......................................................... 96 Willingness to Pay for TNC Surcharge, DRS, and Autonomous RideHailing ......... 97 Crowdsource Reporter Analysis ............................................................................. 98 Substitution by Mode and Trip Type During L Train Closure .................................. 99 First/Last Mile Shuttle Use and Support for Commuter Rail Infill Development .... 100 TNC Surcharge and Congestion Pricing Support Among CBD Workers/Residents ............................................................................................ 101 5 DISCUSSION ....................................................................................................... 115 Ride Hailing Impacts ............................................................................................. 115 Ride Hailing Countermeasures ............................................................................. 118 Dynamic Ridesharing and Autonomous RideHailing ........................................... 120 L Train Substitution by Mode and Trip Type During Planned Closure .................. 123 First Last Mile Transit Connections and Support for Infill Development ............... 125 Limitations ............................................................................................................. 126 Recommendations ................................................................................................ 128 Reinforcing the Subway and Embracing Mobility as a Service ....................... 128 Implementing Congestion Charging to Reduce Traffic and Fund Transit ....... 131 Use Shared Mobility to Open Opportunities for Affordable Housing Infill ....... 132 6 CONCLUSION ...................................................................................................... 134 APPENDIX: SURVEY INSTRUMENT ......................................................................... 137 LIST OF REFERENCES ............................................................................................. 149 BIOGRAPHICAL SKETCH .......................................................................................... 154
8 LIST OF TABLES Table page 2 1 Transit/Ride Hailing Company Pilots and Public Private Partnerships. .............. 64 3 1 Summary of Comparisons Across Sample Sources. Adapted from Kees et al., 2017, p. 151. ................................................................................................. 87 3 2 Primary study area groups/subgroups by transit access thresholds and size. ... 88 3 3 Survey questions by subject, inherited logic, ZIP Code group, and source. ....... 89 4 1 Respondent Demographics: NYC Transit Zone and All New York State. ......... 102 4 2 Shared mobility and sharing economy (time since last use). ............................ 103 4 3 Impact of ridehailing on other modes. ............................................................. 103 4 4 Reasons for substituting ridehailing for transit use. ......................................... 104 4 5 Willingness to pay for TNC surcharge, DRS and Autonomous RideHailing .... 104 4 6 Variation in willingness to use SAVs depending on time of day. ...................... 104 4 7 TNC use, mode shift, and likely SAV users by subway walk distance. ............. 105 4 8 L Train substitution by mode and trip type. ....................................................... 105 4 9 Existing means of traveling to commuter rail stations. ...................................... 105 4 10 Willingness to pay for FLM shuttles to subway or commuter rail stations. ........ 106 4 11 Extent of support for infill development at commuter rail stations. .................... 106 4 12 Extent of support for TNC surcharge and congestion pricing (CBD). ............... 106
9 LIST OF FIGURES Figure page 3 1 Location of MTurk pilot study respondents relative to US population density. .... 90 3 2 Map of ZIP Code groups and subgroups in the primary study area. ................... 90 3 3 Crowdsource Reporter application adapted to collect intersection locations nearest to respondents homes. ......................................................................... 91 3 4 Distribution of Uber pickups and Airbnb Hotspots in New York City. .................. 92 3 5 Distribution of trip origins and destinations in New York State. ........................... 93 3 6 Elements of the survey branching process. ........................................................ 94 3 7 TurkPrime survey description presented to MTurk respondents. ........................ 94 4 1 Modes substituted for with use of ride hailing and ridehailing induced travel demand (232 respondents). ............................................................................. 107 4 2 Willingness to use an autonomous ridehailing serv ice (338 respondents). ..... 107 4 3 Distribution of New York State MTurk respondents relative to major highways and trip origins and d estinations (2006 2010). ............................................... 108 4 4 Distribution of respondents in the New York Metro area by ZIP Code group and subgroup. ................................................................................................... 109 4 5 Location of New York City respondents by subway station walk distance. ....... 110 4 6 New York City respondents who have used ridehailing within the last month by subway walk distance. ................................................................................. 111 4 7 New York City respondents who use the subway less since beginning to use ride hailing by subway walk distance. ............................................................... 112 4 8 New York City respondents likely to use SAVs by subway walk distance. ....... 113 4 9 Distribution of Uber pickups and Airbnb hotspots relative to likely SAV users. 114
10 LIST OF ABBREVIATIONS API Application programming interface AV Autonomous vehicle CBD Central business district DRS D ynamic ridesharing FHV For hire v ehicle FLM First/ last m ile GIS Geographic information system GPS Global Positioning System HIT Human Intelligence Task MaaS Mobility as a service MTA Metropolitan Transportation Authority MTurk Amazon Mechanical Turk LIRR Long Island Rail Road P3 Public private partnership R&D Research and development RPA Regional Plan Association SAE Society of Automotive Engineers International SAV Shared autonomous vehicle TLC New York City Taxi and Limousine Commission TNC Transportation network company V2I Vehicle to infrastructure V2V Vehicle to vehicle VMT Vehicle miles traveled WTP Willingness to pay
11 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Urban and Regional Planning USING CROWDSOURCED AND OPEN DATA TO EXPLORE EMERGING MOBILITY TRENDS: A NEW YORK METROPOLITAN AREA CASE STUDY By LESLIE BROWN May 2018 Chair: Ruth Steiner Major: Urban and Regional Planning The Regional Plan Associations (RPA) landmark Fourth Regional Plan highlights a number of the New York metropolitan area s most pressing issues and proposes a series of policy solutions to address these. Chief among these issues are the deteriorating condition of New York Citys subway system, increased gridlock in the citys Central Business District and chronic affordable housing shortages. Fixing these issues is key to the regions success. The rise of Uber and other ride hailing companies in New York City have coincided with recent declines in subway ridership, which reverses a decades long trend of transit oriented grow th, as well as increased traffic congestion, which some have attributed to these companies. As these firms plan to deploy driverless technologies the impacts of ridehailing are expected to increase. Yet, transit companies have an opportunity to use the efficiency and mobility gains offered by these services and technologies to improve outcomes in each of these problem areas. This study uses RPAs recommendations as a framework to study these issues and emerging ones A novel methodology was developed to survey area residents (n=500) regarding issues both granular and regional in scale. Additionally, a number of
12 open data sets were explored to bring further context to these areas The study addresses issues about the relationship between transit and ride hailing use; willingness of current ridehailing users to embrace emerging technologies and new mobility models; and addresses a host of urban policy issues of immediate and future concern for the New York Metro area.
13 CHAPTER 1 INTRODUCTION Since the 1990s, New York Citys subway system has absorbed much of the citys growing travel demand, reversing an earlier period of autooriented growth (Mahler, 2018). This period of transit oriented growth, which saw the c ity move from an era of post industrial decline to become a powerhouse of the information economy, and which Mahler (2018) contends was driven by earlier investments in the subway system, represents one of the great urban renaissance stories of our modern era (n.p.). However, this transit driven success story has undergone a recent reversal of fortune, recalling the troubles of an earlier era. In 2012, many of the citys subway tunnels were inundated by storm surge from Hurricane Sandy corroding parts of a system that rely on badly aged infrastructure (Nir, 2017) These issues have led to growing system wide delays and safety issues, as well as the planned closure of the citys L Train route for 15months beginning in April 2019. This closure will reroute 400,000 daily riders including many from a number of Brooklyn neighborhoods that have fueled some of the citys fastest and most visible growth a challenge for the city eclipsed in scope in recent history perhaps only by the transportation challenges following the September 11, 2001 terrorist attack and immediately after [Hurricane] Sandy ( Nir, 2017, n.p.). In 2016, the subway witnessed its first ridership decline in decades owing in part to issues like those discussed above (Schaller, 2017). However, the subways chronic maintenance and operations issues paint only half the picture. In the spring of 2011, Uber launched in New York City after its public debut in San Francisco weeks earlier (Uber, n.d.). From 2013 2016, Schaller (2017a) estima tes Uber drivers traveled more than 600 million miles on the city s streets, while passenger travel in the citys for hire
14 vehicle (FHV) fleet which includes Uber, competing ridehailing services like Lyft and the citys yellow cab fleet increased by 52 million passengers who benefit from the numerous mobility benefits offered by these ondemand point to point services that operate at a significant savings over the citys taxi fleet. Coinciding with this rapid growth in ridehailing has been growi ng gridlock. In 2017, the citys central business district (CBD) recorded its slowest traffic speed on record averaging less than 7 miles per hour during the day (Schaller, 2017b). A series of recent studies by Schaller (2017a, 2017b) have blamed much of this recent growth in traffic and part of the decline in subway ridership on TNCs. With Waymos recent deployment of a driverless vehicle fleet on public roads in a Phoenix suburb without safety drivers behind the wheel, and plans from automakers like GM to deploy autonomous ridehailing pilots in cities like San Francisco by 2019, the automation of TNCs may be forthcoming, and the present impacts of ridehailing services may increase in proportion to the expected cost and operational efficiency gains offered by these technologies (Schaller, 2017b) McKinsey predicts that autonomous vehicles are likely to penetrate markets of dense high income cities such as New York City first (RPA, 2017b, p. 21). Research by Krueger, Rashidi and Rose (2016), which suggests y oung, educated multimodals may be among the earliest adopters of shared autonomous vehicles, as well as fierce price competition among the citys ride hailing companies that may drive further innovation, would seem to support McKinseys contention (RPA, 2017b). The timeline of the L Train shutdown, which begins in April of 2019 and will last for 15 months coincides with the expected deployment of urban autonomous ridehailing pilots by the likes of GM and others.
15 Schallers (2017a) analysis of Uber trip data shows rapid growth in ridehailing trips in much of the affected L train catchment area an area known for its high concentration of young, educated multimodals (Nir, 2017). The citys transportation commissioner expects much of the displaced L Train c ommuters to be absorbed by buses and alternate subway lines (Nir, 2017) however, the mobility vacuum this situation creates may induce further ridehailing use and could provide ideal conditions for testing autonomous ridehailing models. In November of 2017, the Regional Plan Association (RPA) a venerated urban research and advocacy group whose work has had significant impact on the New York metropolitan regions transportation decision making for nearly a century, published i ts landmark Fourth Regional Plan In this document, RPA (2017a) observe that the New York metro area is losing population at a faster rate than any other large metro area in the country. RPA (2017a) contend that the deteriorating conditions in the subway, growing gridlock on the regions roads, and the areas perennial housing affordability issues represent serious threats to the regions economic competitiveness and negatively affect quality of life for its residents. Among the many fixes suggested by the Fourth Regional Plan, a few stand out. RPA (2017a) suggest s a $111 billion package of fixes for the subway that address issues of safety, reliability system throughput and extension of service to underserved areas of the city that are highly transit dependent. RPA (2017a) propose part of the funding for this will come from implementing road pricing that will simultaneously discourage excessive ridehailing uses. Yet, past efforts to impl ement congestion charging in the city have failed and ride hailing first/ last mile (FLM) connections may offer a more pragmatic alternative than adding prohibitively
16 expensive track miles to the citys chronically underfunded subway system with its substantial backlog of deferred maintenance needs (Mahler, 2018; Hu, 2017) Others, have proposed implementing a flat surcharge, however, Schaller (2017b) claims that ride hailing demand is relatively inelastic and that a $3 surcharge would do little to dissuade excessive use of these services. To address some of the regions housing affordability issues, RPA (2017c) proposes using partially subsidized autonomous FLM shuttles to replace park and rides for the regions two commuter rail systems: Long Island Rail R oad and Metro North. This would open up a substantial amount of land adjacent to these stations for mixeduse development that could have a dedicated portion of affordable units set aside and use real estate value capture to offset subsidies for autonomous FLM shuttles and other associated costs (RPA, 2017c; Mahler, 2018). However, this would require significant changes to local zoning codes and the support of area residents, who would also need to be willing to use these autonomous shuttles in place of par k and rides. Research Question Using the major themes explored above declining subway conditions, growing congestion, forthcoming vehicle automation, and chronic affordable housing needs and the Fourth Regional Plan recommendations as a research framework, this study reviews the contemporary academic literature and policy papers in each area to identify research gaps, particularly as they pertain to the New York metro area. Gaps identified in this search include: Willingness of current TNC users, as well as those expected to use TNCs as fares become more cost competitive with transit, to use and pay for autonomous ride hailing services assuming they were proven relatively safe and resulted in some degree of cost savings;
17 Concurrent participation of ride hailing users in the wider sharing economy, and when these phenomena coincide spatially, how this might predict areas that will be among the first to embrace shared autonomous vehicles; The extent that ride hailing may be impacting the New York metro areas other modes (e.g., subway ridership) ; The relationship between residential proximity to subway stations and ride hailing use; Extent of support for congestion pricing and/or a ridehailing surcharge among those who would be most impacted (i.e., resi dents of and commuters to the Manhattan CBD); Receptivity of New York City residents currently underserved by transit to use an FLM shuttle and their willingness to pay for this service; Receptivity of metro area residents who live within a 15minute drive of commuter rail stations to use FLM shuttles and their willingness to pay for this service, as well as their willingness to support infill development at the stations nearest to their homes (and if not, their reasoning why). How L Train riders expec t to travel by mode and trip type for the duration of that lines closure; To address these research gaps across varying scales and geographies, a novel geographically targeted survey methodology was devised to reach a population of New York metro area res idents This involved the use of desktop and online geographic information systems (GIS) and the programming of an online survey application in order to survey a population of Amazon Mechanical Turk (MTurk) workers located across the state of New York. The target sample size for this survey is 610. Additionally, open datasets that shed further understanding on some of the above questions are explored and compared qualitatively through proximity analysis to residential locations of respondents captured using the novel crowdsourced online survey methodology. This methodology is explained in detail in subsequent chapters. This is followed by an account of the study results, then a discussion of these results in context of the relevant
18 literature, then concluding with a discussion of the studys advantages and disadvantages as well as areas for future research. Study Area Downstate New York is an unofficial designation for the area that includes New York City, Long Island and the Lower Hudson Valley, and which comprises over half of the New York metropolitan areas population and a majority of the population of New York State. This region was selected for three primary reasons. First, the Downstate New York area is the most transit rich region in the US with le vels of transit use that are significantly higher than other American metropolitan areas. Further, the New York City subway system shares many of the same challenges faced by other dense, transit rich regions. Second, New York City has been the target of a number of recent shared mobility innovations and has seen exponential growth in ridehailing use since these services; RPA (2017b) predict that New York City will be among the first areas impacted by autonomous ridehailing services. Lastly, the Downstate New York region was chosen for practical reasons The MTurk platform allows survey administrators to narrow sample targets by state, but not by region or city (with the exception of Washington, DC) Ideally, the sample would have included residents of New Jersey and Connecticut who live in the New York metro area. However, due to time and resource constraints, only residents of New York State were surveyed.
19 CHAPTER 2 LITERATURE REVIEW Transportation and Housing Context for the New York Metropolitan Area The subway has played a pivotal role in the history of New York City, enabling densities that supported the development of the citys iconic skyscrapers, vibrant street life and the formation of agglomeration economies that turned the city into a center of global finance and culture (Mahler, 2018). In the 1960s and 1970s, as a significant portion of the citys tax base fled for the suburbs and the city grew insolvent, the subway fell into disrepair and disuse. However, policymakers of the era under stood the importance of the subway to the citys economic health and took steps to rebuild the citys ailing subway system (Mahler, 2018; Schaller, 2017a). According to Mahler (2018), it was this investment in the subway that led the citys financial recovery, allowing it to repurpose its built form to meet the needs of the late20thcentry and 21stcentury economies, representing one of the great urban renaissance stories of our modern era (n.p.). With respect to outcomes on the citys transportation sy stem, Schaller (2017a) observes: As transit ridership increased, auto ownership and use stabilized, first in Manhattan and then throughout the city. The result was an historic shift in travel from autooriented growth, which characterized the post World W ar II era, to transitoriented growth. Starting in the 1990s, the bus and subway system absorbed most of the growth in travel in New York City which was generated by growth in population and economic activity.31 By the mid2000s, transit was accounting for not just most but all of the growth in travel citywideAs the city reached all time highs in employment and population, city officials increasingly recognized the importance of continuing to absorb increases in travel through the transit system and by wal king and cycling, which also make efficient use of limited street space. (17) The city is once again facing a series of crises. During the winter of 2017, for example, seventy five percent of the citys subway lines were chronically late (Mahler,
20 2018). M uch of the subway systems infrastructure and stations are badly dated and in some cases visibly decaying This has been aggravated by a train switching system that is nearly a hundred years old and, more recently, by storm water inundation that flooded th e citys subway tunnels following Hurricane Sandy (Mahler, 2018). The latter issue has led to the planned closure of the citys L Train route for 15months beginning in April 2019. This closure will reroute 400,000 daily riders in a number of Brooklyn neighborhoods that have fueled some of the citys fastest and most visible growth (Nir, 2017). Nir (2017) observes this closure represents a challenge for the city eclipsed in scope in recent history perhaps only by the transportation challenges following the September 11, 2001 terrorist attack and immediately after [Hurricane] Sandy (n.p.). The Commissioner of the New York City Department of Transportation anticipates that 80 percent of these riders will shift to alternate subway lines, that buses will accom modate an additional 15 percent, and that the number of people walking and biking across a bridge that links the affected area with Manhattan will double (Nir, 2018). Notably, the transit oriented growth trend that prevailed in New York City from 1990 2 014 saw a reversal as New York Citys for hire vehicle (FHV) fleet which includes ride hailing companies like Uber, yellow cabs and similar services began to absorb additional growth (Schaller, 2017a). In 2016, subway ridership witnessed its first decl ine in many decades (Figure 11) The cit ys buses have also seen ridership declines dating back a number of years further (Tsay et al., 2016). These trends have followed nationwide declines in transit use and concurrent growth in per capita vehicle miles travelled (i.e., more driving of personal vehicles) (Schaller, 2017a). This same
21 period also witnessed exponential ridership growth fo r companies like Uber and Lyft, with use of the cit ys FHV fleet increasing by 29 million trips since these companies launched in New York City as well as significant growth in traffic congestion, which is especially pronounced in Manhattans Central Business District (CBD) (Schaller, 2017a). Studies by Schaller (2017a, 2017b) indicate that companies like Uber and Lyft are significantly contributing to this congestion and that they may also be competing with transit, bicycling and walking. Schaller ( 2017b) observes that the adoption of forthcoming autonomous driving technologies by companies like Uber and Lyft may accelerate these impacts, while also offering the opportunity to improve the city s transportation network. Regional Plan Association (RPA) an urban research and advocacy organization whose history in the region dates back to the 1920s and who publish an update to their influential Regional Plan roughly every three decades, released their Fourth Regional Plan in November of 2017. In this document RPA observe that more people are fleeing the New York metro area than any other metropolitan area in the country. Chief among the concerns RPA attribute to this decline, and which they believe represent a serious threat to the regions economic vi tality, are subway and congestion issues the region is choking on its traffic (RPA, 2017a, p. 148) and a chronic shortage of affordable housing. To remedy the former concern, RPA (2017a) prescribe a combination of subway system fixes upgrades and ex tensions that total in excess of $100 billion, along with road pricing initiatives that will discourage excessive driving while generating funds for some of the proposed subway improvements. Among a number of possible solutions to address the regions chronic housing affordable housing shortages, RPA (2017a)
22 propose that automated shuttle services offer an opportunity to add dense, mixeduse infill developments around underdeveloped suburban commuter rail stations in the region by eliminating a significant amount of parking need. Using a wide body of research and policy analysis briefing papers, this literature and the research that follows focus es on: 1. The i mpact of ride hailing companies like Uber and Lyft on the New York metro regions transportation system, as well as variables that influence use of these services. 2. Deployment trajectories, market penetration variables and possible impacts of autonomous ridehailing services and shuttles in the region. 3. Policy responses to the regions contemporary housing and transportation challenges and strategies that could be employed to address anticipated impacts of vehicle automation. Overview of Transportation Network Companies Transportation Network Companies (TNCs) also cal led ridehailing, ridesourcing and ridesharing companies, are ondemand FHV services that match passengers with drivers using a GPSenabled smartphone app. ( The terms TNC and ride hailing will be used interchangeably in this manuscript.) Uber, the larges t of these companies and an early pioneer in the ride hailing industry along with Lyft and the now defunct Sidecar, officially launched in San Francisco in 2011 As of 2018, Uber is operating in 633 cities worldwide with a valuation rivaling many of the worlds biggest auto manufacturers (Uber, n.d.). According to Gurumurthy, Kockelman, and Hahm (2017) 32.5 percent of Americans have personal ridehailing experience; of this group, 27.3 percent have shared their rides with strangers using dynamic ridesharing (DRS) services like UberPool and Lyft Line that reduce trip fares by splitting costs with one or more
23 passengers picked up en route. Globally, Clewlow and Mishra (2017) estimate that within the fiveyear period since these companies began operations, the number of unique users grew to 250 million. By comparison, global membership to carsharing services like Zip car, which was founded in 1999 and represents an earlier model of shared mobility, stands at roughly 5 million (Clewlow & Mishra, 2017, p. 1). New York City Market Penetration In New York City, four TNC companies are currently operating: Uber, Lyft, Via, and Juno. Uber currently has a 72 percent share of the citys ridehailing market (Schaller, 2017). Since Uber opened for business in New York City in May of 2011, the second city the company expanded to after launching in San Francisco a few weeks prior TNCs have grown exponentially In 2017, TNCs represented 68,000 of the 103,000 FHVs regist ered with New York Citys Taxi and Limousine Commission (T LC) the agency responsible for the oversight of these vehicles. 65,000 of the 68,000 TNC vehicles are affiliated with Uber (TNC drivers are often affiliated with multiple companies ). In comparison, the citys iconic yellow cabs are restricted by law to less than 13,600 (Hu, 2017). TNC ridership doubled annually in New York City from 2013 2016, and is currently on track to surpass yellow cab riders hip levels (Schaller, 2017). Composition of Trip Types An intercept study of San Francisco TNC users by Rayle, Dai, Chan, Cervero and Shaheen (2015) found that 67 percent of TNC trips were taken for social or leisure trips, 16 percent of tri ps were work commutes, and 4 percent of trips were to or from the airport. The same study found that 40 percent of trips were homebased, while 47 percent began elsewhere. The authors of this study acknowledge that the intercept locations, which were said to be high traffic areas for ride hailing, as well as the times of
24 day in which the surveys were conducted, may have bi ased their results. The findings of a study conducted by Henao (2017), a University of Denver PhD candidate who worked as a driver for Uber and Lyft and who distributed surveys to his ridehailing passengers seem to support the above findings of Rayle et al. (2015). Babar and Burtch (2017) found that the primary use of TNC trips is for short distance travel. This may be due in part to diminishing cost and travel time advantages for longdistance TNC trips when compared to transit services with dedicated r ights of way (Schaller, 2017a; Tsay, Accurardi, Schaller & Hovenkotter, 2016). Trip Volume by Time of Day and Time of Week In San Francisco, Rayle et al. (2015) found that nearly half of all ridehailing trips occurred on Friday and Saturday with concentr ations in the evening. Schaller (2017a), who analyzed data from millions of New York City Uber trips made public following a Freedom of Information Act request, also found that most growth in TNC mileage occurred in the evenings and weekends. In Manhattan, Schaller (2017a) found that trip growth was concentrated during the morning and evening peak periods and with yellow cab driver changes of shift, which occur at a set time of day governed by the TLC. U ser Demographics The popular conception of TNC users tends to be one of a young and educated urban elite. Much of the literature support this assumption. The intercept surveys of Rayle et al. (2015) found that TNC users were generally younger and better educated than the general population of San Francisco and represented higher than average socioeconomic brackets. Clewlow and Mishra (2017), who conducted a comprehensive two phase travel and residential survey with a representative sample of urban and suburban populations in seven major U.S. cities found that 36 percent of 18 to 29 year
25 olds have used ridehailing services. In comparison, the study authors found that just 4 percent of those aged 65 or older had used these services. Similar disparities hold true for education, income, and neighborhood type. C lewlow and Mishra (2017) found that college educated, affluent Americans use TNCs at double the rate of their less educated, lower income counterparts. Likewise they found that 29 percent of TNC users who live in more urbanized areas of cities have adopted ride hailing and use these services more regularly, whereas just 7 percent of suburban residents of major metropolitan areas use them to travel around their areas and regions. Nearly a quarter of TNC users in major metro areas who have adopted these serv ices use them on a weekly or daily basis (Clewlow & Mishra, 2017). Concurrent p articipati on in other s haring e conomy p latforms. Feigon and Murphy (2016), who conducted a convenience sample of individuals who have used one or more shareduse modesincludi ng transit in seven major metro areas (p. 6), looked at respondents use of multiple shared mobility modes Given shared mobility is a neologism, and typically understood to reference newer mobility options like bike sharing and ridehailing rather than traditional pay per use services like taxis, the inclusion of transit in this grouping lends ambiguity to their findings. As such, the connection between TNC use and the use of other shared mobility modes that have been enabled by recent innov ations in smartphone technology is unclear Further, it is unclear how use of these technology enabled shared mobility services correlates with the use of other sharing economy platforms These other sharing economy platforms include online peer to peer marketplaces like Airbnb and VRBO that match consumers seeking short term rental properties with property owners that rent all or part of their
26 properties on a short term basis as well as services like TaskRabbit and Thumbtack that match consumers with short term labor needs with freelance laborers that provide the requested services. M any of these sharing economy services debuted contemporaneously many rely on similar smartphone technologies, and many share similar business models Given this, it seems fair to assume there is some correlation between the use of ride hailing services, the use of other technology enabled mobility services, and the use of other sharing economy services like Airbnb T he value of exploring potential connections between shared mobility and the wider sharing economy will be revisited later in this manuscript. Impact of Transportation Network Companies The literature on TNCs reports a host of benefits offered by these services both to individual users and to the greater society These include increased mobility, travel time savings, reduced need for parking in congested areas, and possible reductions in drunk driving. TNCs may also improve sustainability outcomes although thi s claim is contested. Conversely, th e literature also reports considerable social detriments stemming from TNCs. These include increases in congestion and overall vehicle miles traveled (VMT); increases in TNC VMT without passengers ( also know n as deadheading) ( Schaller, 2017b, p. 12); and questionable business practices that undercut existing regulatory mechanisms and potentially affect the longterm viability of these companies financial models.
