Citation
Impacts of ridesourcing - Lyft and Uber - on transportation including VMT, mode-replacement, parking, and travel behavior

Material Information

Title:
Impacts of ridesourcing - Lyft and Uber - on transportation including VMT, mode-replacement, parking, and travel behavior
Creator:
Henao, Alejandro ( author )
Language:
English
Physical Description:
1 electronic file (121ges) : ;

Subjects

Subjects / Keywords:
Local transit -- Planning -- United States ( lcsh )
Local transit -- Social aspects -- United States ( lcsh )
Genre:
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )

Notes

Review:
The transportation sector is currently experiencing a disruption with the introduction and evolution of technology and transportation services such as bikesharing, carsharing, on-demand ridesourcing (e.g. Lyft, Uber), and microtransit (e.g. Bridj, Chariot). As these new layers of technology-based transportation options begin to flourish, it is important to understand how they affect our transportation systems and society. This doctoral dissertation analyzes the impacts of ridesourcing on several areas of transportation including: efficiency in terms of distance – Vehicles Miles Traveled (VMT) versus Passenger Miles Traveled (PMT) – and travel times, mode replacement, VMT increase, parking, transportation equity, and travel behavior. Realizing the difficulty in obtaining data directly from Lyft and Uber, this research employs an innovative approach by the author becoming an independent contractor to drive for both companies; this allowed the author to gain access to exclusive data and real-time passenger feedback. The datasets include actual travel attributes – such as times, distances, and earnings – from 416 rides (Lyft, UberX, LyftLine, and UberPool), and travel behavior and socio-demographics from 311 passenger interviews. This dissertation estimates a low ridesourcing efficiency rate compared to other modes, mix of modes replacement, overall increase in VMT, decrease in parking demand, low wages (i.e. net earnings) for drivers, travel behavior changes for users, as well as relationships between modality style, trip purpose, and stated reasons for mode replacement. These results give us insights into the impacts of ridesourcing on several key aspects of transportation. This, in turn, will help cities and transportation organizations better account for ridesourcing in their planning and engineering processes (e.g. travel demand models) as well as policy decisions.
Thesis:
Thesis (Ph.D.)--University of Cdorado Denver
Bibliography:
Includes bibliographical references.
System Details:
System requirements: Adobe Reader.
Statement of Responsibility:
by Alejandro Henao.

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
999349807 ( OCLC )
ocn999349807

Downloads

This item has the following downloads:


Full Text
IMPACTS OF RIDESOURCING LYFT AND UBER ON TRANSPORTATION
INCLUDING VMT, MODE REPLACEMENT, PARKING, AND TRAVEL BEHAVIOR
by
ALEJANDRO HENAO B.S., University of Colorado Boulder, 2006 M.S., University of Colorado Denver, 2013
A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Doctor of Philosophy Civil Engineering Program
2017


This thesis for the Doctor of Philosophy degree by Alejandro Henao has been approved for the Civil Engineering Program by
Bruce Janson, Chair Wesley E. Marshall, Advisor Jo Ann Silverstein Carolyn McAndrews Debbi Main Kevin Krizek
Date: May 13, 2017


Henao, Alejandro (Ph.D., Civil Engineering Program)
Impacts of Ridesourcing Lyft and Uber on Transportation including VMT, Mode
Replacement, Parking, and Travel Behavior
Thesis directed by Associate Professor Wesley E. Marshall.
ABSTRACT
The transportation sector is currently experiencing a disruption with the introduction and evolution of technology and transportation services such as bikesharing, carsharing, on-demand ridesourcing (e.g. Lyft, Uber), and microtransit (e.g. Bridj, Chariot). As these new layers of technology-based transportation options begin to flourish, it is important to understand how they affect our transportation systems and society. This doctoral dissertation analyzes the impacts of ridesourcing on several areas of transportation including: efficiency in terms of distance Vehicles Miles Traveled (VMT) versus Passenger Miles Traveled (PMT) and travel times, mode replacement, VMT increase, parking, transportation equity, and travel behavior. Realizing the difficulty in obtaining data directly from Lyft and Uber, this research employs an innovative approach by the author becoming an independent contractor to drive for both companies; this allowed the author to gain access to exclusive data and real-time passenger feedback. The datasets include actual travel attributes such as times, distances, and earnings from 416 rides (Lyft, UberX, LyftLine, and UberPool), and travel behavior and socio-demographics from 311 passenger interviews. This dissertation estimates a low ridesourcing efficiency rate compared to other modes, mix of modes replacement, overall increase in VMT, decrease in parking demand, low wages (i.e. net earnings) for drivers, travel behavior changes for users, as well as relationships between
iii


modality style, trip purpose, and stated reasons for mode replacement. These results give us insights into the impacts of ridesourcing on several key aspects of transportation. This, in turn, will help cities and transportation organizations better account for ridesourcing in their planning and engineering processes (e.g. travel demand models) as well as policy decisions.
The form and content of this abstract are approved. I recommend its publication.
Approved: Wesley E. Marshall
IV


DEDICATION
I dedicate this work to my wife Gusty. She has been a constant supporter, contributor, and has carried a lot of weight in our lives. Without her, this project would not have been possible.
To my boys, Tomas and Andres, who are my everyday inspiration.
To my parents, who instilled in me the unmeasurable value of education and have supported me unconditionally in all aspects of my life.
v


ACKNOWLEDGMENTS
I would like to thank Dr. Wesley Marshall for mentoring me and providing extensive support and advice by sharing his time, knowledge, and funding for this doctoral dissertation. I thank Dr. Bruce Janson, Dr. Joann Silverstein, Dr. Carey McAndrews, Dr. Debbi Main, and Dr. Kevin Krizek for contributing to my education, for providing feedback and mentorship, and for serving on my Thesis Committee.
I would like to thank all members of the Active Communities Transportation (ACT) Research Group and the Integrative Graduate Education and Research Traineeship (IGERT) Program for their contributions to my research work.
I also thank the transportation organizations (e.g. the Institute of Transportation Engineers CO/WY Section) and local professionals for their advice and financial support; as well as the Mountain-Plains Consortium, the Eisenhower Fellowship Program, and the National Science Foundation through their IGERT Award and the Bridge to the Doctorate Program for providing funding for this work.
Finally, I am grateful to all the Lyft and Uber passengers. Without their patience, time, and willingness to share, this project would not have been feasible.
vi