27 Travel Time Savings and Increased Mobility Rayle et al. (2015) argue that ridehailing serves a latent demand for urban travel by offering short wait times and point to point service, and that this is especially true for the young and well educated city dwellers An analysis of travel time savings, which factored for both wait time and inve hicle travel time, found that 66 percent of San Francisco TNC trips would have taken twice as long if they were taken by transit (Rayle et al., 2015). In San Francisco, Rayle et al. (2015) find that TNCs fill gaps in the existing transportation network, ex tending mobility to peripheral areas of th at city presently underserved by taxis and transit. Likewise, Schaller (2017a) found that TNCs helped to expand service to peripheral areas of Manhattan that were historic ally deficient in taxi service and to surrounding suburbs and even rural areas that traditionally lacked FHV services altogether. RPA (2017a) observe that more than a third of New York City residents live outside of subway or commuter rail walksheds For residents of these areas without access to p ersonal vehicles, TNCs offer potential to increase access to economic opportunity and essential goods and services. Further, for captive transit riders, the door to door service offered by TNCs can be particularly valuable for trips that involve transporti ng large and/or heavy items that would be difficult to carry for long distances. Notably, Schaller (2017a) found that while absolute growth of TNC trips was greatest in the Manhattan Central Business District (CBD), relative growth was greatest in peripher al areas of New York City traditionally underserved by FHVs like the Bronx, the southern part of Brookly n, and the eastern part of Queens. Reductions in Drunk Driving and Parking Need In surveys examining motives of TNC users, Clewlow and Mishra (2017) found that avoiding drinking when driving accounted for 33 percent of trips for those who own
28 vehicles but opt to use ridehailing services as opposed to driving themselves. Similarly, Henao (2017) found that avoiding drinking and driving was among the top reasons for ride hailing us e, while Schaller (2017a) observes that avoiding drinking and driving accounts for at least part of the reason these services are used disproportionately on evenings and weekends. Avoiding parking is another primary motive for TNC use. Clewlow and Mishra (2017) found that avoiding parking was the primary reason urban TNC users who own vehicles substitute these services for driving themselves; for this group, avoiding parking was the primary motive for 37 percent of trips. Henao ( 2017) recorded similar findings with greater than a quarter of all users surveyed reporting avoiding parking as a motivating factor in their decision to use ridehailing versus driving themselves. Reductions in drinking and driving and reductions in the need for parking benefit not just the consumers of these services, but the greater society. The former for obvious safety reasons, while the latter offers the opportunity to put urban land to more productive uses than the temporary storage of vehicles. Othe r Purported Social Benefits In addition to the above benefits, which are well supported by the literature on ride hailing, supporters of these services have made a number of other claims. Lyft, for example, claims that they are helping to reduce the carbon footprint left by our countrys dominant mode of transportation driving alone (Schaller, 2017a, p. 3). Likewise, when Uber announced $5 flat fares for its UberPool service in Manhattan, the company said [our] goal is simple: take one million cars off the road in New York City and help eliminate our citys congestion problem for good (Schaller, 2017a, p. 7). However, as the next section explores both these claims are dubious.
29 Increased Congestion and Vehicle Miles Traveled According to AlonsoMora, Samaranayake, Wallar, Frazolli, and Rus (2017), congestion costs American more than $120 billion each year this figure represents approximately one percent of the total gross domestic product and includes more than 5 billion hours of lost time and near ly 3 billion gallons of wasted fuel (AlonsoMora, Samaranayake, Wallar, Frazolli, & Rus, 2017). When factoring for other externalities like emissions, INRIX (2017), a global traffic analytics firms, has estimated the economic costs of congestion to be as m uch as 20 times higher than the figure quoted by AlonsoMora et al. (2017). According to INRIX (2017), New York was the second most congested city in the nation following Los Angeles and had the nations highest concentration of traffic hotspots. INRIX (2017) estimates the direct and indirect price of congestion will cost residents of the New York metro area the equivalent of $64 billion over the next decade. Extrapolating the findings of his Denver area study, Henao (2017) estimates that TNCs add an extra 5.5 billion VMT to the nations roadways. A study of TNC trip data by the San Francisco County Transportation Authority found that much of that citys recent increases in traffic congestion are attributable to the rapid growth of ride hailing services (SFCTA, 2017). Further, the SFCTA (2017) study found that ridehailing trips were concentrated in the densest and most congested parts of the city on busy arterials and traveled extensively through neighborhood streets and along major public transit lines; in aggregate, the agency found that TNC trips in San Francisco account ed for approximately 570,000 VMT on a typical weekday. In New York City, shared use mobilit y is a key element of the current administrations plans to increase environmental sustainability and promote various
30 livability initiatives like increased opportunity for walking and bicycling (Schaller, 2017a, p. 5). However, Schallers longitudinal analysis of TLCs Uber data suggests the opposite outcome (2017a, 2017b). Adjusting for VMT declines in competing FHVs, like taxis and car services, and estimated decreases in personal vehicle use by TNC passengers, Schaller (2017a) estimates a net gain of 600 million VMT from TNCs, even when accounting for DRS options like UberPool that are claimed to reduce mileage. In Manhattan and the inner ring (i.e., Western Queens and Western Brooklyn), TNCs added approximately 7 percent to existing VMT by all vehicles an overall gain of 352 million VMT in these areas from (Schaller, 2017a). According to this study, this figure represents an increase of the same magnitude as [a] 2007 congestion pricing proposal would have decreased in vehicle miles travelled (p. 1). By pushing New York Citys roadway network over capacity, Schaller (2017a) claims this traffic growth in already congested areas of the city has a multiplier effect, creating more congestion than would be expected by the proportion added. In 2017, TNCs and taxis account for more than onehalf of all traffic in the Manhattan CBD and contribute to the slowest traffic speeds on record in this area, averaging less than 7 miles per hour during the day (Schaller, 2017b). Induced t ravel d emand For better or worse, research findings suggest that TNCs are creating new travel demand. Henao (2017) found that 12.2 percent of passengers surveyed would not have traveled were TNCs not avavilable at the time ; Rayle et al. (2015) determined 8 percent of trips would not have been taken in the absence of TNCs. Research by Clewlow and Misra (2017) also supports the conclusion that TNCs are inducing travel demand. While added VMT is an obvious downside of this new travel demand, one could also argue
31 that latent demand served by TNCs stimulate s local consumer economies given a significant proportion of TNC travel happens on evenings and weekends and for social or leisure purposes (Rayle et al., 2015; Schaller, 2017a; Henao, 2017). Low e fficiency and m iles t raveled w ithout p assen gers Given TNC drivers are free to work where and when they please, Schaller (2017b) claims the TNC balancing dynamic at play in cities results in a gravitational pull towards high traffic areas While this creates short wait times for passengers, it also means that drivers in these areas spend a majority of their time driving without passengers while waiting for their next trip request compounding congestion issues as they do so. In the Manhatt an CBD, this means more than a third of FHVs are empty at any given time on weekdays (Hu, 2017) According to Schaller (2017b), the economic underpinnings of this phenomenon suggest it is happening not just in New York, where it has been observed directly but in cities across the US. Using a log of his own trip statistics while working as a driver for Uber and Lyft in the Denver area, Henao (2017) found that only 41.3 percent of the miles he traveled while working as a TNC driver were traveled with a pass enger in the vehicle. This accounted for factors like travel to and from his home to areas of adequate ride hailing demand. Comparing TNCs to other modes, Henaos (2017) analysis of this trip data found that TNCs were less efficient than all but two alternatives taxis and getting a ride (p. 99). Critique of Business Practices Criticisms levied against TNC business practices include circumvention of supply restrictions, lack of data transparency, and questions about the longterm viability of
32 price models underlying the rapid growth in ridehailing. This section explores these issues. Exemption from s upply r estrictions RPA (2017a) argue that TNCs are competitive in part because pricing of these services is largely unregulated and because there are no growth caps on TNCs, allowing them to expand indefinitely to meet demands. In contrast, taxis in New York City are capped at approximately 13,000 (Schaller, 2017a). Traditionally, the city has applied this cap along with high parking fees and various bridge and tunnel tolls to discourage excessive driving in a city where space is at a premium (Schaller, 2017a; Hu, 2017). By circumventing these regulatory mechanisms, critics argue that TNCs have unfairly gamed the system in their favor and do not pay their fair share for the negative externalities they impose like added congestion and pollution (Schaller, 2017a) Data r eciprocity and t ransparency Lack of TNC data reciprocity has been an ongoing issue for municipalities impacted by TNC operations; these same companies benefit greatly from open public data offered by municipalities and government agencies ( Feigon & Murphy, 2016; Tsay et al., 2016) SFCTA (2017), for example, was not able to get data from the California Public Utilities Commission, the state regulatory agency legally mandated to collect this information. Instead, the agency had to work with a university to scrape data from these companies Web application programming interfaces (APIs) in order to estimate TNC travel activity in San Francisco. ( Data scraping in this context is defined as the extraction of large amounts of data from websites using open APIs provided to developers; this is often done without the explicit permission of the data source in
33 question and typically falls into a legal grey area short of hacking.) Data requests for validation of the studys findings were declined by the TNCs in question (SFCTA, 2017) Ostensibly, this is done to project customers privacy (Schaller, 2017a). According to Schaller (2017a), however, large amounts of detailed anonymized data can be made public with no evident harm to user privacy or company interests (p. 24). While some cities like Boston and Portland have established data sharing agreements with TNCs, the walled garden model where TNCs provide no data or data of only superficial value to cities, is much more common (Feigon & Murhphy, 2016, p. 33). According to Clewlow and Mishra (2017) and Feigon and Murphy (2016) cities need to collect this data in order to understand the i mpact of these services on travel choices and infrastructure needs, and to support informed transportation policy and decisionmaking. Given the estimated magnitude of TNC impact, and the substantial benefits accrued by these companies from the use of publ ic goods, this seems like a reasonable exchange. Clewlow and Mishra (2017) observe that similar datasharing mandates exist elsewhere in the transportation sector such as the airline industry. In 2017, the TLC unanimously approved a mandate requiring Uber and Lyft to provide the City of New York with more detailed data than they presently provide; this data will be aggregated at the neighborhood level In addition to providing both pick up and dropoff data the citys TNCs previously did not disclose drop off data the new rule will also track driver hours to measure possible fatigue and whether the trip is taken in a DRS option like UberPool or Lyft Line (Meyer, 2017). It is not clear when this rule will go into effect, or whether it will require further approval, however, the TLCs example can provide an early model for municipalities seeking to enact similar data
34 sharing mandates ( Meyer, 2017). According to the TLC Commissioner, this data will pose no threat to individuals privacy (Meyer, 2017). Longt erm v iability of p rice s tructures TNCs have recently begun experimenting with fares that are cost competitive with transit (RPA, 2017a). According to RPA (2017a), this is presumably done to lure choice transit riders taking short trips. In New York and a handful of other cities, Uber has begun offering $5 flat fares in CBDs during rush hour (Schaller, 2017a). Solomon (2016) contends that Ubers aggressive price cutting is an attempt to kill off rival services like Lyft. In San Francisco, where there i s especially fierce competition among TNCs, Lyft was forced to follow suit with Ubers recent price cuts, slashing per minute and per mile costs by 62 and 29 percent respectively. According to Solomon (2016), [Uber is] using their [$7 billion] capital adv antage to try to win all the marbles (n.p.). In instances where other TNCs have not followed Ubers pricecutting lead, drops in these companies ridership numbers have been significant. Solomon (2016) observes that both Uber and Lyft are losing money l ike its going out of style (n.p.). According to leaked documents, Uber lost nearly $1 billion in the first half of 2015 (Solomon, 2016). Aggressive price cutting strategies and significant research and development (R&D) expenditures into emerging transportation technologies like autonomous driving systems are surely contributing. There have been serious questions about the longterm sustainability of this model (Schaller, 2017a). Large influxes of investment capital and aggressive competition for riders m ay be causing considerable distortions in TNC fares Assuming ride fares eventually stabilize, fares will presumably need to recalibrate to justify these considerable capital expenditures In the meantime, however, TNCs pursuit of choice transit riders may
35 erode the financial solvency of public transit agencies. (This issue is explored further in subsequent sections of the literature review.) Equity concerns There has been considerable documentation of discriminatory practices among taxi services (Ge, Kni ttel, MacKenzie & Zoepf, 2016). Ridehailing companies like Uber and Lyft contend their services extend mobility for minority populations often shunned by taxi drivers. However, Ge et al. (2016) identify a number of ways that TNC drivers could engage in discriminatory behavior including: (i) drivers could choose not to drive in or around certain neighborhoods, (ii) drivers could decline to accept a ridehailing reservation after a passengers identifying details have been revealed, and (iii) drivers could cancel a pickup after a passengers identifying details have been revealed. Using 1500 rides on controlled routes in Seattle and Boston, Ge et al. (2016) devised a study to measure the extent of this phenomenon and whether Uber and Lyfts assertions about mobility benefits for minority populations held true. The researchers found TNC passengers with AfricanAmerican sounding names experienced 35 percent greater wait times in Boston and twice the cancellation rate across trips in all areas. Additionally, fe male TNC passengers were taken for longer and more expensive rides in Boston. Recent news reports have also highlighted the danger of sexual harassment and sexual assault for female TNC passengers. Despite the above, Ge et al. (2016) note that TNCs may nonetheless represent a net mobility improvement for minority populations who often live in areas where transit fails to adequately meet their mobility needs. Of the onethird of New Yorkers who live outside of the walking distance of a subway or train stati on, a disproportionate number live in low income neighborhoods, and a disproportionate number of these low income
36 neighborhoods are majority minority (RPA, 2017a). Looking at Schallers (2017a) data on areas of high trip growth, including traditional minor ity and low income communities like northern Manhattan, this mobility extension seems to hold true. However, in the absence of demographic data on riders in these areas, it is difficult to discern who are the true beneficiaries of TNC services. Some have expressed concerns about whether elderly populations will be excluded from the mobility benefits offered by TNCs owing to lower rates of digital fluency and smartphone access (RPA, 2017b). Likewise, access for disabled passengers who may require a wheelchair lift or other vehicle modifications to accommodate their needs may be excluded from the benefits of TNCs (Rayle et al., 2015) Clewlow and Mishras (2017) findings on age demographic disparities among TNC users support the notion that elderly Americans are largely excluded from the increased mobility benefits offered by TNCs. Impact of Transportation Network Companies on Other Modes Recent research has examined potential impacts of TNCs on driving of personal vehicles and various forms of public transit including buses, subways, and commuter rail. This section documents these findings Driving and Ownership of Personal Vehicles Schallers (2017a) analysis found relatively unchanged levels of private auto use due to TNCs Clewlow and Mishras (2017) nati onwide travel and residential surveys yield a number of interesting findings with respect to TNC impacts on personal vehicle use and ownership: TNC users who use transit own slightly more vehicles than those who only use transit.
37 For those who do not use transit, there are no measurable differences in vehicle ownership rates between TNC users and nonusers. 91 percent of TNC users have not made any changes to their vehicle ownership since beginning to use these services. Those who drive less or who have r educed the number of cars they own have substituted those trips with increased TNC use. Public Transit Service According to Pendall, Blumenberg, and Dawkins (2016) shared mobility works best in the same moderateto highdensity neighborhoods where transit works best (p. 4). Given transit and TNCs often serve the same neighborhoods, researchers have sought to uncover whether the two primarily complement or compete with one another. The literature suggests evidence for both findings, with significant variation between different transit modes and with quality and frequency of transit service as an important moderator. These findings are explored in further detail bel ow. Increased t ransit a ccessibility Feigon and Murphy s (2016) travel time of day analysis found that ridehailing services are most frequently used for late night social and recreational trips when transit frequency is reduced or unavailable and therefore serves to fill gaps in the network Babar and Burtch (2017) who evaluated the impacts of TNCs on transit by looking at the spatiotemporal staggering of Ubers entry in US cities as well as the simultaneous nationwide rollout of Google Maps TNC trip booking feature, find that TNCs are unlikely to compete with longdistance fixed guideway transit systems like subway s and commuter rail and that they may help New Yorkers connect to subway stations at the outer edges of the system by helping t o overcome the first/last mile gap. Rayle et al. (2015) found that 28 percent of trips in San Francisco started or ended within 400
38 meters of rail transit. However, whether these trips primarily represent FLM connections to transit or substitution for tran sit is not clear. Increased h eavy r ail r idership Looking at positive impact s on transit ridership, Babar and Burtch (2017) found that Ubers entry is associated with a 2.59 percent increase in subway use and a 7.24 percent increase in commuter rail use. Respondents to Clewlow and Mishras (2017) surveys reported 3 percent greater use of these heavy rail systems in aggregate. Publicp rivate p artnerships Agencies across the US have implemented or are exploring partnerships with ride hailing companies to complement existing services or substitute for low performing ones (Table 21) (Tsay et al., 2016; Masoud, Nam, Yu & Jayakrishnan, 2017; RPA, 2017b) While many of these programs are still in pilot phases, early results have shown promise to increase transit agency cost savings and to help overcome the FLM gap (Tsay et al., 2016; RPA, 2017b). Decreased b us and l ight r ail r idership Tsay et al. (2016) remark that recent fare reductions that have made TNC fares cost competitive w ith transit are raising red flags for some in the industry (p. 17). The recent period of rapid TNC growth has overlapped with a period of declining trans it ridership ( this is especially true for bus ridership) (Harrison, 2017). However, Tsay et al. (2016) observe that local bus ridership in New York has been in decline for over a decade, predating the arrival of TNCs in the city. Further, Tsay et al. (2016) contend that Cities across the US for example, Seattle, Nashville, and Jacksonville have seen steadily increasing transit ridership in the past six years in spite of the coincident rise of TNCs and other emerging mobility services (p. 17). However, the findings of a
39 number of studies that have looked directly at mode shift causation among current or former transit riders who also use TNCs seem to contradict some of the above assumptions. A list of these follows : Nationwide, Babar and Burtch (2017) found that Ubers entry led to a 1.05 percent decline in bus service utilization and also found that c ities hosting a ridehailing service at the time Google Maps unveiled its TNC trip booking feature had much larger ridership losses relative to those cities who did not. Nationwide, Clewlow and Mishra (2017) found that 49 61 percent of ridehailing trips would not have been taken or would have been taken by transit walking or bicycling; they also found a 6 percent reduction in bus use and 3 percent reduction in light rail use. In San Francisco, Rayle et al. (2015) found 33 percent of surveyed riders would have used bus or rail were TNCs unavailable for the trip in question. In the Denver area, Henao (2017) found that only 5.5 percent of all TNC trips were used to connect to other modes Further, many carpooling trips replaced by ride hailing were FLM connections whereas the ridehailing trip was not. In New York City, Schaller (2017a) alleges that TNC induced traffic congestion is further eroding bus ridership. Anecdotally, Hu s (2017) st ory of a bus rider whose trip from his home in the Bronx to his Manhattan workplace took more than two hours due to traffic congestion support s Schallers claim however, a direct causal relationship has not been established. Transit q uality as a m oderat in g v ariable While the above findings seem to suggest a net negative impact on transit, and a number of these researchers have suggested this is the case, measur ing net impacts precisely is difficult because some portion of TNC trips are likely induced. Vari ations in quality of transit service between and within municipalities, and also by time of day and time of week, further complicate this relationship. Clewlow and Mishras (2017) analysis and Babar and Burtchs (2017) analysis both determined that the substitutive versus
40 complementary nature of this relationships varies widely and is affected significantly by the prevalence and quality of transit in a given city and area. As such, data aggregated at the national or even municipal level may fail to convey c onsiderable local variation. Promise and Shor tcomings of Dynamic Ridesharing DRS options like UberPool and Lyft Line offer considerable sustainability and cost saving promise. According to Quarles and Kockelman (2017), DRS is one of the few ways the worlds transportation future becomes environmentally sustainable (and relatively noncongesting), while still ensuring much personal travel freedom (p. 11). Alonso Mora et al. (2017) suggest a similar set of benef its. Likewise, public relations efforts by Uber and Lyft have claimed these services offer sustainable transportation alternatives and promoted low cost flat fares for their DRS services (Schaller, 2017a). Research by AlonsoMora et al. (2017) and an earl ier study at Columbia University have used simulations to demonstrate the potential for a perfectly efficient DRS model to eliminate over threequarters of New York Citys FHV fleet (RPA, 2017b). Despite the promise and hype surrounding DRS however, excl usive ride trips predominate, and according to Schaller (2017a), most DRS ridership in New York comes from transit, walking and biking, leading to net VMT growth regardless of whether trips are shared or not. Henao (2017) point s to passenger unwillingness, lack of sufficient participation, and extra hassle for drivers as reasons these services in their existing iterations seem to have floundered when compared to exclusiveride options The introduction of autonomous vehicle technologies has the potential to change the DRS equation by eliminating inefficiencies, extending services, and lowering costs. However, there is no guarantee it will do so. This is explored further in subsequent sections of the literature review
41 Overview of Shared Autonomous Vehicles T erminology and Assumptions The Society of Automotive Engineers (SAE) (2016) Taxonomy and Definitions for Terms Related to Driving Automation Systems for OnRoad Motor Vehicles (also known as J3016) which has been adopted by the US Department of Transportation and is considered the definitive guidance on the matter, includes a six degree scale of vehicle automation that ranges from no automation (Level 0) to full automation (Level 5). Highly or fully autonomous vehicles (AVs) also called automated vehicles, driverless vehicles or self driving vehicles refer here to onroad motor vehicles that can conduct all aspects of the dynamic driving task without human intervention. These are considered Level 4 and Level 5 on the SAE scale, respectively (the latter can also operate in all roadway and environmental conditions whereas the former may not be able to) ( SAE, 2016 ) For the purpose of this discussion, references to AVs and shared autonomous vehicles (SAVs) should be assumed to refer to vehicles that are Level 4 or Level 5 on the SAE scale. Connected vehicle technologies complement AV technologies and these two sets of technologies are developing along convergent timelines So, it should also be assumed that AVs referenced in this discussion will have some degree of vehicleto vehicle (V2V) and vehicleto infrastructure (V2I) connectivity, meaning they can communicate info such as vehicle speed and proximity to other vehicles and the infrastructure around them ( and vice versa) SAVs combine elements of ridehailing and conventional carsharing with AV technologies (Krueger, Rashidi & Rose, 2016). Here, t he same technological assu mptions made about AVs also apply to SAVs. A variety of uses and business models have been proposed for SAVs, but this discussion primarily focus es on their
42 anticipated role s in the ridehailing and DRS contexts, and in particular the potential transportat ion system and land use impacts of these applications in the New York metropolitan area. ( The terms SAV and autonomous ridehailing are used interchangeably in this manuscript; SAV DRS and autonomous DRS are also used interchangeably.) Deployment Trajectories Deployment of AV and SAV technologies are progressing rapidly. Waymo, the AV unit of Google, has logged millions of onroad testing miles with employees behind the steering wheel who are able to take control of the vehicles in case of emergenc y. In November of 2017, Waymo started AV testing in Chandler, Arizona, a suburb of Phoenix, without the safety driver behind the wheel the first company to do so at a large scale on public roads in uncontrolled conditions (RPA, 2017b). The company has announced plans to deploy these vehicles throughout the Phoenix area in near future and may also deploy a ridehailing model General Motors Cruise, a self driving electric ride hailing service, expects to deploy in dense urban areas like San Francisco by 2019 (RPA, 2017b). Additionally, Uber recently signed an agreement to purchase 24,000 SAVs from Volvo, who provided the vehicles and technology for the companys recent autonomous ridehailing pilot in Pittsburgh (Isaac, 2017). The conversion of the vehicle fleet from gasoline to electric and various road pricing schemes and gas tax alternatives such as mileagebased user fees which are further enabled by V2I technologies, are advancing along similar trajectories and may eventually converge (RPA, 2017b). Gurumurthy and Kockelman (2017) anticipate that personal AV ownership will not be commonplace for many years. Considering the above developments, some of
43 the first deployments of these technologies may arrive soon as SAV fleets operated by ride hailing com panies in urban areas Gurumurthy and Kockelman (2017), Gurumurthy, Kocelman and Hahm (2017), Schaller (2017a), and RPA ( 2017b), support this presumption, noting that economies of scale, labor issues, and other cost concerns will drive TNCs increasingly towards SAVs and especially DRSenabled SAVs, as these companies compete fiercely for market share in areas like San Francisco and New York. RPA ( 2017b) state s that the future of TNCs is surely autonomous (p. 24) Schaller (2017a) believes SAVs are likely to arrive soon, however, he speculates that TNCs will likely be composed of mixed fleets of SAVs and traditional ridehailing services for many years to come, as these companies are loath to depend entirely on a new technology until it is completely proven and shown to overcome operating limitations (p. 18). Potential Impacts of Shared Autonomous Vehicles According to Schaller (2017a), one can glimpse the SAV future by observing the impact of TNCs today: The modeling shows that changes in travel and vehicle mileage are generated primarily from the combination of demandresponsive service and shared use of the vehicles, with automated operations being of secondary importance. In other words, the SAV future can arrive with continued growth of TNCs driv en by actual people. That future may thus be evident today (Schaller, 2017a, p. 5). This finding has important implications for those seeking to understand both the market penetration patterns and possible impacts of SAVs While TNCs may portend the impac ts of automation, there is also reason to believe they could magnify or fundamentally alter urban mobility ecosystems and land use patterns for better or worse (Alonso Mora et al., 2017; Gurumurty & Kockelman, 2017; RPA, 2017b; Schaller,
44 2017a) Possible change in the New York metro area SAV and SAV DRS contexts explored here include congestion impact s; FLM improvements; urban infill and changes to the street hierarchy; and transit competition and ridership attrition. Con gestion and Vehicle Miles Traveled The research of AlonsoMora et al. (2017) and similar studies on highcapacity SAV DRS have determined the potential for these services to dramatically reduce the size of the FHV fleet in New York City. Using New York City taxi data to validate a mathematical model of SAV DRS in the Manhattan CBD Alonso Mora et al. (2017) found that 2,000 10person capacity autonomous DRS vehicles (15 percent of the citys current taxi fleet) could serve 98 percent of FHV demand in this area within a mean waiting time of 2.8 minutes and a mean trip delay of 3.5 min (p. 1). The researchers found that increasing the maximum allowable delay to 5 minutes or more significantly increased the service rate, decreased wait time, and reduced travel distance of each SAV DRS vehicle. Notably, Alonso Mora et al. (2017) claim the results are robust with respect to density of requests and could therefore be applied to other cities (p. 1). The research of AlonsoMora et al. (2017) applied an additional layer of methodological rigor to an earlier study done at Columbia University that determined approximately 9,000 autonomous DRS vehicles could replace 50,000 FHVs (RPA, 2017b). A simulation of DRSenabled SAVs in the Orlando area by Gurumurthy and Kocklem an (2017), which used cellphone location data for experimental validation, found: Nearly 70 percent of the singleperson trips could be shared with other persons traveling solo and with less than 5 minutes added travel time (to arrive at their destinations ), and this value climbs to 90 percent for 15 to 30 minutes of added wait or travel time (Gurumurthy & Kockelman, 2017, p. 1).
45 An earlier agent based simulation in Austin, TX found similarly dramatic congestion and VMT reduction results with DRS enabled S AVs (Gurumurthy & Kockelman, 2017). While simulation models have demonstrated the efficacy of highcapacity autonomous DRS in a variety of urban contexts, the congestion reduction potentials, high ridership volume, and grid network of the Manhattan CBD mak es an especially attractive location for this model. While these studies use robust mathematical models, and the implications for congestion and VMT reduction in urban environments are considerable, they do not account for the various human factors that may impact actual use of these services (this is explored further in subsequent sections of the literature review ) On the flip side, if riders largely shun the DRS enabled SAV options and exclusiveride SAV trips prevail the opposite effect could occur. Fi rst/Last Mile Transit Connection Pakusch and Bossauer (2017) believe that SAVs could complement public transit networks by providing an FLM connection to trunk line routes and substituting for less frequently used routes. As explored in the section on transit/TNC P3s, this is already happening to some extent and the pace at which these types of complementary partnerships are forming appears to be accelerating (Tsay et al., 2016). In the New York metro area context, RPA (2017a) observe that SAVs and DRS enabled SAVs could increase accessibility to trunk line transit in lower density communities. Additionally, they can increase transit service efficiency by providing overnight services on low volume routes and could potentially provide paratransit service with the assistance of aids (RPA, 2017b). Assuming these services took the form of a P3, the potential efficiency gains and cost savings offered by SAVs could help offset some of the subsidies provided. Again, existing transit/TNC P3s such as the City of Centennials
46 FLM light rail connection and Pinellas Countys low performing bus route replacement offer useful models (Tsay et al., 2016). Infill D evelopment According to a widely cited SAV DRS simulation model of Lisbon, Portugal, SAVs could eliminate as much as 95 percent of public parking need in urban areas (Schaller, 2017a). SAV DRS simulations in Austin and elsewhere have made similar findings (Gurumurth & Kockelman, 2017). RPA (2017c) estimate that as much as three quarters of trips to and from the regions commuter rail stations could be in SAVs, which would open up opportunities to develop a significant portion of the almost 1 million square feet of land used solely for surface parking withi n a mile radius of each of [the regions] commuter rail stations (p. 8). This would offer an ample opportuni ty for the region to build midto highdensity mixeduse transit oriented development that would address a significant portion of the regions current and f orecast housing need (RPA, 2017c ). According to Lu, Du, Dunham Jones, Park and Crittenden (2017) and RPA (2017c), subsidies for such SAV services could be offset by: (i) increased commuter rail ridership revenues and (ii) real estate value capture. In a New York Times interview, Mahler (2018) offers evidence to support the assumption that major New York area real estate developers would be supportive of these arrangements. However, most of these changes would require zoning changes (RPA, 2017c), and the zoning changes would be subject to the whims of local politics. Historically, proposals to bring midto highdensity infill to suburban areas of New York, especially where there is a low income housing component have met considerable local resistance. The battle to des egregate public housing in the city of Yonkers is a famous example of this.
47 Changes to the Street Hierarchy In addition to urban infill opportunities opened up by SAVs, RPA (2017a) observe that 80 percent of New York Citys street space could be allocated to walking, bicycling, and public transit, up from approximately 25 percent today (p. 136). Through geofencing and dynamic routing based on the time of day and week, this could take the form of: Sidewalk widening Installation of cycle tracks and dedicat ed onstreet rights of way for transit Dynamic pedestrian plazas and car free streets Curb free streets that will allow users to comingle in a low speed environment Transit Competition and Ridership Attrition According to Pakusch and Bossauer (2017), RPA (2017b) and Tsay et al. (2016), traditional transit systems will continue to play a vital role in dense urban areas like New York after the arrival of SAV s. This is especially true of transit systems with dedicated rights of way and high carrying capaciti es like heavy rail. However, Krueger et al. (2016) observe the following: [SAV systems] pose a threat to [public transit] systems because SAVs could provide a more convenient user experience at a competitive rate. The overall ride experience of SAVs without DRS would be much smoother, as no transfers would be necessary and the vehicle would not have to stop to let passengers board or egress the vehicle. SAVs would offer more privacy and intimacy, seating availability would be guaranteed and walking times would be significantly reduced. As a consequence, travelers could make more efficient use of their travel time than on [public transit] (Krueger et al., 2016, p. 345). Pakusch and Bossauer (2017) observe that the potential for SAVs to amplify the disruptive effects of TNCs are putting growing pressures on transit agencies. As a result, existing public transport business models are increasingly being reevaluated. This is particularly true for suburban services and inefficient bus routes (Pakusch &
48 Bossauer, 2017). Given the above, RPA (2017b) caution that governments should take steps to guard against SAVs skimming off wealthier choice transit riders and leaving captive transit riders with underfunded and diminished options. Shared Autonomous Vehicle Sentiment As explored above, various mathematical models and simulations make a compelling case for DRS enabled SAV. However, the impacts and longterm viability of these services will be affecte d significantly by both the traditional economic factors that influence mode choice and by a host of cognitive factors that transcend the utilitarian assumptions of these mathematical models (Krueger et al., 2016). Both of these considerations are critical to understanding which cohorts are most likely to adopt SAVs and SAVs with DRS and where this is most likely to occur first. Economic Factors and Willingness to Pay According to Krueger et al. (2016), the traditional transportation service attributes like travel cost, travel time, and wait time remain important determinants to SAV and the acceptance of DRS. In a survey, Krueger et al. (2016) found that respondents perceived SAVs with DRS and SAVs without DRS as two distinct options; these suggest configurations and fares will be crucial determinants to the acceptance of DRS, especially where SAV without DRS is offered as an alternative. From the literature on willingness to pay (WTP) for SAVs and DRS enabled SAVs a number of findings stand out: Gurumurthy, Kockelman, and Hahm (2017) found just 62.5 percent of Americans were willing to share rides with strangers assuming no delay is accrued; the pool of respondents who were willing to use the SAV DRS option reported an average WTP of $0.87 per mile. At $0.60 per mile, Quarles and Kockelman (2017) found respondents were willing to share less than 19 percent of their SAV miles with strangers versus $1 mile for
49 the nonDRS SAV option; respondents reported they would use the DRS option regularly at $0.44 per mile. Gurumurthy and Kockelman (2017) found that only 4.4 percent of Americans are willing to use the DRS option at night with a rider unknown to them. Quarles and Kockelman (2017) found SAV use appears unpopular at $4 per ride to access rail stations. Gurumurthy, Kockelman, and Hahm (2017) also observe that actual system experience will significantly impact WTP and give the example that walking to ones own personal vehicle takes a predictable amount of time, whereas traffic patterns ; access and egress of other passengers; and roadway or environmental conditions are among the variables that could affect how quickly an SAV arrives. Influence of Cognitive Factors and Modality Styles Gurumurthy, Kockelman, and Hahm (2017) note that att itudes to things like sharing rides with strangers at night could evolve as users become accustomed to this. Shifting social norms may also impact attitudes and adoption rates For example, the idea of renting a room in ones home to a rotating cast of str angers probably appealed to few before Airbnb s arrival. Likewise, the idea of driving at 70 miles per hour against oncoming traffic or flying at 40,000 feet may have frightened many who reached maturity before car and airplane technologies did. Similarly, the idea of giving up control of a car to a computer may be frightening to many contemporary Americans; these same individuals may have no qualms about traveling on a commercial airliner that operates primarily in autopilot. If one were to administer a survey to the average American about their intent to purchase an automobile at the turn of the 20th century when paved roads were sparse and cars were pr ohibitively expensive and offered questionable utility, this survey would
50 probably not predict the rat e of adoption seen in the years that followed, when the Ford Model T debuted and a host of other developments made car ownership much more attractive. In a similar vein, Pakusch and Bossauer (2017) who administered surveys to users of autonomous subways a nd driverless shuttles in Europe, found that most who were skeptical before taking a trip on one of these autonomous modes generally have very positive impressions afterward; this finding held true regardless of respondents age. These findings are supported by other research that has shown prior experience with technology increases the acceptance of that technology (Pakusch & Bossaeur, 2017, p. 57). In a survey exploring receptivity to using SAVs as a FLM shuttle to connect to train stations, Yap, Correa, and Van Arem (2016) found a high level of disutility associated with AV travel time. This runs contrary to the assumption that AVs offer greater travel time utility by offering the opportunity to work or relax while traveling rather than having t o drive. The study authors suggest that perceived discomfort with these technologies may account for this counterintuitive finding (Yap et al., 2016). Notably, Pakusch and Bossauer (2017) report high levels of satisfaction with autonomous trains such as Denmarks subway system. They observe that the most common reason riders on these systems report feeling secure is because these systems are closed and controlled and there is no oncoming traffic. Regarding the findings of their study, Yap et al. (2016) conc lude: [The] usually referred advantages of automobile automation may not yet be perceived as such by todays travelers. The importance of attitudinal factors in the mode choice leads to uncertainty on how people will react when AVs are introduced in pract ice. It shows that psychological factors can pl ay an important role in the choice of travelers to use automated vehiclesSince automated driving is a quite new and innovative way of
51 mobility, the classic instrumental attributes like travel time and cost do not tell the whole story (Yap et al., 2016, p. 15). Pakusch and Bossauer (2016) contend existing transportation mode choice is likely to affect predisposition towards use of SAVs and use of SAVs with DRS. The authors suggest that public transit users are accustomed to being passengers and typically use these services for utilitarian motives alone. On the other hand, drivers are accustomed to having locus of control and may consider sensation seeking aspects part of the appeal of driving (Pakush and Bo ssauer, 2016, p. 57). The findings of Krueger et al. (2016) suggest a strong link between an individuals modality style and their likelihood to switch to SAVs. Krueger et al. (2016) propose that multimodal travelers frequently reevaluate their decisions t o optimize their utility and are therefore relatively more open to explore novel mobility options (p. 353). Conversely, individuals whose modality is mostly and almost exclusively centered around the use of the private car may be reluctant to use SAVs (Krueger et al., 2016, p. 353). Quarles and Kockelman s (2017) survey findings seem to support these ideas The study authors found that the largest group of respondents in all possible scenarios intend to rely on personal vehicles after SAVs are available to them and do not intend to use these services. On the other hand, they observe notable shifts for those currently using nonautomobile modes, suggesting mode shift towards SAVs may come largely fromthose currently using public transit, walking and bi cycling (Quarles & Kockelman, 2016, p. 11). Krueger et al. also suggest a potential link between higher order orientations and the propensity to use SAVs with DRS (p. 353). Regarding this higher order orientation, the authors offer the following: Empir ical evidence suggests that use of the private car is not only influenced by utilitarian considerations, but also by symbolic affective
52 motives such as the use of the car as symbol of social status and self expression as well as feelings of autonomy, freedom and flexibilityThe impersonal or collective nature of SAVs suggest that for some individuals, SAV services may not satisfy the symbolic a ffective ends to the same degree as the private car can. Therefore, it can be assumed that individuals who put a hi ghvalue on the nonutilitarian motives of mobility might not choose SAVs even though configurations of an SAV service may be objectively superior to the mobility offered by a private car (Krueger et al., 2016, p. 345). Demographic Profile of Likely Users Quarles and Kocklelman (2016) find that younger and better educated respondents are more likely to adopt SAV and DRS enabled SAV. Four additional studies cited by Quarles and Kockelamn similarly conclude that younger people are more likely to use AVs. In addition to young people, Krueger et al. (2016) find that multimodals are more likely to adopt SAVs. Pakusch and Bossauer (2017) find men are significantly more willing to use autonomous transport than women, regardless of the type of transport. In sum, ac cording to the available literature, the population group most likely to be early adopters of SAV and DRS enabled SAV are young, well educated multimodal males. Policy Responses Congestion Pricing and RideHailing Surcharge Clewlow and Mishra (2017) observe that in the absence of significant policy action, congestion and emissions are likely to grow assuming current TNC growth trajectories continue. Given this, they suggest congestion pricing and enforcedpriority lanes as effective ways to prioritize increasingly scarce road space. Historically, policy makers in New York have used a number of actions to discourage driving and excessive FHV use including: (i) caps on the number of taxis allowed to operate in the city (ii) pricing of taxi fares that was on average 4.5 times higher than subways and (iii) parking supply limitations (Schaller, 2017a). New York State Governor Andrew Cuomo