TABLE OF CONTENTS
CHAPTER
I. INTRODUCTION..........................................................1
Specific Aims.....................................................4
Study Organization................................................5
II. BACKGROUND............................................................6
III. LITERATURE REVIEW................................................... 10
IV. RESEARCH METHODS.....................................................19
Driving for Lyft/Uber and Driver Dataset.........................20
Driving Strategy and Passenger Survey............................23
Study Area.......................................................25
V. DATA.................................................................27
Driver Dataset...................................................27
Passenger Dataset................................................28
VI. DRIVER PERSPECTIVE: TRAVEL TIMES, DISTANCES, AND EARNINGS...........31
Chapter Related Literature.......................................32
Chapter Data and Analysis........................................34
Travel Distances and Times...................................36
Ridesourcing Efficiency Rate.................................37
Ridesourcing Earnings........................................38
Chapter Results..................................................39
Ridesourcing Efficiency Rate.................................41
Ridesourcing Earnings........................................42
Chapter Conclusions..............................................48
VII. VMT IMPACTS..........................................................53
vii


Chapter Related Literature.............................................56
Chapter Data and Analysis..............................................57
Chapter Results........................................................62
PMT/VMT Efficiency.................................................62
VMT/PMT Ratio......................................................63
VMT before and after...............................................63
Chapter Conclusions....................................................66
VIII. PARKING IMPACTS...........................................................68
Chapter Data and Analysis..............................................69
Chapter Results........................................................69
Parking Demand.....................................................69
Locations, Trip Purpose, and Connectivity to Transit Stations......72
Parking as a stated reason to choose Ridesourcing..................75
Chapter Conclusions....................................................78
IX. TRAVEL BEHAVIOR CHANGES...................................................80
Chapter Literature Review..............................................81
Chapter Data and Analysis..............................................83
Chapter Results........................................................87
Mode Frequency and Travel Behavior Changes.........................87
Relationships between Drive Frequency and Other Variables..........90
Modality Styles....................................................92
Chapter Conclusions....................................................94
X. OVERALL RESULTS...........................................................96
Driver Dataset.........................................................96
Ridesourcing Times and Distances...................................96
viii


Ridesourcing Earnings....................................96
VMT.........................................................97
Parking.....................................................97
Travel Behavior.............................................97
XI. SUMMARY CONCLUSIONS AND FUTURE WORK.............................99
REFERENCES.........................................................103
IX


LIST OF TABLES
Table V-I. Origin Destination (O-D) Matrix......................................28
Table V-II. Demographics of Ridesourcing Passengers...............................30
Table VI-I. Travel Times and Distances Summary Statistics.........................40
Table VI-II. Time and Distance Efficiency.........................................41
Table VI-III. Lyft/Uber Fares and Driver Commission...............................42
Table VI-IV. Passenger Cost, Driver Earnings, and Actual Commission...............43
Table VI-V. Gross Earnings........................................................44
Table VI-VI. Gross Earnings Lyft compared to Uber...............................44
Table VI-VII. Ridesourcing Expenses...............................................46
Table VI-VIII. Net Earnings (Gross Earnings minus Expenses).......................47
Table VI-IX. Net Earnings Lyft compared to Uber.................................48
Table VII-I. PMT, VMT Replaced, and Ridesourcing VMT..............................62
Table VII-II. PMT/VMT, before and after...........................................64
Table VII-III. VMT by Mode Replacement, before and after..........................65
Table VII-IV. Extra VMT per year in the U.S. due to Lyft/Uber.....................67
Table VIII-I. O-D Matrix (Driving Trips Replaced).................................73
Table VIII-II. Connectivity to Transit Stations...................................75
Table IX-I. Bi-modality Style Classification......................................94
x


LIST OF FIGURES
Figure II.I. Lyft and Uber Timeline.................................................7
Figure II.II. LyftLine serving cities...............................................7
Figure LV.I. Lyft and Uber Driver Profiles.........................................21
Figure FV.II. Smartphone Apps......................................................21
Figure IV.III. Driver Data Collection Form.........................................22
Figure IV.IV. Car Sign for Passenger Survey........................................24
Figure V.I. Ridesourcing Data......................................................27
Figure VI.I. Travel Distances and Times of a Lyft/Uber Driver......................35
Figure VI.II. GPS Tracking of a Lyft/Uber Ride.....................................36
Figure VII.I. Taxis in Cali, Colombia (Source: ElPais.com.co)......................54
Figure VLI.II. Taxi Tracks in Cali, Colombia (Source: ElPais.com.co)...............55
Figure VII.III. Mode Replacement (Q5)..............................................59
Figure VILLI. Ridesourcing Replacing Driving Trips (Parking).......................71
Figure VLH.II. Travel Behavior Change (Driving today compared to the past).........72
Figure VIII.ID. Ridesourcing Trip Purpose (All respondents and those Driving less).74
Figure VIII.IV. Main reason for choosing Lyft/Uber for Actual Ride.................76
Figure VIII. V. Driving Frequency and Trip Purpose.................................77
Figure IX.I. Travel Demand Framework to Study Ridesourcing.........................83
Figure IX. II. Mode Frequency......................................................88
Figure LX.ILL Travel Behavior Changes, Driving & Public Transportation.............89
Figure IX.IV. Driving Frequency and Trip Purpose...................................90
Figure IX. V. Driving Frequency and Stated Reasons.................................91
Figure IX. VI. Modality Style Classification.......................................93
xi