53 recently launched a state task force, called Fix NYC, to examine a number of range of possible solut ions that would address two of the citys biggest issues: growing congestion and a deteriorating subway system (Mahler, 2018). The Fix NYC task force is studying possible implementation of a congestion pricing plan that would charge all vehicles entering t he CBD. The revenue from CBD congestion pricing would be used to fund improvements to the citys mass transit system (Mahler, 2018). Past CBD congestion pricing proposals including the recent MoveNY effort and former Mayor Michael Bloombergs failed 2008 i nitiative have met opposition in Albany and among some outer borough New Yorkers (Schaller, 2017b ). Significantly, Bloombergs 2008 congestion pricing proposal, proposed three years before the arrival of Uber in New York City, was slated to reduce traffic congestion in the CBD by seven percent the same amount by which TNCs are estimated to have increased traffic in the CBD since their arrival. The magnitude of the citys current congestion and mass transit problems may be changing outlooks however. Mayor Cuomo has indicated bold action is in order to solve these two issues that are increasingly seen as threats to the citys economic competitiveness (Hu, 2017). Schaller (2017b) believes the public has little desire to curtail TNC operations given the mobi lity benefits they bring, however, one approach under consideration would involve charging New York City TNC users a $2 $5 fee per ride. According to Schaller (2017b), a $3 surcharge on every TNC and taxi trip that starts in the Manhattan core would generate $475 million in annual revenues. Schaller (2017b) believes this would do little to curtail FHV trips. To support this, he cites a study that found a 10 percent increase in taxi fares is only likely to produce a 2 2.5 percent reduction in
54 ridership; therefore, applying this $3 fee to TNC fares would only be expected to produce a 3 4 percent reduction. Importantly, Schaller (2017b) assumes equal price elasticity for both taxi and TNC trips, whereas taxis have traditionally been considered a premium t ravel option in New York and much of the recent growth of TNCs have been driven by aggressive cost cutting and fares that are increasingly transit competitive. Therefore, the relative inelasticity of taxi fares may not apply for TNCs. Schaller (2017b) notes there have been no studies done to date on price sensitivity of TNC fares. A third a pproach would be to focus on reducing ridehailing VMT between passenger trips by applying a ridematching system similar to what is used in the citys airports (Schaller, 2017b). Interestingly, of all the above, Uber has advocated for the adoption of congestion pricing schemes that would apply to all vehicles traveling in congested areas (Hu, 2017). The company argues levying a surcharge on all TNC trips would unfairly penalize outer borough and ridehailing passengers who may not be contributing to the citys traffic hotspots. Transit Improvements Accordin g to RPA (2017a) at least $111 billion is needed to upgrade the New York City s ubway to a modern and functional system (Mahler, 2018) This list includes : modernizing a train signal ing system so dated that repairs require custom fabrications ($27 billion ); renovation of stations most desperately in need ($14 billion); eliminating sharp turns to increase system throughput ($5 billion); and extending the system track miles to serve densely populated low income areas of the city who are heavily transit dependent yet lack easy access to the subway (RPA, 2017a; Mahler, 2018). This does not include expenses such as subway platform doors that would improve rider safety and decrease the track debris that presently plagues the system, nor does it include the
55 cost o f switching to a driverless metro that would further increase system efficiency and throughput (cities throughout Europe have begun to successfully implement these driverless train systems) ( RPA, 2017a; Mahler, 2018). While the subways problems and the c ost of fixing them seem monumental, it is important to consider the issue in both its historic context as an engine of economic generation and regeneration, and alongside the example of peer cities London, for example, whose subway system is older than al l others, faced similar ly intractable issues. Today, the city has a well functioning system that is moving towards driverless trains (Mahler, 2018). Los Angeles, long the quintessential car city, has committed $120 billion to its transit sys tem over the next four decades. Meanwhile, China has been constructing new metro systems across the country in a bid to ensure development in the heavily populated country remains transit oriented (Mahler, 2018). According to Mahler, the New York metro area is responsibl e for $1.7 trillion in gross metropolitan product, the equivalent of 9 percent of the gross domestic product. Columbia University geophysicist Klaus Jacob, who authored a study on the economic impact of storm surge on the city s subway system, estimates th at losing the subway for a month would cost the city about $60 billion in lost economic output (Mahler, 2018, n.p.). Tsay et al. (2016) argue the sheer peoplemoving capacity of mass transit remains unparalleledand no smartphone app can change that simple fact (p. 17). The citys recent TNC driven car oriented growth trend is fundamentally unsustainable (Schaller, 2017a) the regions present population densities were enabled by a network of extensive high capacity rail systems, and most critically, by the subway itself (Mahler, 2018). Given New York City is largely an archipelago, it would be impossible to
56 retrofit the city to accommodate personal vehicles or lowcapacity ridehailing as primary modes without razing a substantial part of the city in the process. The city is simply too built up and there is too little space. This was attempted once before by Robert Moses, and much of the city s recent planning efforts have focused on reversing his legacy of urban renewal and car oriented megaprojects. Regardless of whether the City is able to solve its growing congestion problems, the good order of the subway remains critical to the economic health of the region and to the physical, economic, and social mobility of its cit izens (Mahler, 2018). Mahler suggests the city should tax a portion of the monumental sums of wealth currently being generated by the citys financial industry and other boom ing sectors to ensure this is the case (n.p.). Failure to do so, according to Mahler (2018), would be a tragic misstep. While the outcomes would not be as immediately catastrophic as Jacobs worst scenario, Mahler (2018) suggests a slower, more insidious r esult: What is the alternative? Heres one possible scenario: New York wont die, but it will become a different place. It will happen slowly, almost imperceptibly, for years, obscured by the prosperity of the segment of the population that can consistent ly avoid mass transit. But gradually, an unpleasant and unreliable subway will have a cascading effect on New Yorkers relationship with their city. Increasingly, we will retreat; the infinite possibilities of New York will shrink as the distances between neighborhoods seem to grow. The gap between rich and poor will widen. As the citys density dissipates, so too will its economic energy. Innovation will happen elsewhere. New York City will be just some city (Mahler, 2018, n.p). Structural I ssues Schaller (2017a) suggests a prosaic solution to the MTAs dilemma: focus on improving existing services rather than vanity projects like the Oculus the $4 billion transit hub built adjacent to the former World Trade Center site and other prohibitively expensiv e ribbon cuttings like the longdelayed Second Avenue subway line, whose
57 completion hastened recent subway declines. However, elected officials are often reluctant to spend limited political capital on low visibility projects like transit operations and maintenance. Tsay et al. (2016) observe that structural issues like these, which include politics and agency governance, have long been and will continue to be the biggest challenges for transit providers regardless of the incursion of competing options. Mahler (2018) observes the following about the relationship between the state and city, and the competing interests within its borders: The subway subsists on an ad hoc patchwork of taxes implemented and overseen by a governor who represents millions of voters well beyond the greater metropolitan areaNew York City is responsible for [an estimated] 55 percent of the states revenues, but that doesnt change a fundamental political reality for its governors: People who dont ride the subway dont want to pay for the subway. A more parochial version of this divide exists in the city itself, where representatives of car centric outer borough neighborhoods continue to fight congestionpricing plans that could deliver hundreds of millions of dollars every year to the M.T.A. (Mahler, 2018, n.p.). While agencies like the MTA face considerable structural barriers not all of the citys peer transit providers are subject to the delays, cost overruns, and maintenance declines endemic to the New York City subway. Paris, for example, has expanded its metro system at a fraction of the cost of additional track mileage in New York and has implemented automated trains (this despite the fact that both systems share a similar provenance) (Mahler, 2018) Taken as a whole, the above suggests a need not just for upgrades to the subway system, but a need for structural change. Evolution of Public Sector Transportation The literature shows that TNCs simultaneously complement and compete with transit with variations in transit mode and quality of service moderating the magnitude of impact (Clewlow & Mishra, 2017; Babar & Burtch, 2017). TNCs fill a mobility vacuum
58 and complement transit by providing: (i) trips to and from low density areas, (ii) latenight trips where service is infrequent or nonexistent or where riders may have safety concerns, and (iii) an alternate mobility option where there is overcrowding during peak travel hours (Rayle et al., 2016). As explored earlier in this manuscript, P3s between transit agency and TN Cs can eliminate low performing routes and offer more cost effective provision of paratransit service (Feignon & Murphy, 2016) This may take the form of reinforcing trunk line transit routes in highvolume corridors through TNC or SAV FLM connections while elim inating low performing alternatives (RPA, 2017b) In New York City and elsewhere, this may call for tough decisions like ceasing 24/7 subway operations and instead subsidizing individual trips during early morning weekday hours where ridership volumes are low (RPA, 2017a) This would also offer the MTA an opportunity to perform regular maintenance without having to reroute passengers or close lines. While RPA (2 017a) call for the addition of 62 subway track miles and a number of new stations to serve areas of the city that are currently underserved, but that have high concentrations of low income transit dependent riders. While this is a laudable goal, recent history illustrates that MTAs major capital projects come at the expense of necessary maintenance (Mahler, 2018). If the core functionality of the subway remains comprised, extending the reach of the subway to peripheral areas will give limited mobility returns. Value capture taxes for real estate ventures around new transit lines may serve to off set some of the capital costs of new subway stations and track miles ( Lu et al. 2017; RPA, 2017a; Mahler, 2018). However there is potential for lower income residents of these areas to be displaced in the process.
59 TNCs and SAVs may offer a more pragmatic alternative that would not require huge capital expenditures but could nonetheless meet these populations mobility needs. Where mode shift from transit to ride hailing is occurring, policy makers should consider not just punitive measures improvement s to the transit system and transit service cuts, but how to adapt the nature of their transportation services to meet the demands of a rapidly evolving mobility ecosystem and fundamental changes to customer s mobility expectations. According to Schaller (2017a), TNCs brightly illuminate what the public wants from transportation services, and provide a seemingly irrefutable argument behind the need to meet the publics demand for fast, reliable, comfortable and affordable transportation services (p. 24). More specifically they highlight the publics desire for real time arrival info, short wait times, and ondemand point to point service (Rayle et al., 2015). Fixedroute transit systems, especially those without dedicated rights of way like traditional b uses come up short in these regards. In light of this fact and evidence of rider attrition the arrival of cost competitive TNCs bankrolled by big influxes of investment capital (Schaller, 2017a) may signal the need for a more fundamental shift. In order to continue serving the public interest and satisfy core transportation system goals, Feignon and Murphy (2016), Tsay et al. (2016), and RPA (2017c) suggest that agencies may need to retool to allocate subsidies for mobility outcomes rather than provision of specific modes. Feignon and Murphy (2016) observe that San Francisco Municipal Transportation Agency and a number of agencies in Seattle have begun to coordinate existing public transit operations with things like bikesharing, carsharing, ridesourcing, shuttles, parking, and curb access (p. 33). The end goal for
60 some is mobility as a service [Maas] which converges several modes of transport services into a single, accessible, ondemand mobility service (RPA, 2017c). According to RPA (2017c), Finland is a leader in MaaS For example, Helsinkis Ministry of Transport and Communications recently partnered with a mobility startup to provide unified payment systems and intermo dal trip planning services that seek to seamlessly connect citywide public transit to regional transit and private shared mobility options (RPA, 2017c). Intermodal fare payment and trip planning systems such as these could streamline connections between modes and even include mileagebased user fees for users who opt to travel by SAV for part of their journey (RPA, 2017c). If successful, these systems could hasten a shift away from private mobility and towards a more sustainable shared mobility future Feignon and Murphy (2016) argue changes like these, which include integr ated fare payment and real time service info, are essential to transportation agency evolution. H owever, they caution th at transit provider s will need to ensure that low income and minority riders with limited access to credit cards, the internet or smartphones are not excluded by these changes. Significance of Shared Autonomous Vehicle Deployment in New York City By many indications, AV technologies are forthcoming and may arrive first for many consumers with fleets of SAVs operated by ride hailing companies who can offset high overhead costs through increased passenger volumes and labor cost reductions (Schaller, 2017b). On the other hand, there are a host of direct and indirect factors that could significantly influence the pace of AV/ SAV deployment and their market penetration patterns Lack of enabling legislation at the federal level as well as thorny legal, ethical, privacy, cybersecurity and liability issues could directly impact the deployment of AVs (RPA, 2017c) Further the literature on A V sentiment suggests
61 consumer demand for these technologies is limited, and that purported benefits like increased utility of travel time are significantly moderated by unease with these technologies (Yap et al., 2016). However, the literature also suggest s that exposure to and familiarity with new technologies can quickly change attitudes (Pakusch & Bossaeur, 2017). Shifting social norms also have the potential to hasten acceptance of new technologies. Additionally, there are a host of exogenous factors th at could significantly impact deployment These include issues such as possible job loss from the approaching wave of automation (especially in the transportation sector) and significant changes to the demographic composition of the U S to name just a few. Given the above, much uncertainty surrounds AVs and their potential to impact transportation systems, land use patterns, the environment, and other issues of concern. H owever, recent developments like the Waymo deployment in the Phoenix area, which is r eported to be followed shortly by the deployment of an autonomous ridehailing service, suggest this may happen sooner than later. Because of its large and concentrated population, many of whom are highincome and well educated, RPA (2017b) contend that New York City and the surrounding inner core will likely come first in SAV adoption, despite the challenging and complex operating environment posed by its streets (p. 25). McKinsey also forecasts that AVs are likely to penetrate markets of dense high income cities such as New York City first (RPA, 2017b, p. 21). The following findings from the literature support this notion: New York City has been a target for and early adopter of recent shared mobility innovations (e.g., it was the second city Uber launched in and among the first North American cities to adopt a large scale bikeshare system) (RPA, 2017c).
62 Competition among TNCs is especially intense in cities like New York and San Francisco (Solomon, 2016) and this is likely to spur further innovation (Schaller, 2017b) High capacity dynamic ridesharing simulations done by Alonso Mora et al. (2017) and a precursor study from Columbia University researchers have demonstrated that DRSenabled SAVs have the potential to deliver great efficiency gains in the Manhattan CBD. The research of Krueger et al. (2016) and others suggest that young, educated multimodals are among the most receptive to SAVs New York City has the highest concentration of these groups of any city in the US by a significant marg in (e.g., New York City has roughly 10 times the population of San Francisco). The shutdown of the L Train for 15 months beginning in April 2019 will create a significant mobility vacuum. This will occur in neighborhoods that have driven a significant amount of the citys recent growth (Nir, 2017) and where ridehailing patronage is high (Schaller, 2017a). Compared to the city as a whole, residents of these areas are disproportionately young and well educated. Further, the timeline of the L Train shutdown coincides with the anticipated deployment of autonomous ridehailing services in urban areas. AV deployment in New York City is likely to take the form of SAVs operated by TNCs; these may or may not be DRSenabled. According to Schaller (2017b) results of mode choice modeling suggest that many of the most desirable characteristics of SAV and SAV DRS already exist in todays ridehailing services. Determining geographic hotspots of shared mobility use and sharing economy participation may be val uable to understanding market penetration patterns of SAV s within New York City. Preparing for the Impact of Shared Autonomous Vehicles Planners and decision makers should engage with emerging transportation technologies to avoid repeating the mistakes of the automobile era (RPA, 2017c, p. 3) Assuming automation increases the efficiency of TNCs ride hailing market share i s likely to grow in the absence of regulations (Clewlow & Mishra, 2017) In this case, the
63 existing impacts of ridehailing may be magnified commensurately. Issues and strategies to consider in advance of the arrival of these technologies include: Scaling VMT fees to the number of passengers per vehicle to discourage excessive single occupancy SAV trips (RPA, 2017c). Providing sufficient curb space for pick ups, dropoffs and deliveries for SAVs by reclaiming onstreet parking spaces (RPA, 2017c). Ensuring SAVs are fairly distributed throughout cities and regions to prevent disparities in access to these technologies (RPA, 2017c ). Subsidizing SAV trips as a supplement or substitute for existing services (RPA, 2017c). Explor ing opportunities for real estate value capture in parking spaces adjacent to commuter rail stations and other transit oriented development opportunities that can make more efficient use of space and reduce auto dependency (Lu et al., 2017; Mahler, 2018). Increase technical capacity of transit agency staff and explore adoption of a MaaS model (Clewlow & Mishra, 2017). Schaller (2017a) contends that the central task of New York City transportation policy is to shift trip growth back to nondriving modes while maintaining the mobility benefits of ridehailing This challenge is likely to grow with the arrival of AVs and SAVs. A ccording to Schaller (2017b), the ar rival of a utonomous driving technologies in cities brings both promise and peril, and should be met with preemptive policy actions: In the long run, SAVs can bring myriad benefits to cities. These range from reduced traffic injuries and fatalities t o reducing the use of singleoccupant vehicles, freeing parking spaces for new housing and commercial buildings, and increased use of electric vehicles. While recognizing those benefits [there are] r isks in the long transition period that precedes a fully autonomous future. These findings thus underscore the important role for public policy in managing traffic impacts as the day of shared autonomous fleets in its major ur ban centers approaches ( Schaller, 2017b, p. 18) .
64 Table 2 1. Transit/Ride Hailing Company Pilots and Public Private Partnerships. Area Agency P3 Type/Value Program Goal Status Pinellas County, FL Pinellas Suncoast Transit Authority TNC subsidy up to $3 Replace low performing bus routes Established Altamonte Springs, FL City of Altamonte Springs TNC subsidy up to 25% FLM commuter rail connection Established Centennial, CO City of Centennial 100% TNC subsidy FLM light rail connection Established Boston, MA Massachusetts Bay Transportation Authority Taxi subsidy FLM transit connection for disabled riders Established Seattle, WA King County Metro Determined by municipalities Determined by municipalities Planned Dallas, TX Dallas Area Rapid Transit N/A FLM transit connection Planned Atlanta, GA Metropolitan Atlanta Rapid Transit Authority N/A FLM transit connection Planned Memphis, TN Memphis Area Transit Authority Integrated transit/TNC fare system FLM transit connection Planned Raleigh, NC GoRaleigh Transit Integrated transit/TNC fare system FLM transit connection Planned
65 CHAPTER 3 METHODOLOGY The literature uncovers a wide range of emerging transportation issues that are currently impacting the New York metropolitan region or are expected to impact the region in the near future. It also highlights a number of research gaps. The connection between TNCs and transit in the New York metro raise questions on the extent that TN Cs are substituted for transit and the motivating factors underlying such decisions ; r esidential subway accessi bility and its impact on TNC use or lack thereof; the magnitude of TNC surcharge that would cause the average user of these services to consider switching to an alternate mode; and the willingness of New Yorkers and CBD commuters to support congestion char ging and/or a flat ridehailing surcharge. Additionally, the literature raises questions on the willingness of current TNC users to embrace emerging transportation technologies and shared mobility models including willingness to use an autonomous ridehai ling service; willingness to use dynamic ridesharing both in the existing TNC context and with expected automation of these services in the near future; and the extent TNC users patronize other shared mobility and sharing economy platforms as well as the extent these activities may coincide spatially. Finally, the literature review highlights a number of gaps in regional policy issues that may be impacted by these emerging transportation technologies and shared mobility models including willingness of the regions residents to use an FLM shuttle to connect to subway and commuter rail as well as willingness of the regions suburban residents to support infill development in parking lots adjacent to the regions commuter rails (assuming these FLM shuttles mak e the need for park andrides obsolete).
66 Summary of Methodology Some of the above questions have a high level of geographic specificity while others apply broadly. In order to address both specific and broad questions across a range of geographic scales a novel survey based methodology was devised. This methodology is composed of a number of connected parts. To survey residents of the New York metro area, the MTurk platform was used. MTurk is an online labor marketplace that offers cost and efficiency advantages over many alternate survey methods. However, the MTurk platform is proprietary and knowledge about the characteristics of this population are limited. To select the appropriate study subareas for geographically targeted questions, a service area analysis of commuter rail drive sheds and subway walk sheds was performed using GIS. To explore how residential subway proximity relates to TNC use, the Crowdsource Reporter application was used. The Crowdsource Reporter application is embedded in the survey instrument. Using this application, MTurk survey respondents are directed to tag the intersection nearest to their home using a randomly generated number that anonymizes their responses. This geotag serves as a proxy for respondents residential location and can be joined back to survey responses using the corresponding random number saved in the survey responses. Three open data sets representing Airbnb locations in New York City, Uber pickups in New York City over a six month period, and 2006 2010 American Community Survey origin and destination data for New York State were used to conduct a proximity analysis of respondent locations collect ed through Crowdsource Reporter This proximity analysis was solely qualitative and intended to lend context to other results and suggest areas for future research.
67 The final step in this process involved the design and distribution of the survey instrument The survey instrument was created using Qualtrics an online survey design and data analysis platform. The survey employed a novel branching logic that prompted respondents with a specific set of questions based on their residential and/or workplace ZIP Codes, as determined by the GIS service area analysis. A dditionally, a d isplay logic was created for questions that require appropriate answers to preceding questions (in some instances the branching and display logic interact to pose highly targeted questions to specific sets of respondents). Finally, the surv ey instrument was distributed using TurkPrime a thirdparty platform used to extend the functionality of MTurk The constituent parts of this methodology are ex plored in further detail below Amazon Mechanical Turk Casler, Bickel and Hackett (2013) descr ibe the Amazon Mechanical Turk (MTurk) platform as follows: MTurk is an Internet marketplace where employers post Human Intelligence Tasks (HITs) for paid workers to complete, with typical HITs consisting of small tasks such as comparing product images, transcribing podcasts, copying business card text into a database, or responding to online questions. The marketplace principally was designed for employers (requesters) to hire workers to do simple jobs that are better suited to human than computer based labor. Of late, scholars also have found it to be a fruitful forum for recruiting participants to complete computer based tasks (Casler et al., 2013, p. 2157). As scholarly use of MTurk has grown in popularity, researchers have raised questions about the external validity of data obtained through this method. Examining the demographic composition of respondents recruited on MTurk, Casler et al. (2013) found MTurk workers to be more diverse on average than participants recrui ted through undergraduate samples and also more diverse than other paid internet panels Examining the validity of MTurk recruitment for political science research, Huff and
68 Tingley (2015) compared MTurk respondent demographics to those of the Cooperative Congressional Election Survey a nationally stratified sample survey administered by the f ederal government each fall. While MTurk respondents were observed to be younger as a whole and more liberal across age ranges in comparison to the control survey the researchers concluded that the two samples were comparable in many respects. While MTurk samples show some demographic stratification, panels recruited in this fashion nonetheless comprise a broad and fairly r epresentative range of the US population (Huff & Tingley, 2015). A number of studies have also compared attentiveness and choice patterns of respondents recruited through MTurk to undergraduate panels (both faceto face and online), paid survey panels recruited via professional marketing resear ch companies, and other subject pool recruitment sources. Kees, Berry, Burton, and Sheehan (2017), Hauser and Schwartz (2016), and Casler et al. (2013) found MTurk offered comparable or superior results in nearly every respect. Using instructional manipulation checks to measure respondent attentiveness Hauser and Schwartz (2016) found that MTurk respondents were consistently more attentive than college subject pool samples ( interestingly, this study included a University of Florida subject pool sample). Us ing a prisoners dilemma task, a priming task, and a framing effects task Casler et al. (2013, p. 2157) found that choice patterns were identical for MTurk and inperson test takers. In comparisons across five samples, Kees et al. (2017) found MTurk to be superior across nearly every variable in question (Table 31). This was especially true with respect to cost per respondent.
69 Regarding MTurk respondents high performance on the instructional manipulation checks, even those that involved novel or minute manipulations, Hauser and Schwarz (2016) hypothesize that by offering MTurk requesters the option to reject worker reimbursement for surveys that fail to meet certain criteria, the MTurk platform incentivizes close attention to detail among workers Addit ionally, requesters can select minimum performance thresholds for MTurk recruits. According to Hauser and Schwartz (2016), the contemporary MTurk study paradigm suggests restricting MTurk samples to highreputation workerswith at least 95% approval rating and 100 or more approved HITs (p. 401). Following this paradigm and taking sensible safeguards, Casler et al. (2013) conclude that integrity of data collected via MTurk represents no greater a concern than data collected through other methods Further, the MTurk method is considerably more affordable than the Webbased survey alternatives referenced by Kees et al. (2017). Pilot Study While a significant number of studies have explored the representativeness and integrity of MTurk panels for use in social science research, only a handful have used MTurk recruitment for transportation research. These studies include Guo, Zhao, Wong, Mishra, and Wymans (2017) study on subway map design conducted on a national sample of MTurk workers; Xus (2017) similar study on subway map design conducted on a sample of Washington, DCbased MTurk workers (2017) ; Sarriera, Alvarez, Blynn, Alesbury, Scully and Zhaos (2017) study of the social aspects of DRS, conducted on a national sample of MTurk workers that use TNCs; K rupa, Rizzo, Epstein, Brad Lanute, Gaalema, Lakkarraju, and Warrenders (2014) study on consumer attitudes towards plugin hybrid electric vehicles, conducted on a national sample of MTurk workers; and
70 Zhao and Lees (2013) attitudinal study of commuting experiences, conducted on a North American population. Of these five peer reviewed transportation studies, only Xus (2017) study targeted a specific geography (Washington, DC metro area ). This study had 371 MTurk participants. No other transportati on, planning or geographic studies were found in the literature scan that used MTurk. Hence, little is known about the geographical distribution of MTurk workers. Further little is known about the total population size of MTurk workers in the U S In December 2016, an MTurk pilot study was conducted on AV sentiment for a course term paper. With respect to the present study, the goals of the pilot study were as follows: (i) gain familiarity with the MTurk platform and learn MTurk best practices; (ii) evaluate resp ondent demographics and integrity of data collected; and (iii) examine the potential to conduct a geographically targeted study at a scale that would be reasonably representative of the target population(s). On December 13, 2016, a survey entitled Recepti vity to Self Driving Vehicles and Emerging Transportation Options was distributed on MTurk to 100 US based workers that had completed at least 1000 approved HITs and had an approval rating greater than 97 percent. Workers were paid $1.25 for each completed survey. All 100 responses were collected in roughly 45 minutes. Pilot study respondents were disproportionately white, well educated, young and male. However, the title of the survey may have resulted in a self selection bias. Given this, demographic fi ndings of the pilot wer e not deemed reliable. N evertheless, the data collected suggest ed respondents engaged thoughtfully with the questionnaire, while the speed with which the survey collected the 100 responses suggest ed a large population
71 of active MTurk workers. Using a geographic information system (GIS) application, distribution of pilot study respondent locations was examined through geolocation of X/Y coordinates associated with individuals internet protocol address es This method does not offer sufficient precision for use at the neighborhood level, however it does offer a rough approximation of regional distribution. These X/Y coordinates were then mapped relative to US population density using data from Esris (n.d.) Living Atlas of the World. Figure 31 offers a qualitative comparison of this distribution where darker orange areas represent population concentrations and blue dots represent MTurk respondent locations. As expected, the distribution of the MTurk sample r oughly follow ed the general population distribution of the US, with clustering in metropolises like New York and Los Angeles and greater concentrations of respondents east of the Mississippi River. Screening Criteria Considerations Following Hauser and Sc hwartzs (2016) recommendations, study qualification was restricted to MTurk workers who had completed at least 100 HITs and who had a 95 percent or greater approval rating. Worker payment per survey was an additional consideration. The literature suggests requesters have an ethical obligation to pay workers at least the equi valent of federal minimum wage ( $7.25/hour in 2018) For example, if a requester estimates that a HIT will take 10 minutes to complete, they should pay at least 1/6th of the federal minimum wage, which would amount to $1.21 per HIT in this case. Additionally, University of Florida Institutional Research Board protocol on MTurk requires that requesters pay respondents the full value of the HIT regardless of the perceived quality of the survey. The planned budget cap for this study was $1000. Factoring for Amazons 40 percent surcharge for HITs with more than 10 respondents
72 this budget would yield 590 HITs assuming the survey took 10minutes and the payment adhered to the minimum wage equi valency Ultimately, the survey was estimated to take approximately 10 15 minutes to complete and respondents were offered $1.35 for each completed survey. Duplicate internet protocol addresses were excluded from the study to prevent ballot box stuffing as were internet protocol addresses that originated outside the target area (more on this below). Geographic screening criteria was an important consideration of this study As mentioned above, no literature on the geography of MTurk workers was found. While MTurk lets requesters restrict respondent location by state, it does not offer the ability to restrict respondent location by city or region. Therefore, in order to study local or regional issues, it is necessary to apply additional geographic screen ing criteria. Initially, Washington, D.C. was considered as an alternate study area given it is an entirely urban area, has high transit accessibility and shares a similar set of population characteristics and transportation issues with New York This wou ld have eliminated much of the need for additional geographic screening. However, because the population of D C is less than a 1/10th that of New York City, the pool of MTurk respondents in the area may not have been sufficiently large. Further, restricti ng the study to Washington, D C would have eliminated the ability to study transportation issues at a state or regional scale. Because this study explores a wide range of local and regional issues that concern, it would have been ideal to sample from par ts of the entire tri state New York metropolitan area, which includes portions of New York, New Jersey and Connecticut However, due to time and budget constraints this was not practical. Instead, the survey
73 was restricted to MTurk workers who lived in Ne w York State Downstate New York is an unofficial designation for the area that includes New York City, Long Island and the Lower Hudson Valley, and which comprises over half of the New York metropolitan areas population and a majority of the population o f New York State. Respectively, Long Island and the Lower Hudson Valley are well served by the Long Island Rail Road (LIRR) and Metro North Railroad commuter rail lines Considering the demographics, urban form and transportation systems of Long Island and the Lower Hudson Valley are similar to those of suburban New Jersey and Connecticut, these areas of the Downstate region were seen as reasonably equivalent to and representative of the suburban areas of the New York metro area as a whole. However, because some of the issues the study explores apply not just to the state or region, but at the corridor and neighborhood levels those who live within the walkshed of the L Train subway line, for example further subdivision of the Downstate region was necessary. ( Henceforth, t his study refers to the Downstate region as the primary study area and the whole of New York State as the secondary study area.) To determine the appropriate target populations for each of these subareas GIS analysis wa s conducted. Combining the target population for each subarea yields the total primary study area population and inform s the selection of an appropriate sample size. This is explored further below. Geographic Information System Analysis Delineation of Study Subareas For geographically targeted questions, five subareas were chosen based on a review of RPAs (2017a) Fourth Regional Plan and other literature sources like Schallers (2017a) UNSUSTAINABLE? The Growth of App Based Ride Services and Traffi c, Travel and the Future of New York City Since the geographically targeted
74 questions focus largely on questions related to transit, transit accessibility thresholds were selected as the method for delineating the subareas (the Manhattan CBD was selected irrespective of transit access, however its spatial extent overlaps with one of the transit threshold areas as will be explained below ). ZIP Codes were chosen as the geographic unit of analysis because these info is easy to collect from respondents The fi ve ZIP Code groups and subgroups are as follows: New York City residents who live in an officially designated Transit Zone Manhattan CBD residents and commuters ( a subgroup of the Transit Zone) Residents of Brooklyn who live within a 30minute walk of the L Train subway line ( a subgroup of the Transit zone) Lower Hudson Val ley residents who live within a 15minute drive of Metro North Long Island residents who live within a 15minute drive of the LIRR Service a rea a nalysis. To measure transit accessibility for the latter three ZIP Code subgroups the appropriate transit and roadway network data was collected. This included collecting ZIP Code data for New York State from Esri (n.d.), New York metro region roadway network data f rom OpenStreetMaps (n.d.) Additionally, station locations for Metro North, LIRR and the New York City subway were collected from CUNY (n.d.) Following the collection of this data, a service area analysis was conducted using the ArcGIS Network Analyst ext ension. The first step in this process was to set the walking speed for the L Train subgroup and driving speed for the Metro North and LIRR subgroups For expediencys sake, network impedance factors like traffic or stop lights were not included in this es timate. The average human walking speed of 3.1 miles per hour was selected for the L Train subgroup. Factoring for both time in motion and time
75 at rest 30 miles per hour was selected as an average drive time for the Metro North and LIRR subgroups Using the walking and driving speed assumptions above, average driving miles per minute and walking miles per minute were calculated. Next, subway and commuter rail station location data and roadway network data w ere loaded into the Network Analyst extension. Us ing this data, the tool is able to create concentric polygons along the network that enclose the respective walking and driving distance to the nearest subway or commuter rail station. The service area analysis process was repeated for each of the three ZI P Code groups and subgroups in question. The wal k time threshold for the L Train ZIP Code subgroup service area analysis was set to 30 minutes, and the drive time thresholds for the Metro North and LIRR ZIP Code groups w ere set to 15 minutes. Next, the Select by Location tool was used to find ZIP Codes that intersected the service area analysis polygons. These ZIP Code groups were then exported to create standalone shapefiles for the LIRR and Metro North groups and the L Train subgroup The Transit Zone shapefile was obtained from the City of New York (n.d.) and encloses areas of the city that are densely populated, transit accessible, and do not have minimum parking requirements. The L Train subgroup is located within the Transit Zone. The Select by Location tool was also used to select the ZIP Codes for Lower Manhattan and Midtown Manhattan, which comprise the Manhattan CBD The Manhattan CBD subgroup is also enclosed within the Transit Zone. The results of this analysis are shown in Figure 32. For the Transit Zone, the service area analysis was used in conjunction with respondents residential location data in order to measure residential proximity to
76 subway stations ( the process of collecting respondents residential l ocation data is discussed in the following section) This involved overlaying point shapefiles representing respondents residential locations with the concentric service area analysis distance bands In this case, the concentric distance bands represent f ive minute walking increments radiating outward from each station in the New York City subway system. Together, these six concentric distance bands enclose an area representing a half hour walk from each subway station within the Transit Zone. Using the Se lect by Location tool, the respondent location point shapefiles enclosed within each five minute walking distance threshold were exported as a group. For example, every respondent in New York City residing within a 5 10minute walk of the nearest subway station were grouped. This was repeated for each 5 minute distance band from 0 5 minutes, 5 10 minutes, 10 15 minutes and so forth, with 30 minutes as the maximum distance threshold. Next, a field called Walk Distance was added to the attribute table of each walk distance polygon to indicate which distance threshold it represents. The residential location point shapefiles were then merged into a single shapefile that contained all New York City respondent locations and the associated subway walk distance for each. This shape file was later joined to the survey data using a unique identifying number common to each. The details of this process are discussed in the following sections. Crowdsource Reporter Application To explore the relationship between TNC use and subway accessibility respondent s residential locations had to be located with relative precision. Esris Crowdsource Reporter application was used to accomplish this Crowdsource Reporter is an online mapping application that is typical ly used by local governments to crowdsource the location of potholes and other public works issues and is marketed by Esri for this use
77 From a user standpoint, this application requires little more technical savvy than that required to use Google Maps. C onsequently, most Internet users are perfectly capable of using the Crowdsource Reporter application. The preparation of the Crowdsource Reporter application required performing the steps below sequentially : 1. An empty feature service layer that contained a numerical field labeled UserID was created using ArcGIS online. 2. An ArcGIS online map was created and the empty UserID layer was loaded into this map. 3. The spatial extent of the map was set to load to the New York metro area by default and an appropriate basemap was selected for maximum legibility of street intersections 4. An ArcGIS online group was created and the UserID layer and online map were loaded into this group. 5. The Crowdsource Reporter Application was uploaded to the ArcGIS online group. B y default the Crowdsource Reporter application loads the maps and layers contained in the online group created in the previous step. 6. The user interface of the application was configured to minimize barriers for a nontechnical audience. 7. A mobile responsive if rame was created using the website iframely.com The resulting code allowed the application to be embedded in the online survey application. Additionally, the iframe was configured to automatically resize its display to optimize for device screen size A number of additional steps were taken to maintain respondents privacy and anonymitiy First, rather than ask respondents to tag their home location in Crowdsource Reporter, they were asked to tag the location of the intersection nearest to their home. This is a reasonable proxy for residential location, especially in urban areas that follow a grid pattern, as New York City does. Second, a string of code was embedded in the Qualtrics survey that generated and recorded a unique random number for each respondent. Respondents were asked to enter this random number in
78 the UserID field within the Crowdsource Reporter application. Users were assured in the IRB and survey instrument that only the primary investigator would be able to link respondents survey responses back to their intersection locations. To record this information in the Crowdsource Reporter app, respondents were asked to perform the following six steps in sequential order (Figure 33 shows the user interface at step 5 in the process) : 1. Click the blue Proceed as Guest icon in the application window below. 2. Click the green Submit a Report button at the bottom of the application window. 3. Enter this number in the UserI D field : [Random Number] 4. Zoom to the intersection closest to your home using the map controls. 5. Tag the intersection closest to your home using the cursor. 6. Click the Report It button once selected. Crowdsource Reporter Pilot Study Because the Crowdsource Reporter process was fairly complex and untested for this use, a pilot study was conducted to determine the feasibility of this method. In the Crowdsource Reporter pilot study, r espondents were asked to follow the six step process outlined above. But rather than ask respondents to tag the intersection closest to their home, they were asked to tag the grocery store nearest to their home. Following the completion of this task, respondents were asked two questions: 1) How difficult did they find the use of the Crowdsource Reporter application; and 2) Would they feel com fortable reporting the location of the intersection nearest to their home if a study asked them to do so. The pilot Crowdsource Reporter survey was distributed to 30 New York State respondents, all of whom completed the mapping task. On cursory examination, many of the crowdsourced grocery store locations appeared to coincide with actual grocery stores. Of the 30 respondents, 23 reported that they found the process very easy or
79 somewhat easy, 3 said they found it neither easy or difficult, while the remaining four reported that they found the process somewhat difficult. Regarding the intersection location question, 19 respondents reported they would feel comfortable, 10 reported that they might feel comfortable, and just 1 reported they would not feel comfortable. The findings of the pilot study validated the use of this method in the s urvey instrument. Qualitative Analysis of Open Data Sets To better understand the spatial distribution of MTurk respondents relative to travel patterns and sharing economy activity a number of datasets were downloaded and mapped using GIS. A rigorous and structured analysis of possible relationships between respondent locations and these indicators are beyond the scope of this research. However, qualitative proximit y analysis of this data may open areas of future areas of inquiry (e. g., areas likely to be early adopters of SAVs or hotspots of sharing economy participation ). For example, Figure 34 shows a high concentration of Airbnb rental units and Uber pickups near the L Train line. Future research could investigate whether there is a correlation between these indicators and the likelihood to adopt future shared mobility and sharing economy innovations like SAV and SAV DRS. Understanding spatial proximity of these phenomena is a first step. Distribution of t rip o rigins and d estinations in New York State To understand how distribution of MTurk workers spatially coincides with travel patterns at state and regional scales, Nelson and Raes (2017) trip origin and desti nation dataset was mapped relative to Esris (n.d.) U.S. Freeways shapefile. The Nelson and Rae data employs American Community Survey travel survey data from 2006 2010. These were both clipped to the boundaries of New York State. The Nelson and Rae (2017) trip origin and destination data are not assigned to a highway
80 network Instead, they indicate the gravitational relationship between various areas across the state (Figure 35 ). Distribution of Uber p ickups and Airbnb h otspots in New York City Little is known about the distribution of shared mobility use or use of the wider sharing economy within cities. Potential hotspots of shared mobility and sharing economy participation may predict areas that will be among the first to adopt future innovati ons like SAV and SAV DRS. To visualize how home locations of New York City respondents who use TNCs compares to distribution of TNC trips and other sharing economy activity two datasets were incorporated. Slee (2017) created a method to scrape Airbnb s API to determine the approximate location of every rental unit hosted on the site for a given time period, as well as characteristics about the rental units themselves (e.g., number of rooms available) A dataset that represent s the totality of Airbnb units on the market in New York City in June 2017 (39,168 rental units) was downloaded as a standalone table. Using the Add X/Y function in ArcGIS, this data was located in space on a map of New York City. Next, an optimized hot spot analysis was performed to det ermine statistically significant concentrations (95 percent) of Airbnb units. Using map symbology, a continuous density gradient was created from the Airbnb optimized hot spot data. These hotspots were mapped relative to the New York City subway system usin g data from Esri (n.d.) The result s of this analysis are shown in Figure 34 below where the magenta surface represents Airbnb hotspots and the colored polylines represent a map of the subway system Notably, these hotspots represent areas with the highes t concentrations of Airbnb units However, a lesser density of Airbnb units is spread across the entirety of New York City.