LIST OF ABBREVIATIONS
DTT Drive to Transit
ETA Estimated Time of Arrival
HOV High Occupancy Vehicles
Hr Hour
IRB Institutional Review Board
MAX. Maximum
MIN. Minimum
MINS Minutes
MPH Miles per Hour
O-D Origin-Destination
PMT Passenger Miles Traveled
Q# Question Number
sov Single Occupancy Vehicle
St. Dev. Standard Deviation
TDM Transportation Demand Management
TNC Transportation Network Company
U.S. United States of America
VMT Vehicle Miles Traveled
WP With-Passenger
WPMT With-Passenger Miles Traveled
WTT Walk to Transit


CHAPTER I
INTRODUCTION
Evolving transportation services such as bikesharing, carsharing, ridesharing, on-demand ridesourcing (e.g. Lyft, Uber), and microtransit (e.g. Bridj) are becoming increasingly popular all over the world. Many factors including social networks, real-time information, and mobile technology allow passengers and drivers to connect through mobile smartphone applications (i.e. apps). In turn, this has led to the creation and popularization of technology companies offering app-based on-demand transportation platforms. As these new layers of technology-based transportation options begin to flourish, it is important to understand how they compete and interact with more traditional modes. Beyond travel behavior, these tools and evolving transportation services can also significantly impact our transportation systems, society, and the environment; yet, very little data is known and the academic research is minimum to understand and measure the impacts of these services regarding outcomes such as vehicle miles traveled (VMT), mode replacement, parking, equity, and travel behavior.
Providing a more diverse array of travel options should theoretically reduce car dependence and lower parking demand; however, there remain unresolved questions about what cities actually gain (or lose) in terms of sustainability-related outcomes including efficiency, congestion, carbon emissions, and transportation equity issues. Even when replacing single occupancy vehicle (SOV) trips, there are negative effects. For example with VMT, there are additional miles traveled by the ridesourcing driver before passenger pickup or after passenger drop-off- over and above the actual trip the passenger would have driven in the first place (Cramer & Krueger, 2016; Henao & Marshall, in press). There is also
1


a theoretical saturation point where higher ridesourcing supply than demand leaves many drivers circulating without riders, which can cause unnecessary VMT, congestion, environmental issues, and other problems that are not yet documented with these new technology-based modal options.
While there is widespread information online regarding companies such as Uber and Lyft, the academic literature on ridesourcing is extremely limited due to the lack of open data on these services. Obtaining data for independent academic research from Lyft and Uber is extremely difficult (Bialick, 2015a; Levitt, 2016) and even when these companies agree to share data, the data is often not adequate for research purposes (Vaccaro, 2016). These private companies cite customers privacy protection and business competitiveness for their lack of data sharing, but perhaps they do to avoid showing the potential negative impacts in our transportation system. City officials and transit advocates have expressed concerns about the lack of open data and potential problems with ridesourcing such as congestion, competition with public transportation, and equity issues (Flegenheimer & Fitzsimmons, 2015; Grabar, 2016; Rodriguez, 2016)
Without appropriate data, measuring impacts is not possible; and even when such data is available, investigating short-term and long-term impacts of ridesourcing on travel behavior such as the travel modes replaced by ridesourcing and why people shifted from a previous mode remains extremely difficult. There are still limitations with regard to measuring new trips that may not have occurred before (i.e. induced travel), modality resources (e.g. car ownership), and modality style (e.g. car-oriented) of users as well as multimodality (i.e. availability of several modes) and intermodality (i.e. combination of various modes for a single trip or mixed-modes). This combination of problems makes
2


analyzing the impact of these services on the overall transportation system exceedingly difficult.
Due to the complexity of this topic, this dissertation first proposes a comprehensive framework aimed at starting the conversation on the type of data that needs to be collected, the questions that researchers need to be asking, and pointing out issues that might arise with conventional research methods. For example, if we ask someone that does not own a car what they would have done without Lyft/Uber for a specific trip, they might answer transit.
In theory, the ridesourcing trip is classified as a negative environmental impact. However, a more comprehensive research framework might reveal that the decision not to own a car in the first place was made in part due to the availability of Lyft/Uber. Considering such longterm car ownership decisions would now expose the ridesourcing trip as a positive environmental benefit.
Beyond looking at the travel modes replaced by ridesourcing, the framework also includes insights from individuals on the process of why a specific mode was selected over the alternatives. For example, what is the role of travel time, travel cost, parking, and other factors in the decision making process? Such insight would help provide researchers with the ability to investigate the impact of ridesourcing on a region or city in terms of VMT and parking demand. It may also facilitate studies across different geographical areas (e.g. urban vs. suburban, city size, density, etc.) where we could find differing impacts in different contexts. In other words, could ridesourcing have, for example, positive impacts in more suburban areas and negative impacts in more urban areas? Or could the contrary be true? The intent is to provide a framework that will allow such questions to be explored, and then to carry out the research.
3


The overall goal of this dissertation is to start filling the gap in the academic literature and help researchers study the effects of evolving services such as ridesourcing and start measuring these impacts on transportation. This, in turn, will help cities and transportation organizations better account for the impacts of evolving transportation services in their policies, transportation planning, and engineering processes.
Specific Aims
The specific aims and key contributions of this research are that I will build upon the existing literature on evolving transportation services by:
1. Developing a comprehensive research framework to study ridesourcing
2. Collecting unique and interrelated datasets of ridesourcing drivers and passengers
3. Developing a ridesourcing survey for passengers seeking Institutional Review Board (IRB) approval
4. Measuring travel distances, times, earnings, and its efficiencies from the driver perspective
5. Measuring the VMT and parking demand impacts of ridesourcing services
6. Investigating travel behavior changes by assessing what travel modes are replaced by these evolving transportation services; and evaluating the factors associated with why people shifted from their previous travel modes and for what trip purposes.
7. Developing a framework for a mode choice model that would allow for integrating ridesourcing services into regional travel models.
4


Study Organization
This dissertation is organized into eleven chapters. Chapter II provides a background for ridesourcing including a history and overview of Lyft and Uber. Chapter III (Literature Review) overviews the topic of evolving transportation services and covers the limited research in this area. In order to better understand how to do research on ridesourcing services, the first step is to develop a comprehensive research framework. Thus, Chapter IV is devoted to this, and includes research methods, city choice, and data collected for its application in this dissertation. Chapter V presents the data. The first three objectives are addressed in Chapter IV and V. Objective four is addressed in Chapter VI (Driver Perspective: Travel Times, Distances, and Earnings), the fifth objective in Chapters VII (VMT Impacts) and VIII (Parking Impacts), and the sixth in Chapter IX (Travel Behavior Changes). Chapter IX is a summary of results and Chapter XI finalizes this dissertation with overall conclusions, recommendations, and future research. Assisting the reader and for better organization, each of the four paper chapters (Chapters VI through IX) includes its own detail section on literature review, specific data and analysis, chapter results, and chapter conclusions for each detail topic.
5


CHAPTER II
BACKGROUND
While Lyft and Uber in their current form are mostly known for their regular Lyft and UberX services, and carpool options: LyftLine and UberPool, they offer other options and have evolved from a variety of services in their history (Figure II.I). For example, Uber started as a black-car limousine service called UberCab, launched in San Francisco in 2010 (McAlone, 2015), while Lyft co-founders Logan Green and John Zimmer previously cofounded Zimride, a true rideshare platform created to connect drivers and passengers through social networking. Green and Zimmer started Zimride in 2007 and sold it to Enterprise Holding in July 2013 (Lawler, 2014). While Lyft was launched in June 2012 with its original regular Lyft service, Uber did not unveil its regular UberX service until July 2012, a couple of years after it started with UberCab. LyftLine and UberPool services started in 2014 but are only available in certain metropolitan cities (Lyft Blog, 2016; Uber Newsroom, 2014, 2016). For example, Figure II.II shows the cities where LyftLine was in service or about to launch as of April 2016 (including Denver).
As of the summer 2016, Uber was already in 450 cities globally, and completed two billion trips in its life span. One billion rides were completed in six years, while the same number of rides were completed in six months (Somerville, 2016). libers estimated valuation continues to grow and currently is at $62.6 billion, making it the most valuable transportation company in the world; and currently, without owning any vehicle, infrastructure, or having to hire drivers as employees. Lyft operates exclusively in the U.S. and is valued at approximately $5.5. billion dollars (B. Salomon, 2016).
6


Figure ILL Lyft and Uber Timeline
o
CD
co O o o -
C* Q o o
o o
Cities Lyft Line is in:
Atlanta New York City
Austin Philadelphia Launching 4/13
Boston San Diego Launching 4/15
Chicago San Francisco
Denver Launching 4/12 San Jose Launching 4/14
Los Angeles Seattle Launching 4/11
Miami Washington. D.C. lyft
Greater Newark Area Launching 4/12
Figure II.II. LyftLine serving cities
(Source: Lyft Blog, Five Days. Six Cities. A Lyft Line First April 5, 2016)
7


One of the latest news releases shows that Lyft is giving rides at a rate of 17 million U.S. rides per month. It is estimated that Lyft has around 20% of the market share, making Uber the ridesourcing company with the highest volume in the U.S. These numbers show the magnitude of Lyft and Uber and their influence on the way people get around. Uber and Lyft path has not been worry free. They have to constantly deal with different situations such as regulations, protests, and lawsuits from taxi companies, city officials, and drivers claiming employment rights. They also have taken advantages of the terminology in their marketing strategies.
The terminology of new and evolving transportation services can be confusing and sometimes ill defined by the transportation sector. Intentionally or unintentionally, many accredited people and companies use the terminology incorrectly, which can mislead public perception and general use of the services. A recent example is the misused word ridesharing when referring to ridesourcing companies in their original form (Goddin, 2014). The Associated Press Stylebook in January 2015 presented an update on the topic: Ride-hailing services such as Uber or Lyft let people use smartphone apps to book and pay for a private car service or in some cases, a taxi. They may also be called ride-booking services. Do not use ride-sharing (Warzel, 2015). While there seems to be a consensus that these services are not ridesharing, there is still no clearly a defined term. Some of the names include: Transportation Network Companies (TNCs), ride-hailing, ride-booking, ridematching, on-demand-rides, app-based rides. In an attempt to be consistent with previous academic research (Rayle, Dai, Chan, Cervero, & Shaheen, 2016) and to allow for possible future variations of such schemes to be housed under the same header, this study
8


uses the term ridesourcing. The definition of ridesourcing is the sourcing of rides from a for-fare driver pool accessible through an app-based platform.
9


CHAPTER III
LITERATURE REVIEW
Lyft and Uber are disrupting urban transportation systems and competing with more traditional modes (i.e. car, taxi, transit, walk, and bike), but a minimal number of U.S. cities has been able to account the impacts of ridesourcing (DuPuis, Martin, & Rainwater, 2015). The introduction of these services has implications for travel behavior and mode shift, as well as impacts on the overall transportation system.
Other services such as bikesharing and carsharing are continuously evolving and increasing users in cities across the globe (S. Shaheen & Cohen, 2012; S. Shaheen, Guzman, & Zhang, 2010). In addition, while the academic literature on carsharing and bikesharing systems has provided insights about these systems user characteristics (e.g. socio-economic demographics, preferences, etc.) and transportation impacts (e.g. car ownership, car use, VMT, reductions of cars on the network, and mode share), the literature on ridesourcing remains very limited.
While there is abundant information online regarding companies such as Lyft and Uber, the academic literature on ridesourcing is very limited, in part due to their novelty and lack of open data on these services. Due to the ridesourcing history, evolution, and similarity to other services, the few academic studies on this topic compared ridesourcing mostly to the taxi industry and ridesharing services (Anderson, 2014; Cramer & Krueger, 2016; Rayle et al., 2016).
Rayle et al. (2016) did a research study comparing ridesourcing and traditional taxis in San Francisco using an intercept survey in spring 2014. The findings from this study indicated that compared to the overall San Francisco population, ridesourcing users tend to
10