81 Schaller (2017a) identified the data source of the Uber pickups used in his analysis as FiveThirtyEight (2016). FiveThirtyEight obtained this data by filing a Freedom of Information Act request with the TLC, who required Uber to report trip pickup locations aggregated to one of 257 designated taxi zones These taxi zones are equivalent to New York City neighborhoods. This data was made available by FiveThirtyEight (2016) on GitHub, a data sharing repository, where it was downloaded as a standalone table. This dataset represents all reported Uber pickups from January June of 2015 by taxi zone. To visualize this data, taxi zone shapefiles were downloaded from CUNY (n.d.) and these were joined to the Uber raw data downloaded from FiveThirtyEight (n.d.). To map the distribution of Uber pickups by taxi zone relative to each other and other data sets a 3D layer was created in ArcGIS Pro This was done by using the Feature to Point function that aggregated taxi zone attributes to a single point within the taxi zone polygons centroid. The number of pickups per zone were then linked to this points Z value. This Z value was then extruded using 3D map symbology Figure 34 shows this data relative to the Airbnb hotspots and subway system map, where the height of the red lines represent the volume of pickups per zone. Two things should be noted here. First, the distribution of trip data is absolute and not normalized by the area of the taxi zone. Second, pickup data only gives half the picture. Schallers (2017a) research shows that dropoffs tend to distribute more in residential areas, whereas pickups concentrate in the CBD and around the citys two airports: JFK and LaGuardia. However, due to privacy concerns dropoff data has not been made publicly available by the TLC
82 Design of the Survey Instrument A number of components went into the design of the survey instrument Qualtrics, an online survey application, was selected for the survey instrument. To direct respondents down the appropriate path according to their ZIP Code group classification, a survey branching logic had to be created in t he backend of Qualtrics Second, the questions themselves had to be selected. Some of these were adapt ed directly from the literature while others built on the literature findings to address research gaps. Finally, a survey display logic had to be created to ensure that respondents met the appropriate screening criteria for questions that required this. These will be explained in further detail below Survey Branching Method Qualtrics does not have a built in function to branch surveys according to ZIP Codes or other numerical data sets. A novel survey branching method was devised to display the appropriate set of geographically targeted questions t o the respective ZIP Code groups and subgroups The core elements of this process are displayed in Figure 3 6 This process followed the following three sequential steps: 1. In the survey flow section of Qualtrics, five embedded data fields were created and the set of ZIP Codes associated with each of the ZIP Code groups and subgroups were added to these fields. 2. A branch field was added within the survey flow section below the survey question block where the ZIP Code question is displayed. This field contains a string of code that establishes a branching logic. 3. A second embedded data field was added under the branch field that associates the embedded data in step 1 with the branching logic in step 2. Survey Design and Display Logic The questionnaire was designed as a stated preference survey and includes 47 questions designed to address some of the issues and research gaps identif ied in the
83 literature review. These questions are included in the appendix and numbered in order of their appearance. In the survey, demographic questions and questions about location are nominal; attitudinal questions use an ordinal Likert scale; and questions regarding respondents willingness to pay (WTP) use a ratio scale. The Crowdsource Reporter application was embedded as t he fifth question in the survey. This was done in part to weed out low performing respondents at the beginning of the survey and in lieu of an instructional manipulation check. Kees et al. (2017), Casler et al. (2013) and others recommend MTurk survey admi nistrators use instructional manipulation checks, attention checks and/or trap questions to weed out respondents who do not follow instructions or who use low effort satisficing strategies (Vannette, n.d.). However, Vannette (2017), the Principal Research Scientist in the Qualtrics Methodology Lab, performed a largescale global experiment to test the efficacy of these methods and found that use of these in the Qualtrics platform can actually degrade data quality and introduce a demographic bias. Consequent ly, Qualtrics has stopped recommending that their customers use these methods (Vannette, n.d.). Heeding this advice, the survey instrument did not employ instructional manipulation checks, attention checks, or trap questions. Instead, respondents ability to successfully complete the complex Crowdsource Reporter task suggests sufficient engagement and attention to detail. Table 33 gives a breakdown of questions by subject area and links these to specific sources in the literature. This table also relates s urvey questions to ZIP Code groups and subgroups where applicable. Finally, this table includes the display logic for each applicable question. For example, questions 10 and 11, which examine
84 substitution of TNCs for other travel modes, inherits the logi c from question 6, which asks respondents about the frequency that they use TNCs. In this instance, only those who report having used a TNC within the last 3 months are asked questions about the mode their last TNC substituted for and how TNC use has affec ted their overall trip making behavior. When combined with the survey branching logic, this question display logic helps to ensure the right questions are being posed to the right groups. This enables specific sets of questions to be asked to specific grou ps in order to address some of the issues and gaps identified in the literature. Survey Distribution Target Sample Size The population of the primary study area is roughly 12 million (Figure 32). In order to sample at the 95% confidence level with a confidence interval of 5, a sample size of 384 was required. Initially, the plan was to conduct a screening survey of New Yor k State MTurk workers to find those that lived in the New York metro area. However, due to budget constraints, there was concern that the low price offered by this screening survey would inhibit full participation. Due to this concern, the survey was distr ibuted to all New York State MTurk workers. Based on a qualitative assessment of MTurk worker locations in the pilot study (Figure 31), the geographic distribution of MTurk workers was assumed to follow the general U S population distribution. Based on t his assumption, the desired sample size for the New York metro area was scaled up proportionally to account for the New York State population (roughly 20 million). Using these assumptions the sample would need to collect responses from 610 New York State MTurk workers in order to reach 384 respondents in the New York metro area. Hence, 610 was chosen as the target sample size. Surveying the entirety of New York
85 State also offers the advantage of collecting additional data for questions that are not specifi c to a given geography This also enables future study of how responses may differ based on variations in population density, urban form and other geographic and socioeconomic factors. In this regard, New York State offers a rich variety of contexts ranging from densely urban New York City, to exurban parts of Downstate New York, to Rust Belt cities like Buffalo, and even incl udes parts of rural Appalachia. Distribution of Survey Instrument Using TurkPrime Using the MTurk worker social science research prot ocol suggested by Hauser and Schwartz (2016), the survey instrument was distributed on January 14, 2018 to MTurk workers who had completed at least 100 HITs and who had at least a 95 percent approval rating for completed HITs. Rather than using the Amazon Mechanical Web Turk website to distribute the survey, the survey was distributed via TurkPrime. TurkPrime adds a secondary set of features that are configured especially for academic studies Further, TurkPrime offers a much friendlier user interface for t his purpose than Amazons MTurk requester interface (TurkPrime claims that over 5000 labs and researchers use their services) The use of TurkPrime requires that requesters link their MTurk requester account to their TurkPrime account using the Amazon Web Services API. While TurkPrime charges requesters an extra fee, it also offers a hyperbatch feature that allows requesters to save half off Amazons 40 percent commission for HITs with 10 or more respondents. Ultimately, this saves requesters money while also simplifying the distribution of the HIT and adding an additional set of features tailored to academic researc h. The HIT description presented to MTurk workers through the TurkPrime interface is as follows: A 10 15minute academic study that involves the use of a mapping application (you will be asked to tag the street intersection nearest to
86 your home). A screen capture of the TurkPrime HIT description is provided in Figure 37. The Q ualtrics survey was closed on February 6, 2018 after running for 23 days. The survey was started 613 times and completed 502 times. 500 MTurk workers submitted the survey completion code on TurkPrime. The number of completed surveys fell short of the sampl e target size by roughly 18 percent. Given more time, it may have been possible to collect all of the 610 desired responses. The total cost for the 502 responses factoring for both Amazons fee and TurkPrimes fee is $943.40. The pilot study cost $21. Ther efore, the study came in under the desired $1000 budget. Median study completion time was 9.3 minutes, which was shorter than the expected 10 15minute estimated completion time.
87 Table 31. Summary of Comparisons Across Sample Sources. Adapted from Kees et al., 2017, p. 151. MTurk Sample Student Sample ( Online ) Qualtrics Sample Lightspeed Sample Cost per respondent $0.75 $3.75 $5.88 Measure reliability Best Better Worst Worst Manipulation checks Best Best Good Good Instructional manipulation check Best Worst Worst Worst Attention checks Best Best Better Worst Ease of data collection Easiest Variable Easier Most Difficult Total time to collect all data Fastest Variable Fast Slowest Number of studies participated in (30 days) Highest Lowest Moderate Moderate Hypothesis testing Consistent Consistent Consistent Consistent
88 a Residential groups and subgroups. b Includes Manhattan CBD residents and all who work in the Manhattan CBD regardless of residential location. Table 3 2. Primary study area groups/subgroups by transit access thresholds and size. Geography ZIP Code Groups Transit System Transit Access mode Access Threshold # of ZIPs Residential Population Manhattan, Brooklyn, Queens, Bronx a Transit Zone Subway Walking 30 min. 122 6,959,109 Long Island a LIRR Long Island Rail Road Driving 15 min. 184 3,593,864 Lower Hudson Valley a Metro North Metro North Driving 15 min. 110 1,438,062 Subgroups North/East Brooklyna L Train L Train (subway line) Walking 30 min. 13 910,949 Lower and Midtown Manhattan b CBD 23 626,670 Total 11,991,035 NY State 19,849,399
89 Table 3 3. Survey questions by subject, inherited logic, ZIP Code group, and source. Question(s) Subject(s) Inherited Logic ZIP Code Group Literature Source 1 5 Respondent Locationa 6 TNC use Concurrent sharing economy participation a Feignon & Murphy, 2017 Henao, 2017 7 Primary travel mode by trip type Quarles et al., 2017 8 9 TNC awareness Reason for not using or infrequent use of TNCs 6 10 11 TNC mode shift and induced demand 6 Feignon & Murphy, 2017 Henao, 2017 12 Influence of transit quality on TNC substitution 11 Clewlow & Mishra, 2017 Babar & Burtch, 2017 13 TNC use by time of day and week 6 Rayle et al., 2015 Feignon & Murphy, 2016 14 TNC surcharge required to induce mode shift a 6 Schaller, 2017b 15 17 WTP for DRS with and without delay 6, 9 Quarles et al., 2017 Alonso Mora et al., 2017 18 TNC users willingness to use automated TNCsa 6, 9 19 20 TNC users WTP for SAV and SAV DRSa 18 Gurumururthy et al, 2017 Quarles et al., 2017 21 24 Interest in using FLM shuttle to subway for those with limited access a Internal Transit Zone RPA, 2017a Lu et al., 2017 26 27 L Train mode shift a Internal L Train Nir, 2017 28 35 WTP for commuter rail FLM shuttlea Internal LIRR Metro North RPA, 2017c Lu et al., 2017 36 38 Likeliness to support infill development at commuter rail stations a Internal LIRR Metro North RPA, 2017c Lu et al., 2017 39 40 Likeliness to support TNC surcharge and/or CBD congestion charginga CBD Schaller, 2017b Mahler, 2018 Hu, 2018 41 47 Demographics a Addresses a gap in the current literature.
90 Figure 31. Location of MTurk pilot study respondents relative to US population density. Figure 32. Map of ZIP Code groups and subgroups in the primary study area.
91 Figure 33. Crowdsource Reporter application adapted to collect intersection locations nearest to respondents homes.
92 Figure 34. Distribution of Uber pickups and Airbnb Hotspots in New York City.
93 Figure 35 Distribution of trip origins and destinations in New York State.
94 Figure 36. Elements of the survey branching process. Figure 37. TurkPrime survey description presented to MTurk respondents.
95 CHAPTER 4 RESULTS Respondent Demographics Table 41 compares demographics of the NYC transit zone respondents with demographics for the entire statewide sample. As one would expect, the NYC Transit Zone group shows considerably more racial and ethnic diversity than the sample as a whole. For example, the proportion of the entire sample that identifies as non Hispanic white is 70.5 percent, whereas the proportion of NYC Transit Zone respondents who identify as nonHispanic white is just 43.8 percent. As a whole, educational attainment for NYC Transit Zone respondents is higher than the entire sample, with 15.7 percent more NYC Transit Zone respondents having obtained a 4year degree. However, h ousehold income does not differ significantly between the two groups, with no more than a few percentage points difference at each income bracket As expected, the population of zero car households is significantly higher for NYC Transit Zone respondents Of this group, 54.7 percent report living in carless households compared to 20.5 percent of households across the entire sample. Time Since Last use of RideHailing and Other Sharing Economy Platforms Table 42 corresponds with survey question 6 in the appendix which asks respondents about their use of shared mobility services and other sharing economy services. This question was posed to all respondents who reported having completed the Crowdsource Reporter task regardless of their residential location. For this question, 507 responses were recorded.
96 Impact of RideHailing on Other Modes and Reasons for Transit Substitution Table 43 corresponds with survey question 11 in the apendix which asks TNC users about changes in their travel behavior since they began using these services. This question was posed to all respondents regardless of geography who reported using a TNC within the last 3 months. Figure 41 below corresponds with survey question 10 in the appendix and indicates how respondents who have used a TNC in the last 3 months would have traveled were a TNC not available at the time, or whether they would not have traveled at all. Of modes being substituted for, taxis and car services were the most frequently substituted for with TNC travel (31 percent), followed by the subway (28.9 percent). Table 44 corresponds with survey question 12 in the appendix which was posed to all respondents who reported using transit less frequently since beginning to use TNCs, and asks about their reasoning for som etimes substituting ride hailing for public transit use. Willingness to Use Autonomous RideHailing Figure 42 corresponds with survey question 18 in the appendix which asks respondents about their willingness to use an autonomous ridehailing service assuming these were shown to be as safe or more so than a traditional TNC, and this resulted in some degree of cost savings. This question was posed to all who have used a TNC in the last three months and to all who indicated their primary reason for not using a TNC was because it was too expensive, not available where they need it, who have used TNCs with friends but have not installed a ridehailing app themselves, or who did not have a smartphone. The latter four groups were included under the assumption they represent potential SAV users. These questions did not account for respondents residential location.
97 Willingness to Pay for TNC Surcharge, DRS, and Autonomous RideHailing Table 45 corresponds with four separate WTP questions. The first row corresponds with survey question 14 in the appendix and asks respondents who have used a TNC in the last 3 months about the amount of tax on a ridehailing trip they would be willing to pay before they considered switching to an alternate mode of transport or not taking the trip. The second row corresponds with survey question 15, and asks respondents how much they would be willing to pay for a DRS trip assuming the same trip taken alone would cost $10 and the DRS trip took no additional time. The third row corre sponds with survey question 16, and asks respondents how much they would be willing to pay for a DRS trip assuming the same trip taken alone would cost $10 and the DRS resulted in a 5 10minute greater trip time. The fourth row corresponds with question 17, and asks respondents how much they would be willing to pay for a DRS trip assuming the same trip taken alone would cost $10 and the DRS resulted in a 10 20minute greater trip time. Questions in the second, third and fourth rows were posed to all who have used a TNC in the last three months and to all who indicated their primary reason for not using a TNC was because it was too expensive, not available where they need it, who have used TNCs with friends but have not installed a ridehailing app themselves or who did not have a smartphone. The fifth row corresponds with question 19, and asks respondents about their willingness to pay for a trip in a self driving Uber assuming the base fare for the same trip in a humandriven Uber was $10, that this trip resulted in no delay when compared to the humandriven Uber, and that this trip was not shared with another passenger. This question was posed to all respondents who indicated in survey question 18 that they definitely would, probably would, or maybe woul d ride in autonomous ridehailing service assuming it
98 was proven as safe or more so than a humandriven vehicle and resulted in some degree of savings. Table 46 corresponds with survey question 20, and asks whether respondents would consider sharing thei r SAV ride with a stranger picked up en route to their destination during the daytime (both by themselves and when traveling with a companion) and during the nighttime (again both by themselves and when traveling with a companion), assuming an additional 30 percent savings. This question was posed to all respondents who indicated in survey question 18 that they definitely would, probably would, or maybe would ride in autonomous ridehailing service assuming it was proven as safe or more so than a humandriv en vehicle and resulted in some degree of savings. Crowdsource Reporter Analysis Locations for all respondents who submitted intersection location information (n=493) are overlaid on a map of Nelson and Raes (2017) trip origin and destination data in Figu re 43 The distribution of the 299 respondents in the primary study area who completed the Crowdsource Reporter task are broken down by ZIP Code group and shown relative to the geographic area corresponding with each of the gr oups and subgroups in Figure 4 4 In a few instances, respondents reported duplicate locations (i.e., there were multiple overlapping intersection location reports associated with the same UserID). In these instances, duplicates were eliminated. After eliminating duplicates, there wer e 128 unique geographic locations in the New York City Transit Zone group, 78 respondents in the LIRR group and 23 in the Metro North group. Figure 45 compares the location of the 128 Transit Zone respondents by walking distance from the intersection nearest to their home to the nearest subway station.
99 Figure 46 shows the location of the 74 Transit Zone respondents who have used a TNC in the last month by subway walk distance. Figure 47 shows the location of New York City respondents who report less subway use since beginning to use TNCs by subway walk distance. Figure 48 shows the location of New York residents by subway walk distance who have used a TNC and who report that they definitely will or probably will use an autonomous ridehailing service, assuming these services were proven to be as safe or more safe than traditional TNCs and use of the autonomous service would result in some undefined amount of savings. Table 47 summarizes the above findings with the number of respondents associated with each subway station walking distance, broken down by the percentage of total respondents for each of the variables in question. Figure 49 compares the distribution of Airbnb hotspots (in magenta), volume of Uber pickups by taxi zone (represented as a red vertical line where greater height indicates a higher volume of pickups ) and by location of likely SAV users where lighter colors indicate shorter walking distances from respondents homes to the nearest subway station and where darker colors indicate longer walk distances. Substitution by Mode and Trip Type During L Train Closure Twelve respondents lived within the L Train ZIP Code subgroup and reported using the L Train to travel to Manhattan at least a little. Importantly, this sample size is too small to be generalizable. Survey question 27 in the appendix asks this group about their intended mode choice for travel between Brooklyn and Manhattan for selected trip types during the time period that the L Train subway line will be closed for 15 months beginning in April 2019. Table 48 displays the results of this question.
100 First/Last Mile Shuttle Use and Support for Commuter Rail Infill Development Table 49 corresponds with survey questions 30 for the MetroNorth ZIP Code group and question 31 for the LIRR ZIP Code group, respectively, and asks respondents who use these commuter rail systems to indicate which mode they typically use to get from their home to the MetroNorth or LIRR station Table 410 concerns respondents WTP for a oneway FLM taxi or shuttle to the subway or commuter rail stations nearest to their homes and back. The variable in the first row corresponds with survey question 25 in the appendix and asks respondents about their willingness to pay for a oneway taxi or shuttle service t hat could bring them to and from the nearest subway station with no more than 5 10 minutes wait time. This question was asked to all respondents in the TransitZone ZIP Code Group who indicated it was neither easy nor difficult, somewhat difficult or extr emely difficult to get to the subway station (survey question 21) and who did not have a physical impediment (survey question 22) and who indicate they would be willing to use such a service at least some of the time (survey question 23). Questions in the second and third rows of Table 410 were asked to respondents who typically drive and park nearby or get a ride/carpool to access the respective commuter rail stations nearest to their homes (survey questions 30/31) and who would use a FLM taxi or shuttle service at least some of the time (survey questions 32/33). Table 41 1 corresponds with questions 36/37, which asks all residents of the LIRR and MetroNorth groups about their willingness to support infill development at the commuter rail stations nearest to their homes. This question employed an interactive slider graphic that allowed users to visualize commuter rail parking lot infill using architectural rendering overlays.
101 TNC Surcharge and Congestion Pricing Support Among CBD Workers/Residents T he firs t row in T able 41 2 corresponds with survey question 39, which asks respondents in the CBD ZIP Code group i.e., residents of the Manhattan CBD and workers who commute to the Manhattan CBD to indicate how much they would support or oppose a $3 TNC surch arge assuming this surcharge would be be used to fund improvements to New York City transit. The second row in Table 41 1 corresponds with survey question 40, which ask respondents in the CBD ZIP Code group to indicate how much they would support or oppose a proposal to toll all vehicles entering congested areas of Manhattan assuming these tolls would be used to fund improvements to New York City transit.
102 Table 4 1. Respondent Demographics: NYC Transit Zone and All New York State. Demographics NYC Transit Zone All New York State % Count % Count Gender Male (18+) Female (18+) 45.2% 54.74% 62 75 43.8% 55.3% 220 278 Race/Ethnicity Non -Hispanic White Hispanic/Latino Black/African-American Asian American Indian/Alaskan Hawaiian/Pacific Islander Two or More Races Other 43.8% 12.4% 19% 16.1% .7% .7% 4.4% 2.9% 60 17 26 22 1 1 6 4 70.5% 7.1% 10% 6.8% .8% .4% 3.6% .8% 354 36 50 34 4 2 18 4 Age 18 24 25 34 35 44 45 54 55 64 65 74 75 84 21.2% 43.1% 17.6% 11.7% 5.8% .7% 29 59 24 16 8 1 15.5% 43.8% 21.1% 10% 8.1% 1.2% .2% 78 220 106 50 41 6 1 Education Less than High School High School Graduate Some College 2 -year Degree 4 -year Degree Professional Degree Doctorate 5.1% 15.3% 6.6% 55.5% 15.3% 2.2% 7 21 9 76 21 3 .2% 6.8% 22.9% 12.6% 39.8% 15.7% 2% 1 34 115 63 200 79 10 Total Household Income Less than $10,000 $10,000 $19,999 $20,000 $29,999 $30,000 $39,999 $40,000 $49,999 $50,000 $59,999 $60,000 $69,999 $70,000 $79,999 $80,000 $89,999 $90,000 $99,999 $100,000 $149,999 More than $150,000 2.9% 8% 6.6% 11% 10.2% 12.4% 8.8% 8% 3.7% 7.3% 12.4% 8.8% 4 11 9 15 14 17 12 11 5 10 17 12 4.4% 7.8% 9.2% 9.8% 11% 8.6% 9.8% 10.2% 4.4% 5.4% 13% 6.8% 22 39 46 49 55 43 49 51 22 27 65 34 Total Cars per Household None 1 2 3 4 or more 54.7% 34.3% 11% 75 47 15 20.5% 33.9% 32.7% 8.6% 4.4% 103 170 164 43 22
103 Table 4 2. Shared mobility and sharing economy (time since last use). Company Week Month 3 Mo. Year 1 5 Yrs. Never Count Uber, Lyft, Juno 3.8% 9.9% 19.3% 12.6% 12.4% 7.3% 34.7% 507 UberPool, Lyft Line, Via 1.8% 3.8% 9.1% 6.5% 5.3% 2.0% 71.6% 507 Airbnb or VRBO 0.8% 1.4% 2.2% 5.1% 17.2% 9.5% 63.9% 507 Zipcar or Car2Go 0.4% 1.0% 0.6% 1.8% 3.8% 6.5% 86.0% 507 Getaround or Turo 0.4% 0.2% 0.2% 0.8% 0.8% 0.8% 96.8% 507 Citibike 0.6% 0.8% 2.4% 3.2% 6.1% 5.7% 81.3% 507 Taskrabbit or Thumbtack 1.0% 0.8% 1.0% 0.8% 2.0% 3.0% 91.5% 507 Table 4 3. Impact of ride hailing on other modes. Much More Somewhat More Same Somewhat Less Much Less Count Go Places 12.8% 21.9% 62.6% 2.3% 1.4% 219 Drive 4.2% 5.2% 56.3% 23.4% 10.9% 192 Subway 3.7% 4.7% 52.1% 24.0% 15.6% 192 Bus 3.1% 3.7% 51.3% 19.4% 22.5% 191 Metro North /LIRR 0.7% 2.7% 69.3% 12.0% 15.3% 150 Walk/Bike 1.9% 6.7% 59.6% 22.6% 9.1% 208
104 Table 4 4. Reasons for substituting ride hailing for transit use. Strongly Agree Somewhat Agree Neither Agree nor Disagree Somewhat Disagree Strongly Disagree Count Services are too slow 41.9% 41.0% 5.7% 9.5% 1.9% 105 Too few stops / stations 25.7% 30.5% 19.1% 15.2% 9.5% 105 Infrequent services 27.6% 36.2% 17.1% 14.3% 4.8% 105 Unreliable services 35.2% 28.6% 14.3% 16.2% 5.7% 105 Too crowded 36.5% 29.8% 16.4% 11.5% 5.8% 104 Safety 8.6% 22.9% 13.3% 32.4% 22.9% 105 Cleanliness 15.2% 22.9% 22.9% 21.9% 17.1% 105 Table 4 5. Willingness to pay for TNC surcharge, DRS and Autonomous Ride Hailing Mean Std. Deviation Variance Base Price Base Price % Change Count Max. TNC surcharge before mode shift $2.56 $2.09 $4.36 +25.6% 226 DRS no delay $4.85 $2.40 $5.74 $10 51.5% 335 DRS 5 10 min delay $3.87 $2.20 $4.85 $10 61.3% 331 DRS 10 20 min delay $2.71 $2.08 $4.32 $10 72.9% 325 SAV w/o DRS (no delay) $6.80 $2.68 $7.21 $10 32% 258 Table 4 6. Variation in willingness to use SAVs depending on time of day. Definitely Yes Probably Yes Might or Might Not Probably Not Definitely Not Count Daytime 27.3% 44.6% 15.4% 8.9% 3.9% 260 Nighttime 15.4% 27.7% 20.8% 22.7% 13.5% 260 Daytime with a companion 35.8% 39.2% 15.0% 6.9% 3.1% 260 Nighttime with a companion 23.9% 32.7% 24.6% 12.7% 6.2% 260
105 Table 4 7. TNC use, mode shift, and likely SAV users by subway walk distance. Subway station walk distance from home 5 min. 10 min. 15 min. 20 min. 25 min. 30 min. Count % of total count All NYC respond. 43.1% 32.5% 11.4% 4.9% 6.5% 1.6% 123 100.0% TNC users 47.3% 31.1% 9.5% 4.1% 6.8% 1.4% 74 60.2% Use subway less since using TNCs 48.6% 31.4% 8.6% 2.9% 5.7% 0.0% 35 28.5% TNC users willing to use SAVs 45.5% 34.8% 10.6% 0.0% 7.6% 1.5% 66 53.7% Table 4 8. L Train substitution by mode and trip type. Trip Type Subway Bus Bike/ Walk TNC Drive Carpool Taxi Count Work 75.0% 8.3% 0.0% 16.7% 0.0% 0.0% 0.0% 12 School 87.5% 0.0% 0.0% 0.0% 12.5% 0.0% 0.0% 8 Shopping 45.5% 9.1% 27.3% 9.1% 9.1% 0.0% 0.0% 11 Personal Business 36.4% 18.2 % 18.2% 9.1% 9.1% 9.1% 0.0% 11 Social/ Recreation 50.0% 16.7 % 8.3% 8.3% 8.3% 0.0% 8.3% 12 Other 40.0% 10.0 % 10.0% 10.0% 10.0% 20.0% 0.0% 10 Table 4 9. Existing means of traveling to commuter rail stations. ZIP Code Group Drive/ Park Get a Ride Bus TNC Walk/ Bike Other FHV Other Count Metro North 45.5% 4.5% 0.0% 0.0% 45.5% 4.5% 0.0% 22 LIRR 55.1% 5.0% 1.5% 2.9% 30.4% 2.9% 0.0% 69
106 Table 4 10. Willingness to pay for FLM shuttles to subway or commuter rail stations. ZIP Code Group Mean Std. Deviation Variance Count FLM Subway Shuttle Transit Zone $2.77 $1.61 $2.59 12 FLM Metro North Shuttle MetroNorth $3.65 $1.64 $2.70 11 FLM LIRR Shuttle LIRR $3.91 $2.13 $4.52 37 Table 4 11. Extent of support for infill development at commuter rail stations. ZIP Code Group Strongly Agree Somewhat Agree Neither Agree or Disagree Somewhat Disagree Strongly Disagree Count LIRR 19.7% 31.6% 25.0% 13.2% 10.5% 76 Metro North 13.0% 30.4% 39.1% 8.7% 8.7% 23 Table 4 12. Extent of support for TNC surcharge and congestion pricing (CBD). ZIP Code Group Strongly Support Moderate Support Neither Support nor Oppose Moderate Oppose Strongly Oppose Count TNC Charge CBD 18.7% 28.6% 13.2% 14.3% 24.2% 91 Congest. charge CBD 16.5% 39.6% 16.45% 4.4% 23.1% 91
107 Figure 41. Modes substituted for with use of ridehailing and ridehailing induced travel demand (232 respondents). Figure 42. Willingness to use an autonomous ridehailing service (338 respondents).