be a lot younger, have higher incomes, have lower car ownership, and frequently travel with companions. This study also shows that compared to taxis, ridesourcing customers experienced shorter waiting times. Participants in this study said that ridesourcing both substitute and complement public transit, walking, and biking; and 8% of survey respondents stated that they would not have traveled (i.e. induced travel effect) if ridesourcing services were not available.
More recently, the Shared-Use Mobility Center investigated the relationship between public transportation and shared modes, including bikesharing, carsharing, and ridesourcing in seven U.S. cities. This report found that the higher the use of shared modes, the more likely people use public transportation, own fewer cars, and spent less on transportation. It also shows that shared modes complement public transportation (Murphy, 2016).
Regarding literature not currently published in academia, the website FiveThirtyEight has published a few articles regarding ridesourcing companies using data acquired via a Freedom of Information Act request. The articles show that in New York, Uber is taking rides away from taxis and generally covers a larger area (Bialik, Flowers, Fischer-Baum, & Mehta, 2015; Fischer-Baum & Bialik, 2015). In another article, FiveThirtyEight argues that for Uber to be worth its $50 billion valuation, it has to complement and attract customers that normally use public transportation. This last article also used data on median income levels by census tract and residential pick up rates showing that lower incomes experienced fewer pickups (Silver & Fischer-Baum, 2015). The article compared general travel cost (using basic assumptions) of public transit, Uber, and the cost to own a car; arguing that Uber in combination with high use (around 65% to 85%) of public transportation can be significantly
11


cheaper than car ownership. Overall, the articles suggested that Uber is affecting mode choice, intermodality, and travel costs (that could in turn affect mode choice).
Since the literature in ridesourcing is extremely limited, is important to review the literature on a similar service that has evolved over the last few years and contains more in-depth studies. This is useful in helping understand ridesourcing and for helping design this newer strand of transportation research.
Carsharing systems provide a fleet of shared vehicles for short-term use where members pay in time increments of minutes or hours. Currently, there are several carsharing models including the following variations: round-trip or one-way (i.e. point-to-point), station-based or free-floating, and peer-to-peer.
Round-trip station-based carsharing is the oldest and most established system, where users need to return the vehicle at the same fixed station it was checked out. Round-trip carsharing started in Europe as early as the 1940s, but more successful programs did not began operating until the mid-1990s (S. Shaheen & Cohen, 2007). While most carsharing research is based on the traditional station-based round-trip carsharing system, the last few years have seen a surge in one-way carsharing research. The first services without any fixed vehicle stations Car2go by Daimler and DriveNow by BMW started in 2009 and 2011, respectively (Firnkom, 2012). As of October of 2014, approximately 4.8 million individuals are members of carsharing programs worldwide with a total fleet of 104,000 vehicles (Shaheen and Cohen, 2014).
There have been a number of studies aiming to evaluate carsharing impacts, but the results are not clear with respect to the effects resulting from changes in the launch of a carsharing system. This is probably due to difficulties with respect to data availability,
12


timelines, confounding effects, as well as research design and methodologies (Fimkom,
2012; Graham-Rowe, Skippon, Gardner, & Abraham, 2011; J Kopp, Gerike, & Axhausen, 2013; Johanna Kopp, Gerike, & Axhausen, 2015; Le Vine, Adamou, & Polak, 2014; Stopher & Greaves, 2007). Numerous carsharing studies focus on determining impacts on transportation, land use, environmental, and social benefits with some mixed results in certain areas and clear evidence on others. As regards to this dissertation, carsharing research on travel behavior can be classified and quantified in the following areas:
Socio-demographics for carsharing users and non-users: Studies suggest that carsharing users do not usually represent the overall population with regard to socioeconomics, demographics, and travel behavior characteristics. Carsharing users tend to be younger, with higher levels of education and income, and live in denser areas with better access to public transportation. Carsharing users also tend to have higher public transit, walking, and biking mode shares and lower car usage compared to the general population (Cervero & Tsai, 2004; J Kopp et al., 2013; Johanna Kopp et al., 2015; Martin, Shaheen, & Li dicker, 2010; Sioui, Morency, & Trepanier, 2012).
Car ownership: Studies revealed that car ownership for carsharing members is lower than the general population and non-members. Empirical evidence has also shown a reduction in private vehicle ownership after joining a carsharing program by getting rid of a vehicle owned or foregoing vehicle purchase (Cervero & Tsai, 2004; Meijkamp, 1998; S. A. Shaheen, Cohen, & Chung, 2009; Steininger, Vogl, & Zettl, 1996). For example, a study on City Carshare in San Francisco indicated that a higher share of members reduced car ownership as compared to a control group of non-members, approximately 29% versus 8%. Two-thirds of members also said they refrain from purchasing a vehicle as compared to 39%
13


of non-members (Cervero & Tsai, 2004). Another study based on a survey in 2010 of members of Communauto, a Montreal carsharing company, concluded that members of the carsharing service have approximately 30% lower car usage compared to the level of those that own a vehicle (Sioui et al., 2012). Another study showed that the average number of vehicles per household dropped from 0.47 to 0.24 (Martin et al., 2010).
Car use and vehicle miles traveled (VMT): A large study across North America on round-trip car share subscribers revealed that while most members drive more with carsharing, the minority that drive less are driving less by a higher order of magnitude, which leads to less driving overall. In this study, VMT declined by 27%, and when including those that decided not purchase a vehicle in the first place, it was a 43% reduction (Martin et al., 2010; S. A. Shaheen et al., 2009). The first year of City Carshare operation in San Francisco suggested an increase in motorized travel for members (Cervero, 2003); however, in the second year of operation, the daily VMT reduced slightly for members and increased for non-members (Cervero & Tsai, 2004).
Reduction of cars on the transportation network: Based on several carsharing reports in the U.S., carsharing helps remove an aggregate of 9 to 23 vehicles from the road (including both shed autos and foregone car purchases) per shared-use vehicle from the transportation network (Lane, 2005; S. A. Shaheen et al., 2009). For example, Cervero and Tsai (2004) estimated that a carsharing fleet of 74 in San Francisco removed approximately 500 vehicles from the streets, equivalent to 6.8 private vehicle per carsharing vehicle. Similarly, a study from Philadelphia found that each PhillyCarShare vehicle replaced an average of 23 private vehicles, 11 vehicles from members giving up a car and 12 vehicles from not acquiring one in the first place (Lane, 2005).
14


Mode share: Studies on station-based carsharing suggest that some of its members change travel behavior towards public transportation and non-motorized modes, while others do the opposite by reducing transit, walking and biking usage; overall, however, most people tend to increase public transit and non-motorized modal use (E. Martin & S. Shaheen, 2011). A study of Ulm, Germany using two different methods reported that after the introduction of a point-to-point carsharing service, members shift modes and reduce the usage of all other modes of transportation including private cars, public transportation, and non-motorized travel (Firnkorn, 2012). Carsharing research on both round-trip and point-to-point carsharing concluded that point-to-point is a substitute for public transport while round-trip carsharing is a complement (Le Vine, Lee-Gosselin, Sivakumar, & Polak, 2014).
Many of the studies on carsharing research rely on sample surveys to gather information on members demographics, current usage of the carsharing service, and prior-to-joining carsharing travel behavior information (Lane, 2005; E. Martin & S. Shaheen, 2011; Martin et al., 2010; E. W. Martin & S. A. Shaheen, 2011). While these studies provide a basic idea on socio-economic demographics and travel behavior patterns at the aggregate level, they are inconclusive on the effects of carsharing because they fail to control for several factors that could affect the results (such as predisposition characteristics of people joining a carsharing system) or by not comparing the study population with a control group. From all the carsharing studies, only a few include a statistical control group in their methodology. Control groups, either on longitudinal or cross-sectional research, allow to correct for some confounding effects that otherwise would be difficult to distinguish from effect results. The best example of the use of control groups is the study over time by Cervero, Golub, and Nee (2007) on City Carshare in San Francisco. After two years of
15


service, VMT for carsharing members decreased, but it decreased even further for nonmembers; so relative to the control group, VMT for members increased. Another example is the research study by Johanna Kopp et al. (2015), where they used a reference group of noncarsharing users using an online and app based travel dairy, MyMobility, to collect individual trips over a 7-day period. This was a relatively well-designed study (with respect to survey instruments, methodology, and clearly stated limitations) of a free-floating carsharing service. The study also implemented a multimodal index by analyzing the distribution of transportation modes of carsharing for users and non-users, and stating future research needs to disentangle the effects of joining a carsharing service on mobility behavior, which this dissertation aims to find.
Although studies that use control groups are considered to have a better statistical methodological research design, there are still some problems to overcome such as confounding biases resulting from carsharing members self-selection and arbitrary choice of non-member sampling that could potentially misrepresent the population. Concerning this dissertation, using latent classes will help understand the modality style of individuals using carsharing in relation to the same classes from the general population. Per the literature review, carsharing members tend to have a more sustainable modality style as compared to the general population, including higher use of non-motorized transportation and lower frequency of private car use. In this case, the fair comparison would be to calculate the difference against non-members that have a multimodal travel behavior.
Another way to compare, track, and measure the impacts of carsharing is using the research design implemented by Fimkorn (2012) using Car2go data. Firnkom used the following two approaches to triangulate toward the impact of carsharing on travel behavior:
16


i) hypothetical travel behavior at present without Car2go; and ii) past mobility travel behavior on top of current behavior with Car2go. Details on the survey methods and methodology from this study are applicable to this dissertation. However, the author states that the two measurement techniques should theoretically have produced the exact same results if they were completely independent. In reality, a persons behavior pre-carsharing could easily be different to what that person would do today without carsharing.
The results from the few ridesourcing studies were similar to carsharing studies suggesting that carsharing users do not usually represent the overall population with regard to socio-economics, demographics, and travel behavior characteristics and users tend to be younger, with higher levels of education and income, and live in denser areas with better access to public transportation. Members also have different mobility resources with fewer cars per households, higher levels of bike ownership and public transportation passes, as well as higher transit, walking, and biking mode shares compared to the general population.
The current carsharing, and ridesourcing literature offers a general idea of the socioeconomic demographics and insights into travel behavior impacts at the aggregate level, but there is no clear understanding at the individual level on the actual motivations why a user chooses a mode over the alternatives. For example from the previous studies, there is no investigation on the role of travel time, travel cost, or convenience (e.g. parking) on the utility and mode choice of travel demand models. There is also no implementation of modality style on the effects of carsharing on travel behavior. The changes cannot clearly be attributed to carsharing or ridesourcing without knowing the members behavior prior to joining a new service (e.g. car-oriented or multimodal) and controlling for the factors that influence travel behavior over time such as individual and household characteristics, location
17


choice, or transportation resources. This dissertation aims to address these problems by implementing a methodology that focuses on a more comprehensive examination of ridesourcing effects on individual travel behavior and overall impacts on the transportation system.
As seen in this overall literature review section, independent research on ridesourcing remains very limited. Each chapter covering specific topics (Chapter VI through Chapter IX) includes a more detail review and related literature to each theme.
18


CHAPTER IV
RESEARCH METHODS
The first step in understanding the impacts of ridesourcing is to develop a framework to guide the research and fill the important gaps in the literature. With Dr. Wesley Marshall, we co-authored the book chapter A Framework for Understanding the Impacts of Transportation recently published in the book Disrupting Mobility: Impacts of Sharing Economy and Innovative Transportation on Cities (Henao & Marshall, 2017). This study lays-out the research framework needed to investigate ridesourcing impacts in transportation, emphasizing the need to employ a combination of travel attributes (e.g. travel times), revealed-behavior data, and stated-response data structures.
Many transportation planners and engineers dream of having ridesourcing data to analyze and make transportation decisions. While it would be nice to have access to this data, we still have not seen any examples of data sharing from these companies for independent academic research. Realizing the difficulty obtaining data directly from Lyft and Uber, I decided to become an independent contractor and drive for both companies; this allowed me to gain access to exclusive data and real-time passenger feedback.
I signed-up to drive for both companies in early 2015, initially doing exploratory analysis to determine how viable this methodology would be for collecting data. After the initial test rides, I decided to continue in this direction by developing the research framework and the passenger survey. I then sought IRB approval and applied for research funding.
There are two interconnected datasets on the data collection: driver dataset and passenger dataset. The first is the exclusive data that Lyft/Uber drivers can obtain by giving rides to passengers. This driver dataset contains information about travel attributes
19