108 Figure 43. Distribution of New York State MTurk respondents relative to major highways and trip origins and destinations ( 2006 2010).
109 Figure 44. Distribution of respondents in the New York Metro area by ZIP Code group and subgroup.
110 Figure 45. Location of New York City respondents by subway station walk distance.
111 Figure 46. New York City respondents who have used r ide hailing within the last month by subway walk distance.
112 Figure 47. New York City respondents who use the subway less since beginning to use ride hailing by subway walk distance.
113 Figure 48. New York City respondents likely to use SAVs by subway walk distance.
114 Figure 49. Distribution of Uber pickups and Airbnb hotspots relative to likely SAV users.
115 CHAPTER 5 DISCUSSION The above findings both support and contradict some of the findings of previous research. Additionally, they address a number of notable gaps and suggest new avenues of research that have yet to be explored in a systematic way. This section will discuss significant findings of the preceding chapter in the context of the literature pertaining to each topic, address the li mitations of this study, and make appropriate recommendations using the selected themes from RPAs (2017a) Fourth Regional Plan as a policy framework Ride Hailing Impacts Compared to previous studies, respondents were found to use ride hailing services at a much higher rate, which could be due to the use of MTurk workers as a sample. This is especially true for respondents in New York City. Gurumurthy et al. (2017) found that 32.5 percent of Americans have personal ridehailing experience, whereas over 65 percent of all study respondents report personal ridehailing experience within the past 5 years (Table 42 ). In New York City, 60 percent of respondents (n=123) report using ridehailing within the past month; presumably, the rate of New York City responde nts who have used ridehailing within the past 5 years is significantly higher. While this finding supports the conclusion of Clewlow and Mishra (2017) and others that ridehailing use is higher in metro areas and urban areas, the extent of use and frequency of use seems to exceed findings on previous cohorts. For example, Clewlow and Mishra (2017) found that just 29 percent of residents in urban areas use ridehailing regularly. Similar to the findings of Rayle et al. (2015), who found
116 that nearly half of all ride hailing trips occurred on evenings and weekends, about 45 percent of respondents reported using ridehailing most frequently after 8pm. Transit While Rayle et al. (2015), Schaller (2017a) and RPA (2017a) discuss the potential for TNCs to extend mobility to areas underserved by transit, and there have been numerous studies that have looked at mode shift from transit as a result of TNC use, there seems to be no academic literature on the relationship between residential transit accessibility and r ide hailing use. The lack of data sharing among TNCs as well as the disaggregated nature of this data when it is shared makes granular spatial analysis of this type difficult. In this regard, the present study has made some intrig uing findings. Table 47 suggests that the closer one lives to a subway station the more likely they are to use TNCs. This higher level of TNC use also correlates with greater declines in subway use. This disparity may correlate with geographic disparities in socioeconomic status, whereby people who are better off can afford to live closer to subway stations where housing costs are higher. The findings of Rayle et al. (2015) and Clewlow and Mishra (2017) on TNC user demographics support the notion that these groups tend to be wealthier and live in more densely urban areas. These wealthier groups with more disposable income may travel more for recreation and leisure during off peak hours where subway service is diminished. Alternately, they may be substituting some peak hour commute t rips where they would otherwise take transit due to overcrowding. Drilling down further into the study results may shed light on the nature of trip substitution and its relation to respondents residential locations. Future research could examine the relat ionship between socioeconomic variables, subway accessibility
117 and perceptions of transit quality as they relate to frequency of ridehailing use and times of the day and week when ridehailing is used. Henao (2017) found that 12.2 percent of trips would not have occurred were TNCs not an option. When asked the same question, just 1.7 percent of respondents to this study reported they would not have travelled. However, nearly 35 percent of TNC users in this study report traveling much more or somewhat more. Studies by Rayle et al. (2015), Schaller (2017a) and others support the conclusion that TNCs are frequently substituted for taxis and other FHVs. Indeed, 31 percent of study respondents reported using TNCs in place of FHVs. Similar to the findings of Clewl ow and Mishra, who found a 6 percent decrease in bus use due to TNCs, this study found that 5.6 percent of TNC rides would have otherwise been taken by bus, and that nearly 40 percent of respondents report using the bus somewhat less or much less since they began using TNCs. Contrary to the findings of Babar and Burtch (2017) and Clewlow and Mishra (2017), however, who found 2.5 7 percent higher use of heavy rail transit systems in aggregate after the arrival of TNCs, this study found that nearly 40 percent of respondents reported less subway use and 27 percent reported less commuter rail use. Likewise, nearly 30 percent of study respondents report decreased bicycling and walking. Notably, 12.6 percent of respondents reported substituting their last TNC tr ip for driving alone and nearly 35 percent of respondents reported driving less since beginning to use TNCs. Substitution for driving alone is presumably occurring with a greater frequency outside of the dense urban areas of the metro area, so this may hav e limited social benefit in New York City where traffic congestion is greatest. However, suburban areas where driving rates are higher may benefit from decreased incidences
118 of drunk driving as the findings of Clewlow and Mishra (2017) and Henao (2017) suggest. Further disaggregation of the data by geography may shed light on this question. Of those who reported diminished transit use since beginning to use TNCs, respondents tended to agree most with the statement that they sometimes substituted TNCs for tra nsit because transit services are too slow, too unreliable, too infrequent and/or too crowded (Table 44 ). Respondents also expressed concern regarding cleanliness and safety of transit services, although to a lesser degree than the above concerns. The ext ent of dissatisfaction with transit may explain why so many more are substituting TNCs for heavy rail transit in the New York metro area than is the case nationally, and why this is especially true for the New York City subway. Mahler (2018), RPA (2017a) and others note the deteriorating quality of service and level of safety on the New York City subway. This seems to support Clewlow and Mishras (2017) and Babar and Burtchs (2017) contention that the substitutive versus complementary nature of transit var ies widely depending on transit quality. Why respondents report diminished commuter rail use since beginning to use TNCs is less clear, however, as the literature supports the idea that TNCs are most complementary to longhaul transit trips (Clewlow & Mish ra, 2017). One possible explanation is that captive riders who may have been using commuter rail for short distance trips in areas where other transit services were lacking may now be using TNCs as an alternate travel mode. However, w hy this would be occur ring in the New York metro area and not in other metro areas is unclear Ride Hailing Countermeasures The scan of literature on TNCs found no research that examined the extent of a surcharge that would be required to induce a shift to alternate modes. Schaller (2017b)
119 speculates that previous research on the effects of taxi surcharges also apply to TNCs, and that a $3 surcharge would have limited ability to deter excessive ridehailing use. His logic is that where the option to travel by subway exists, FHVs are premium travel modes. Therefore, he contends, fare prices are rather inelastic in their ability to induce mode shift as the typical clientele of these services are presumably well off. However, the findings of this study would seem to suggest this is not the case, and that fare prices are elastic with respect to their ability to induce mode shift. When the 226 study respondents who currently use TNCs were asked about the most they would be willing to pay for a surcharge before they considered traveling by an alternate mode or not traveling at all, the study found a mean value of $2.56. This suggests a relatively small price increase is needed to decrease overall TNC use. However, applying such a surcharge across all geographies, irrespective of the availability of alternate modes, may unfairly penalize users on the urban periphery with limited mobility and limited means. Even without this surcharge, however, the longterm sustainability of current pricing models is questionable given TNCs sustained financial losses and the sharp price drops that have coincided with recent ridership gains. Regarding support to implement such a surcharge among those most affected (i.e., Manhattan CBD residents and commuters), opinions seem to be split. Roughly 47 percent of respondents would support a $3 surcharge if it were used to fund improvements to New York City transit, while 38.5 would oppose it. However, approximately 6 per cent more respondents strongly oppose than strongly support such an initiative. Support for congestion charging, which would toll all vehicles entering congested areas of Manhattan, has greater overall support with 56 percent of the CBD
120 ZIP Code subgroup s upporting such an initiative to some degree. However, the number who strongly oppose this initiative still outweigh the number who strongly support it. Further disaggregation of this group by respondent geography and primary travel mode may shed further li ght on the underlying reasons for this variance. Dynamic Ridesharing and Autonomous RideHailing Findings on respondents willingness to pay for DRS, even in cases where no delay is incurred, should give pause to those such as Alonso Mora et al. (2017) who envision a future where highcapacity autonomous DRS solves much of the traffic and emission problems in the CBD by aggregating overlapping trips. Respondents are willing to pay less than half as much for DRS as they would be for the same trip that is not shared with another passenger, even when the shared trip incurs no additional delay (see Table 45 ). A 5 10minute delay reduces WTP by an additional 10 percent when compared to the hypothetical $10 base fare, while a 10 20minute delay reduces WTP by roughly 10 percent further. Assuming a 50 70 percent reduction off a fare that may already be artificially low due to fierce price competition and large venture capital investments (Solomon, 2016), it may be exceedingly difficult for TNCs to make the D RS model financially sustainable in the long term unless they can reach sufficiently high volumes. Present traffic levels in the CBD would be a significant hindrance to the efficiencies required by such a model. However, if supported by policies such as cordon charging that made travel in areas progressively more expensive as levels of traffic increased, this model could become more feasible. Previous studies on SAV sentiment, both in the solo travel context and with SAV DRS, have surveyed the general population with mixed and varying results (Gurumurthy & Kockelman, 2017; Quarles & Kockelman, 2017; Gurumurthy, Kockelman, & Hahm,
121 2017). However, some of these same studies observe that attitudes are likely to evolve with exposure. Given less than a third of Americans have personal experience with ridehailing services it may be too much of a conceptual leap for many to imagine a future where they neither drive or own the vehicle they travel in on a daily basis. The literature on cognitive factors and modalit y styles by Pakusch and Bossauer (2016), Krueger et al. (2016) and Yap et al. (2016) support the idea that receptivity will vary considerably with prevailing travel modes, exposure to similar economic models, and age. Schaller (2017a) contends that most of what will be appealing about SAV and SAV DRS already exists in todays TNC. Therefore, it makes most sense to survey individuals who are already using TNCs and accustomed to shared use economic models; who are conditioned to traveling by modes that do not offer a locus of control or satisfy sensation seeking behaviors (i.e., transit); and who live in areas of sufficient density and where shared mobility models have proven successful. In all likelihood, the cohort of young, educated urbanites who have embraced shared mobility and the wider sharing economy will be among the first to embrace autonomous ridehailing and highcapacity DRS, especially in areas where transit use is high. New York City offers an ideal population in many of these regards. The survey results support many of the above findings and assumptions. Only 23 percent of respondents said they would not use an autonomous ridehailing service assuming they were shown to be as safe or more so than a traditional TNC and offered some degree of cost savings. The majority (57.7 percent) said they probably or definitely would use an autonomous TNC in these circumstances. Looking specifically at New York City, where 60 percent of all respondents reported using a TNC in the past
122 month, 89 percent of TNC users probably or definitely would use an autonomous TNC in the above circumstances. Similar to the phenomena observed with TNC use and mode shift from the subway to TNCs, respondents who live closest to subway stations and who currently use TNCs are proportionately more likely to use SAVs (94.7 percent) than groups who live farther away from the subway. This may be due to the same set of socioeconomic factors driving greater TNC use among this group. Further research may shed light on these relationships. When comparing variations in respondent willingness to use SAV DRS based on time of day to the findings of Gurumurthy and Kockelman (2017), the distinctions between the general U.S. population and the sample of current TNC users becomes clear just 4.4 percent of Americans would be willing to use SAV enabled DRS at night with a rider unknown to them, while 63.8 percent of respondents to this study said they definitely, probably, or maybe would be willing to do this (assuming the DRS option resulted in 30 percent fare savings when compared to solo travel by SAV). The above observations seem to support findings of the literature on the influence of modality style, cognitive factors, geography, age and socioeconomic status on receptivity to SAV and SAV DRS. Q ualitative analysis of likely SAV users when compared to the location of Uber pickups and Airbnb hotspots suggests a possible correlation between shared mobility use, the wider sharing economy and receptivity to SAV and SAV DRS. Roughly half of likely SAV users live within an Airbnb hotspots. Additionally, the proximity analysis shows significant spatial coincidence of Airbnb locations and Uber pickups. Given all respondents are MTurk workers, and all participate in the gig economy, the labor
123 market count erpart to the sharing economy, it is not surprising that study respondents use TNCs at nearly twice the rate of the general population and report considerable use of services like bikeshare (almost 18 percent of all respondents), carshare (14 percent of al l respondents), and Airbnb or VRBO (36 percent of all respondents). Where participation in these services and physical location of these services coincide in space, as demonstrated in Figure 49 it may be possible to predict areas that will be among the f irst to adopt new sharing economy and shared mobility innovations like SAV and SAV DRS, and also where competition with TNCs for choice transit riders may be highest. While the proximity analysis conducted for this study lacks sufficient rigor to propose any such relationship with confidence, it does suggest something that transcends population density and economic activity alone. Further research could explore demographics of these areas and whether they aligned to some degree with the socioeconomic, demographic, modality styles, and cognitive factors believed to correlate with willingness to use SAVs. The spatial coincidence of the L Train with Airbnb hotspots, Uber pickups and a significant number of likely SAV users suggests the forthcoming shutdown of this subway line may offer an ideal proving ground for highcapacity DRS and/or SAVs. TNCs may have their eye on this development. Access to their proprietary trip data would give a more complete picture regarding the extent of ridehailing activity in the L Train walkshed. Schallers (2017a) research suggests that a higher number of dropoffs occur in these neighborhoods than the pickup data would imply. L Train Substitution by Mode and Trip Type During Planned Closure Just 12 responses were recorded for the quest ion corresponding with Table 48 which asks how Brooklynites who live within a 30minute walk of the L Train line and
124 who use the L Train to travel to Manhattan at least a little intend to travel during the planned closure of this subway line. Given this small sample size, and potential demographic bias, these results should not be considered statistically valid. However, if the population of this area share similar demographic characteristics and travel preferences with those of the wider group of MTurk respondents who live in New York City, who exhibit high rates of ride hailing use compared to the US population, some degree of substitution with auto modes should be expected. The New York City Department of Transportation Commissioner anticipates that during the planned 15month closure of the L Train line, 80 percent of all 400,000 daily riders will shift to alternate subway lines, 15 percent will shift to buses and most of the remaining five percent will shift to walking or biking (Nir, 2017) However, t here is no mention of a shift towards auto modes like TNCs or other FHVs. Nevertheless, the City does plan to implement a threepassenger vehicle occupancy minimum on the affected bridge crossing (Nir, 2017). While respondents who lived within the 30minute L Train walkshed and who reported using the L Train to travel to Manhattan at least somet ime of the time is too small to be valid for research purposes, there are nonetheless some interesting findings. For example, nearly 17 percent of respondents report they plan to use TNCs to commute to work, while approximately 10 percent plan to shift to driving alone for school, shopping, personal business and other trip types where they would normally use the L Train to travel. The Citys assumption th at no significant auto mode shift will occur as a result of the planned closure of the L Train is questionable.
125 First Last Mile Transit Connections and Support for Infill Development The survey instrument did not explicitly state that ondemand FLM taxis or shuttles would be autonomous or that they would even be TNCs. Rather, the study made the assumption that if driverless technologies are shown to be as safe or more so than traditional vehicles, that growing exposure to these technologies will increase f amiliarity and comfort with them. Pakusch and Bossaeur (2017) contend that prior experience with technology tends to increase acceptance of that technology and the above findings on ride hailing users willingness to use autonomous ridehailing services se em to support this notion. Of the 12 New York City respondents with limited transit accessibility who would use the subway more frequently if the MTA provided such a service, average willing ness to pay for a oneway trip that would connect from their homes to the nearest subway station and vice versa is $2.77. However, the number of respondents to this question is too small for this finding statistically significant. Of the 11 respondents in the MetroNorth ZIP Code group who currently carpool to get to the nearest commuter rail station or drive and park and nearby all would use an FLM service at least some of the time in lieu of the mode they currently use to get to the MetroNorth station. Mean W TP for such a service among this group is $3.65. Likewise, the number of respondents to this question is too small for this finding statistically significant. Of the 37 respondents in the LIRR ZIP Code group who were asked the same question and who met the same screening criteria, all but six indicated they would use such a service at least some of the time. Mean WTP for this service among this group is $3.91. Regarding extent of support for infill development at suburban commuter rail stations, just 23.8 percent of the LIRR ZIP Code group would disagree with this
126 proposal while 17.4 percent of the Metro North ZIP Code group would disagree. When asked why they would not support such a proposal, respondents indicated that concerns about use of public funds would top the list, followed by concerns about changes to the character of the neighborhood, concerns about traffic, and concerns about parking availability. Given the extent of support for infill development and willingness to use FLM shuttles as a substi tute for parking provided they were offered at a moderate price, P3s with TNCs, especially where SAV and SAV DRS lowered operating costs could offer a valuable opportunity to address regional affordable housing shortages. Real estate value capture on MTA owned property currently being used for parking could help offset the cost of TNC/SAV subsidies and/or the cost of subsidizing a ffordable units (Mahler, 2018). Limitations The results of the study present a number of significant limitations. First, 23 days was not a sufficient amount of time to reach the target sample size of 610. This suggests one of two things: (1) the population of MTurk workers who live in New York State is not much bigger than the number of responses collected (n=502); or (2) the requi rement that re spondents tag the intersection location nearest to their home ( as shown in Figure 37 ) dissuaded many from participating. Second, the geographic groups and subgroups failed to collect sufficiently representative sample sizes, especially with respect to the hyperlocal questions (e.g., the L Train ZIP Code subgroup or respondents on the periphery of New York City limits with limited subway accessibility ) Third, the sample demographics skewed young, white, welleducated, wealthy and female. Given this, the sample is a convenience sample rather than a representative one and the findings have limited generalizability.