from actual trips including date, time of the day, origin and destination (O-D) locations, travel times, travel distances, passenger cost, and driver earnings. The second dataset is the information gathered by surveying passengers during the actual rides (i.e. passenger dataset). Since I would be surveying passengers, I needed to obtain IRB approval to conduct this research. In the spring of 2016,1 submitted a research proposal to the Colorado Multiple Institutional Review Board (COMIRB), obtaining IRB approval to interview passengers (COMIRB Protocol 16-0773, Exception APP001-3).
Driving for Lyft/Uber and Driver Dataset
I conducted my data collection using a sedan vehicle 2015 Honda Civic and a smartphone iPhone 5s to drive as an independent-contract for both Lyft and Uber (Figure IV.I). The main apps in the smartphone used for this research were Lyft, Uber-driver Partner, GoogleMaps, and My Tracks (Figure IV.II). GoogleMaps and MyTracks helped me to track and record ridesourcing travel data. Passengers completed the online survey using their own smartphone or via a tablet device, Samsung Galaxy Tab A, that I provided.
20


Alejandro
5.0
From Cali Valle del Cauca, Colombia
Favorite music is Salsa
t Soccer fanatic, transportation researcher...let's talk about it!
Figure IV.I. Lyft and Uber Driver Profiles
O /S4
Lyft Uber Partner Google Maps myTracks
o.
Figure IV.II. Smartphone Apps
I used the data collection form presented in Figure IV.III to help guide the travel attributes data collection process for the driver dataset. The ridesourcing driver data includes information for each ride such as date and time of the day, weather, pick-up and drop-off locations, driver earnings, and times and distances broken down by waiting/cruising for a ride, en-route to passenger, waiting for passenger, and actual ride. When I was done with driving for the day, I then recorded the end of shift travel time
21


and distance, as shown in Figure IV.IIL Additionally, I collected information about parking; including cruising to park time and cost to determine parking difficulty at destination. Chapter VI includes a more detail description of each segment for the driving travel times and distances.
For the origin and destination locations, I collected the closest cross-streets, rather than the address, to maintain confidentiality. As mentioned previously, I used Google Maps and myTracks GPS apps to track times, distances, and locations, which allowed me to double-check the data recorded.
From Request to
Waiting/Cruising Pick-up (en-route Waiting for From Pick-up to
for a ride to passenger) Passenger Drop-off (WP ride)
From last Drop-off to End location
Driver Initials: BREAKS Mins: END Time:__________
__________________________DRIVER DATA COLLECTION___________________________________________
Date:________ Time:_______ Odometer:_________ LOG-IN Location:__________________________
_________ Miles:________________ Last ride to LOG-OUT time & dist:____________mins (___mi)
Odometer:________ Location:______________________ Log-out to End time & dist:_mins (__mi)
Figure IV.IIL Driver Data Collection Form
22


Driving Strategy and Passenger Survey
On a typical driving day, I turned on both Lyft and Uber apps and waited until a passenger requested a ride. To be conservative, I generally minimized unnecessary driving; thus, I accepted most of the requests unless there were problems with the app or the pick-up location was more than 15 miles away from the driver location (again, this is to minimize driving without a passenger). Once the ride was accepted, I turned off the driving mode for the other service. For example, if it was a Lyft request, the Uber driver mode was turned-off; or vice versa. Then, I traveled to the pick-up passenger location and waited until the passenger got into the car to travel to the desired destination.
I, as a driver, invited passengers to participate in a short survey about ridesourcing both verbally and with signs in the car (Figure IV.IV). The car sign reads: Hi rider, I am a grad student doing research on transportation. Would you help me by doing a short survey (~6 minutes) about this ride? You can use my tablet or go to this link www.ride-survev.com. Thank you! As the sign indicates, passengers had the option to take the survey on a tablet provided by me, the driver, or use their own device by going to a pre-defined website. In some cases, I conducted a verbal interview with the passenger that covered all the questions included in the survey. I waited until the ride was over to take notes and record the interview. Once the ride ended at the destination location, I turned on the other app and waited for a new passenger request. Once the passenger got out of the car, I tried to find the closest parking space available with the intent to record parking data, and again, to minimize cruising distance without a passenger in the car. Driving for both Lyft and Uber helped minimize the waiting times and cruising distance. For example, there were occasions where new requests came in even before I finished parking. I did all of the data collection by myself
23


to eliminate bias between drivers, to control travel without a passenger (i.e. deadheading minimization), to reduce surveyor errors, and to ensure data quality.
Figure IV.IV. Car Sign for Passenger Survey
The passenger survey included three groups of questions:
Specific Trip Questions (Q1 Q10): The first section asks passengers questions regarding the specific Lyft/Uber ride and includes questions such as trip purpose, travel mode replacement, and reasons to shift from a previous mode.
General Use Questions (Oil 025): The second part of the survey covers broader questions about travel behavior in general such as modality resources (e.g. car ownership, transit pass, etc.), general ridesourcing use, frequency of use for different modes, travel behavior changes, and more general trip purposes and reasons.
24


Demographic Questions (026 037): The third section of the survey includes questions regarding characteristics of the individual and household (i.e. socio-economic demographics).
All survey questions are included in Appendix A. Chapter V, about data, as well as Chapters VII, VIII, and IX include a more detailed description of the survey questions in this dissertation.
Study Area
While Lyft and Uber originated in what they considered an unregulated space, Colorado was the first state in the U.S. to legislatively authorize Lyft and Uber services to operate with a bill signed by Governor John Hickenlooper in June 2014 (Vuong, 2014). This helped make Denver and the surrounding cities an innovative and welcoming location for these evolving transportation services. The Denver metropolitan region comprises a variety of places, covering both urban and suburban areas. For example, it contains very urban places like Union Station in downtown Denver, as well as low-density areas such as those surrounding the Denver International Airport (DIA), located about 24 miles north-east of Union Station. This metropolitan area also includes a college town like Boulder and suburban cities like Westminster or Broomfield in between Denver and Boulder. This diversity of characteristics (e.g. density, race diversity, income levels) makes the Denver region an ideal place to study ridesourcing.
Another positive factor in the research design was the randomness of the passenger destinations. As the driver, I did not know where each ride would end up; thus, I drove all over the study area and visited many of the places previously described. The only location
25


that I had control over is where I turned on the app at the beginning of the shift. Thus, I varied my starting location.
26