127 For the sake of surveying TNC users potential SAV users and sharing economy participants however, the MTurk sample offers a fairly ideal demographic Ride hailing use among MTurk respondents who reside in New York City is nearly double the percentage reported by Clewlow and Mishra (2017) in their survey of urban and suburban populations in seven major U.S. cities. The sample demographics across all geographies align closely with Rayle et al (2015) and Clewlow and Mishras findings that TNC users tend to be young, collegeeducated and affluent. In addition to these demographic and socioeconomic characteristics, the sample is disproportionately multimodal, with over 20 percent of all respondents reporting that they live in zerocar households. Quarles and Koc kel man (2016) and Krueger et al. (2016) find that young, welleducated multimodals are among the most receptive to SAVs and t hat this cohort is likely to be among the first to adopt these technologies. The sample demographics also align closely with the cohort anticipated to be most receptive to SAV use. Additionally, since all MTurk workers participate in the gig economy, a close cousin to the sharing economy, they may offer an ideal sample for studying sharing economy phenomena (the high rate of participation in other sharing economy services seems to support this assumption). In addition to the limitations of the sampling met hodology, the analysis itself presents limitations. Because the study was by nature broad and exploratory, a format supported by the survey based methodology, there was not sufficient time to thoroughly explore any one area. That said, while the hyperlocal questions presented sampling challenges, they largely eliminated the need for inferential analysis, which comes with its own set of assumptions. Additionally, the wide net this study cast did not allow for
128 sufficient time to clean the entire data set e. g., eliminating responses that suggest insufficient engagement or to disaggregate data where there was conflation. For example, the WTP results could be disaggregated by those in the New York metro area and those outside and could be further disaggregated by those within New York City and those who live in the surrounding suburbs (i.e., the LIRR and MetroNorth ZIP Code groups). This would not be hard to do, but given the scope of this analysis there was not sufficient time to drill down further in each ar ea. While the study has created a novel and promising methodology for survey research with a geographic component, and has also addressed a number of research gaps, the MTurk sampling methodology has its limitations. Further research could build upon the methods and results of this study. Recommendations While it is important to consider the studys limitations, results of the research have a number of important implications for planners and policy makers in the New York metro area. These both support and challenge some of the recommendations made by RPA (2017a) in the Fourth Regional Plan. This section will make recommendations regarding the four major themes explored in this manuscript i.e., the regions subway problem, congestion problem, affordabl e housing shortages and the potential for emerging transportation technologies to impact these issues in light of the literature in each area and the findings of this study. Reinforcing the S ubway and E mbracing M obility as a Service Schaller (2017a) obs erves that the fundamental challenge for the regions policy makers is to switch growth back to a transit oriented pattern while retaining the mobility gains offered by ridehailing. Results of the survey show broad dissatisfaction with the state of the subway, with a high number of respondents indicating they are switching to
129 alternate modes because it is too slow, too unreliable, too infrequent and/or too crowded. This likely accounts for some of the declining ridership and simultaneous growth in ride hailing. Yet, while ride hailing may serve as a release valve for peak hour overcrowding and fill service gaps during off peak hours, substitution for auto modes increases above ground congestion. This growth pattern is neither fiscally or environment ally sustainable. Previous research has found that ridehailing plays a complementary role with respect to heavy rail transit. Yet, this study found the opposite. Many respondents report using the subway less frequently since they began to use TNCs. With the expected cost and efficiency gains of automation, this competition is likely to increase, especially for riders who can afford alternatives that offer more direct, reliable, and faster services. Given this, it is imperative that the MTA reinforce the c ore functionality of the subway by prioritizing backlogs of def erred maintenance, increasing throughput, upgrading the systems signals, and modernizing subway stations These fixes alone will cost tens of billions of dollars. This sum seems daunting, however peer cities like Los Angeles have undertaken even more financially ambitious projects. Unlike LA, however, where traffic is spread throughout the region and LA County voters supported a referendum to address the issue, the subway system is reliant on funding from the state, which is subject to the political whims of many who are not directly affected by deteriorating subway conditions or localized traffic hotspots. Therefore, it is important that the city exercise fiscal constraint and focus limited funding on addressing the most pressing needs. Given this and the extraordinarily high cost of building new track miles in the city, it does not seem wise to pursue RPAs
130 recommendation to expand the subway into underserved low income areas, which could also have the unintended consequence of displacing those residents. Nor should the city spend limited funds on highprofile ribbon cuttings with limited mobility outcomes like the Oculus the $4 billion train station built at the Ground Zero site. Rather, low income areas currently underserved by transit can be served by partially subsidized FLM shuttles that use integrated fare payment to connect riders with trunk line bus and subway lines In the long run, an autonomous FLM shuttle may serve these needs at a relatively moderate cost. Respondents in underserved areas expressed a desire to use FLM shuttles and a willingness to pay for them. Further steps may be required to fix the subway while maintaining its longterm financial health. The city is famous for its 24/7 subway service, but this schedule makes regular maintenance difficult. It may prove more cost effective to close the subway during hours of low ridership (e.g., Monday Thursday from 2am 5am) and provide subsidize d ride hailing as an alternativ e. Given 60 percent of survey respondents use a TNC at least once a month, 54 percent are open to using SAVs assuming they are shown to be safe, and 43 percent are willing to share their SAV ride at night with a stranger (a number that is likely to grow wi th shifting social norms), this seems like a reasonable solution. The L Train shutdown may provide an opportune moment for the agency to pilot such a P3 that serves affected residents needs through DRS or SAV or some combination of the two. One challenge this poses however is with Title VI requirements, where customers may lack smartphones that would allow them to hail a TNC or where language barriers prevent their use.
131 Shifting the MTAs focus from transit provider to mobility provider may require some fundamental changes to the agency. Given the governance and structural issues that have plagued the agency for decades, an overhaul may be in order. Further, if the region is to remain economically competitive and stem the tide of out migration, it may need to reform the political structure that yokes the subways finances to residents who benefit only indirectly from its continued good order and who may not see this as bei ng the case. Implementing Congestion Charging to Reduce Traffic and Fund Transit New Yorkers have reaped a number of important benefits from TNCs. However, their unchecked growth poses a serious issue for the region. Importantly, these companies have deep pockets and are more invested in short term financial gain then the longterm health of the regions transportation system or environment. The question of whether current TNC business models and fare structures are sustainable is debatable, as is the pace and extent to which these companies will deploy autonomous driving technologies and the impacts that may result Regardless, the city and region cannot afford to take a wait andsee approach. Policy makers can offset fares that may be artificiall y low and that are inducing TNC demand by charging an appropriate user fee to disincentivize use. So as not to unfairly target TNCs, these fares could adjust dynamically to incentivize DRS rides over singlepassenger ones and only apply for pickups and drop offs in congested areas The research suggests that a $2. 75 surcharge is sufficient to shift travel back to nonauto modes. These funds could also be used to pay for muchneeded transit improvements. While past initiatives that have looked to penalize v ehicles for traveling in congested areas have failed, the results of the survey indicate solid support for both a
132 TNC surcharge and for CBD congestion pricing that would be applied broadly. These findings should strengthen the r esolve of the Fix NYC commit tee, which is currently investigating such measures (Mahler, 2018). Use Shared Mobility to Open Opportunities for Affordable Housing Infill The third and final of the major themes from RPA (2017a) explored in this manuscript concerns the regions chronic affordable housing shortages. Building mixeduse infill at the regions commuter rail station with a minimum allotment of affordable units could help to address this issue while ensuring residents have access to New York Citys economic opportunities and other amenities. The results of the research support RPAs (2017c) recommendation that planners should pursue these opportunities. A majority of those who currently drive and park at commuter rail stations or who carpool to the station would be willing to use an FLM instead at least some of the time. On average, these individuals would be willing to pay $3.50 $4 for a oneway service that shuttled them to and from their homes to the nearest LIRR or MetroNorth station. The results suggest less than 25 percent of either the LIRR or MetroNorth ZIP Code groups would actively oppose such an infill initiative. Of those who would be opposed, it may be possible to allay concerns regarding parking, traffic and use of public funds by instituting value capture and providing these partially subsidized FLM shuttles that featured integrated fare payment This could make the need for parking around stations largely obsolete and simultaneously reduce congestion in the surrounding areas especially if these shut tles were DRSenabled. Real estate value capture could offset some of the costs of providing these shuttles. The efficiencies promised by autom ated driving technologies, especially where these are DRS enabled, could make this a cost eff ective and sustainab le approach. Further, by making more
133 productive use of urban land, these dense mixeduse transit oriented developments could stimulate local economies if developed as attractive, pedestrianoriented environments.
134 CHAPTER 6 CONCLUSION The purpose of this s tudy was to provide insight into a number of the New York metro areas most pressing transportation and housing issues and to anticipate some of the impacts of emerging transportation technologies in these areas using RPAs (2017a) Fourth Regional Plan as a policy framework. For the regions long term economic health and the wellbeing of its residents, it is imperative that policy makers and planners find solutions to these issues. In this respect, the study addresses a number of important research gaps. The study also sought to contribute to the literature on TNC impacts and SAVs by addressing gaps in contemporary knowledge. While the New York metro area presents a context that is unique, some of the studys findings in these areas have significant implic ations for other large cities with heavy rail transit. Finally, the study sought to validate the use of MTurk for the study of geographic phenomena, and for transportation and planning use more generally, and developed a novel methodology in order to test this use. In many of these respects, the study was a success. Some of the most significant findings of this research include the following: Contrary to the findings of national studies from Clewlow and Mishra (2017) and Babar and Burtch (2017), this study finds that TNCs are widely being substituted for heavy rail transit. However, this seems to validate these researchers claim that transit quality serves as a moderating variable. Overall ride hailing use, substitution of subway use with TNCs, and recepti vity to using autonomous ridehailing services all correlates positively with residential proximity to subway stations (i.e., the closer one lives to a subway station the greater likelihood they are to do the above). No previous studies have look ed at this specific issue, however, this fi nding would seem to support the notion that the benefits of TNCs accrue disproportionately to the wealthy and that competition for higher income subway riders is significant and likely to increase with SAVs.
135 By addressing t he magnitude of surcharge required to induce a mode shift away from ride hailing this study provides insight both to the regions policy makers and more broadly on the price sensitivity of ridehailing fares. Proximity analysis of shared mobility and wider sharing economy participation seem to coincide spatially, suggesting the presence of sharing economy hotspots. Many of these areas are adjacent to the L Train. The impending shutdown of this line will create a significant mobility vacuum and ideal conditi ons for testing novel shared mobility models. Presently this study appears to be the only to study the attitudes of urban multimodal TNC users towards autonomous ridehailing. The literature on modality styles, cognitive factors, and higher order orientat ions towards vehicle ownership suggest this demographic is significantly more likely to adopt these services than the general population. Results imply most urban multimodal TNC users are willing to use an autonomous ridehailing service providing they are shown to be reasonably safe and lead to some degree of cost savings. The study also shows much higher willingness to use DRS enabled SAVs in a variety of scenarios than Gurumurthy and Kockelmans (2017) study of the general population. Residents of the New York metro area who live adjacent to the regions commuter rails are receptive to using FLM shuttles to get to these stations in place of driving and parking. They are also receptive to mixeduse infill development in these parking lots. As deployment of SAVs and DRSenabled SAVs make widespread use of FLM shuttles to heavy rail transit increasingly feasible, the regions planners and policy makers should begin to think about how best to leverage these development opportunities. While this study addressed a number of significant research gaps, it had two primary limitations: (1) a small sample size; and (2) limitations to disaggregating the data geographically and analyzing it accordingly. The first of these could be addressed by continuing to collect data via the New York State MTurk population (assuming a sufficiently large sample could be collected over time). Alternately, the geographically targeted survey methodology could be applied to an alternate sampling methodology (e.g., another paid study cohort in the New York metro area). Another approach would be to administer the study in Boston, Washington, D.C., Chicago and other cities that have similar densities and transportation networks. Using the methodology developed in
136 this study, such a survey could be deployed relatively quickly and much less cost than comparable paid survey platforms. While the demographics of the New York State MTurk population are biased, they present an ideal group for studying shared mobility, given demographics similarities to current TNC users. The findings of this study support that connection. This methodology shows much promise for researchers exploring the impacts and geography of shared mobility and emerging transportation technologies
137 APPENDIX SURVEY INSTRUMENT 1 2 3 4
139 6 7 8 9
140 10 11 12
141 13 14 15 16
142 17 18 19 20
143 21 22 23 2 4 25 26
144 27 28 29 30
145 31 32 33 34 35
146 36 + 37 38 39
147 40 41 42 43 44
148 45 46 47
149 LIST OF REFERENCES Alonso Mora, J., Samaranayake, S., Wallar, A., Fra zzoli, E., & Rus, D. (2017). On demand highcapacity ridesharing via dynamic tripvehicle assignment. Proceedings of the National Academy of Sciences 201611675. Babar, Y., & Burtch, G. (2017). Examining the impact of ridehailing services on public transit use. SSRN. Retrieved from https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3042805 Casler, K., Bickel, L ., & Hackett, E. (2013). Separate but equal? A comparison of participants and data gathered via Amazons MTurk, social media, and faceto face behavioral testing. Computers in Human Behavior 29(6), 21562160. City of New York. (n.d.). NYC open data [GIS d ata] Retrieved from https://opendata.cityofnewyork.us/ Clewlow, R. R., & Mishra, G. S. (2017). Disruptive tr ansportation: The adoption, utilization, and impacts of rideh ailing in the United States (No. U CDITS RR1707). Research Report UCDITS RR1707, Davis: University of California. Esri. (n.d.). Living atlas of the world [GIS data]. Retrieved from https://livingatlas.arcgis.com/en/ Feigon, S. & Murphy, C. (2016). Shared mobility and the transformation of public transit. TCRP Report 188. Washington, DC: Transit Cooperative Research Program. FiveThirtyEight. (2016). Uber TLC FOIL response [Data set]. Retrieved from https://github.com/fivethirtyeight/uber tlc foilresponse/tree/master/uber trip data Ge, Y., Knittel, C. R., MacKenzie, D., & Zoepf, S. (2016). Racial and gender discrimination in transportation network companies (No. w22776). National Bureau of Economic Research. GIS Lab at the Newman Library of Baruch College, City University of New York (CUNY). (n.d.). NYC mass transit spatial layers Retrieved from https://www.baruch.cuny.edu/confluence/display/geoportal/NYC+Mass+Transit+ Spatial+Layers Guo, Z., Zhao, J., Whong, C., Mishra, P., & Wyman, L. (2017). Redesigning subway map to mitigate bottleneck congestion: An experiment in Washington DC using Mechanical Turk. Transportation Research Part A 106158169. doi:10.1016/j.tra.2017.09.017 Gurumurthy, K.M., & Kockleman, K.M. (2017). Analyzing the dynamic ridesharing potential for shared autonom ous vehicles using cellphone data from Orlando, Florida. Manuscript submitted for publication.
150 Gurumurthy, K.M., Kockleman, K.M. & Hahm, H. (2017). Deeper understanding of American s autonomous vehicle preferences: Questions of longdistance travel, ride sharing, privacy, & crash ethics Manuscript submitted for publication. Harrison, D. (2017, August 12). America s buses lose riders, imperiling their future. The Wall Street Journal. Retrieved from https://www.wsj.com/articles/americas city buses lose momentum 1502539200 Hauser, D. J., & Schwarz, N. (2016). Attentive t urkers: MTurk participants perform better on o nline attention checks than do subject pool participants. Behavior research methods 48(1), 400407. Henao, A. (2017). Impacts of ridesourcing Lyft and Uber on transportation including VMT, mode replacement, parking, and travel behavior (doctoral dissertation) University of Colorado Denver, Denver, Colorado. Hu, W. (2017, December 26). Your Uber car creates congestion. Should you pay a fee to ride? New York Times. Retrieved from https://www.nytimes.com/2017/12/26/nyregion/uber car congestionpricing nyc.html Huff, C., & Tingley, D. (2015). Who are these people? Evaluating the demographic characteristics and political preferences of MTurk survey respondents Research & Politics 2 (3), 2053168015604648. INRIX. (2017, September 27). INRIX identifies the worst traffic hotspots in the 25 most congested U.S. cities [Press release]. Retrieved from http://inrix.com/press releases/us hotspots/ Isaac, M. (2017, November 20). Uber strikes deal with Volvo to bring self driving cars to its network. New York Times. Retrieved from https://www.nytimes.com/2017/11/20/technology/uber deal volvo self driving cars .html?_r=0 Kees, J., Berry, C., Burton, S., & Sheehan, K. (2017). An analysis of data quality: professional panels, student subject pools, and A mazon s Mechanical Turk. Journal of Advertising, 46(1), 141155. Krupa, J. S., Rizzo, D. M., Eppstein, M. J., Brad Lanute, D., Gaalema, D. E., Lakkaraju, K., & Warrender, C. E. (2014). Analysis of a consumer survey on plugin hybrid electric vehicles. Transportation Research Part A 641431. doi:10.1016/j.tra.2014.02.019 Krueger, R., Rashidi, T. H., & Rose, J. M. (2016). Preferences for shared autonomous vehicles Transportation Research Part C: Emerging Technologies, 69, 343355.
151 Lu, Z., Du, R., Dunham Jones, E., Park, H., & Crittenden, J. (2017). Data enabled public preferences inform integration of autonomous vehicles with transit oriented development in Atlanta. Cities 63, 118127. Mahler, J. (2018, January 3). The case for the subway. New York Times. Retrieved from https://www.nytimes.com/2018/01/03/magazine/subway new york city public transportationwealth inequ ality.html Masoud, N., Nam, D., Yu, J., & Jayakrishnan, R. (2017). Promoting peer to peer ridesharing services as transit system f eeders. Transportation Research Record. Meyer, D. (2017, February 2). TLC votes to require Uber and Lyft to disclose trip data. Streetsblog NYC. Retrieved from https://nyc.streetsblog.org/2017/02/02/tlc votes to requireuber andlyft to disclose trip data/ Nelson, G .G.D. & Rae, A. (2017, January 27 ). United States commutes and megaregions for GIS [GIS Data]. Retrieved from https://figshare.com/articles/United_Sta tes_Commutes_and_Megaregions_data_ for_GIS/4110156 Nir, S.M. (2017, Dec 13). Rerouting thousands: City plans for L Train closure. New York Times. Retrieved from: ht tps://www.nytimes.com/2017/12/13/nyregion/l train closing plan.html?_r=0 OpenStreetMap. (n.d.). Relation: New York (61320) [GIS Data]. Retrieved from https://www.openstreetmap.org/relation/61320 map=10/41.1078/ 73.8817 Pakusch, C., & Bossauer, P. (2017). User acceptance of fully autonomous public transport. Proceedings of the 14th International Joint Conference on E Business and Telecommunications (Vol. 4, pp. 5260). Pendall, R., Blumenberg, E., & Dawkins, C. (2016). What if cities c ombined c ar b ased solutions with t ransit to i mprove a ccess to o pportunity?. The Urban Institute, Metropolitan Housing and Communities Policy Center. Quarles, N. & Kock elman, K.M. (2017). Americans plans for acquiring and using electric, shared and self driving vehicles Manuscript submitted for publication. 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, 168178. Regional Plan Association (RPA). (2017a). The fourth regional plan: Making the region work for all of us. New York, NY: Regional Plan Association. Regional Plan A ssociation (RPA). (2017b). New mobility: Autonomous vehicles and the region. New York, NY: Regional Plan Association.
152 Regional Plan Association (RPA). (2017c). Untapped potential: Opportunities for affordable homes and neighborhoods near transit New York NY: Regional Plan Association. San Francisco County Transportation Authority (SFCTA). (2017). TNCs today: A profile of San Francisco transportation network company activity. San Francisco, CA: San Francisco County Transportation Authority. Sarriera J. M., lvarez, G. E., Blynn, K., Alesbury, A., Scully, T., & Zhao, J. (2017). To share or n ot t o share. Transportation Research Record: Journal of the Transportation Research Board, 2605, 109117. doi:10.3141/260511 Society of Automotive Engineers International (SAE). (2016, September 30). Taxonomy and definitions for terms related to driving automation systems for onroad motor vehicles (J3016_201609). Retrieved from https://www.sae.org/standards/content/j3016_201609/ Schaller, B. (2017a). UNSUSTAINABLE? The growth of appbased ride services and traffic, travel and the future of New York City Brooklyn, NY: Schaller Consulting. Schaller, B. (2017b). Empty seats, fu ll streets: Fixing Manhattans traffic problem Brooklyn, NY: Schaller Consulting. Slee, T. (2017). Airbnb data collection: Get the data [Data set]. Retrieved from http://tomslee.net/Airbnb datacollectionget thedata Solomon, B. (2016, January 25). Is Uber trying to kill Lyft with a price war? Forbes. Retrieved from https://www.forbes.com/sites/briansolomon/2016/01/25/is uber trying to killlyft with a price war/ 3a6c53446573 Transit Center. (2017, December 18). Unsustainable: Traffic 2018 [Video file] Retrieved from https://vimeo.com/247866390 Tsay, S., Accuardi, Z., Schaller, B. & Hovenkotter, K. (2016). Private mobility, public interest: How public agencies can work with emerging mobility providers New York, NY: Transit Center. Uber. (n.d.). The Uber story Ret rieved January 4, 2018, from https://www.uber.com/ourstory/ Vanette, D. ( 2017, June 28). Using attention checks in your surveys may harm data quality. Qualtrics [Blog post]. Retrieved from https://www.qualtrics.com/blog/usingattentionchecksin your surveys may harm dataquality/ Xu, J. (2017). Map Sensitivity vs. m ap d ependency: A case study of subway m aps i mpact on p assenger r oute choices in Washington DC. Behavioral Sciences (2076328X) 7 (4), 1 14. doi:10.3390/bs7040072
153 Yap, M. D., Correia, G., & Van Arem, B. (2016). Preferences of travellers for using automated vehicles as last mile public tr ansport of multimodal train trips. Transportation Research Part A: Policy and Practice, 94, 11 6. Zhao, J., & Lee, J. (2013). Experience or memory: Happiness of commuting (No. 131243).
154 BIOGRAPHICAL SKETCH Leslie Brown has a Bachelor of Arts in h is tory from the University of California, Santa Barbara. He is currently completing his Master of Urban and Regional Planning at the University of Florida with specializations in l and u se and t ransportation, and i nformation t echnologies for p lanning. His academic research has focused on examining the impacts of emerging transportation technologies. As a graduate research assistant in the University of Florida Center for H ealth and the Built Environment, he has worked in a number of areas including the tran sportation and residential location preferences of the millennial generation; rural Safe Routes to School implementation; and bicycling in active retirement communities. While a student, Leslie did a summer internship with Renaiss ance Planning in Tampa and served on the City of Gainesvilles Bicycle and Pedestrian Advisory Board. He was selected for a number of prest igious awards including the American Institute of Certified Planners Outstanding Student Award and was granted an Eno Fellowship from the Eno Center for Transportation. Prior to pursuing a masters degree, he worked with Sam Schwartz Engineering, the Institute for Transportation and Development Policy, and Transportation Alternatives in New York City, where he lived for seven years. He is currently working as a transportation planner and policy analyst with ICF in Washington, D C whe re he lives with his wife Jill.