CHAPTER V
DATA
Since I signed-up for Lyft and Uber in 2015 including the rides in exploratory analysis I gave around 500 rides, transporting over 650 passengers. This dissertation includes 416 rides for the driver dataset and 311 surveys for the passenger dataset collected over a period of 14 weeks mostly during the fall 2016. The flowchart in Figure V.I shows the datasets description to help guide the two types of interconnected datasets.
RIDESOURCING DATA
DRIVER DATASET PASSENGER DATASET
Travel Attributes: ft Survey Questions:
Travel Times Specific Trip (Q1-Q10)
Travel Distances General Use (Q11-Q25)
Earnings Demographics (Q26-Q37)
Figure V.I. Ridesourcing Data
Driver Dataset
The distribution of the 416 rides for the different services was:
198 regular Lyft rides
164 UberX rides
39 LyftLine rides
15 UberPool rides
For this dissertation, I drove a total of 4,950.7 miles, spent a total of 15,529 minutes
(or 258 hours and 49 minutes) working as a driver, and earned a total of $4,062.08, including
27


tips. More details on the summary statistics for travel times, travel distances, and earnings can be found in Chapter VI (Table VI-I & Table VI-IV).
Passenger Dataset
As stated before, the passenger dataset from 311 surveys include three types of questions. I analyzed responses to specific trip questions and general ridesourcing usage in Chapters VII through IX. To give the reader an idea of the origin and destination (O-D) combinations, I created the O-D matrix shown in Table V-I. Of all O-D combinations, the three most common were from Home to Work, from Home to Going out/Social and from Going out/Social to Home. Originally, there were many more responses for Other Write in but with further analysis, I disaggregated this category and included those with common origin and destination. They are Hotel/Airbnb and Family/Friend.
Table V-I. Origin Destination (O-D) Matrix
DESTINATION Home Wo Hi School Shopping/ Going Out/ Airport Hotel/ Family/ Other ORIGIN Errands Social Airbnb Friend Totals
Home 2 36 16 7 34 18 0 4 12 129
Work 21 8 1 1 1 2 6 0 1 41
School 5 0 0 3 0 0 0 2 0 10
Shopping/Errands 11 1 0 3 1 0 0 0 0 16
Going Out/Social 30 1 0 3 10 0 3 3 1 51
Airport 3 0 0 0 0 0 2 0 0 5
Hotel/Airbnb 0 2 0 0 7 4 0 0 4 17
Family/Friend 10 1 0 0 1 1 3 1 2 19
Other 8 3 0 2 2 1 3 1 3 23
Totals | 90 | 52 | 17 | 19 | 56 | 26 | 17 | 11 | 23 311
Table V-II provides description statistics from all 311 passengers surveyed. Comparing the summary statistics to the Denver population, the sample seems very representative of the population. Previous studies have shown that the ridesourcing
28


population (and carsharing) does not usually replicate the area they represent with higher incomes, low minority representation, and younger users (Murphy, 2016; Rayle et al., 2016). The authors from these research papers suggest that these services mostly serve certain populations but I believe is mostly due to the location of the intercept surveys. My research has the advantage of being random by design since I did not know the passengers destination location. Thus, allowing this study to cover a larger area and include populations that are usually not represented in this type of studies. The sample has a very close split of male-female population. Passengers were mostly younger adults but compared to other studies, I had higher participation from persons of ages 55 to 64, and 65+ years old people. While two thirds of the sample stated being of white race, I obtained representation from different races and ethnicities. In contrast to previous studies, income is better distributed between different ranges, and not very far from the Denver population.
29


Table V-II. Demographics of Ridesourcing Passengers
Denver Rule soiirc ins Population3 Responses (%) (%) Denver Ride sourcing Population3 Responses (%) (%)
Gender Marital Status
Female 145 46.9% 50.0% Single or never married 185 62.7% 41.7%
Male 162 52.4% 50.0% Married or in a family relationship 80 27.1% 39.2%
Prefer not to answer 2 0.6% Separated, divorced, or widow 28 9.5% 19.1%
n 309 Other 2 0.7%
n 295
Residency
Local Resident 254 82.2% - Household sizeb
Visitor 55 17.8% - 1 65 22.3% -
n 309 2 129 44.2% -
3 56 19.2% -
Age 4 30 10.3% -
18-24b 78 25.2% 10.0% 5+ 12 4.1% -
25-34 132 42.7% 21.8% n 292
35-44 56 18.1% 15.4%
45-54 30 9.7% 11.7% Children in household
55-64 7 2.3% 10.5% Yes 47 20.5% 25.1%
65+ 6 1.9% 10.7% No 182 79.5% 74.9%
n 309 n 229
Race/Etchnicity Education
Asian 24 7.8% 3.5% Less than High School 9 3.0% 13.9%
Black/African American 16 5.2% 9.4% Graduated high school or equiv. 49 16.5% 17.7%
Hispanic or Latino 39 12.7% 30.9% Some college, no degree 58 19.5% 18.3%
White 206 66.9% 53.1% Associate or Bachelor's degree 124 41.8% 32.5%
Other 16 5.2% 3.1% Advanced degree (Master's, PhD) 57 19.2% 17.6%
Prefer not to answer 7 2.3% n 297
n 308
Employment Status
Household Income0 Working (Full-time or Part-Time) 246 81.7% 70.9%
$30K or less 34 11.5% 28.3% Volunteer 1 0.3% -
$3 IK $45K 56 18.9% 14.0% Unemployed 15 5.0% 6.3%
$46K $60K 58 19.6% 11.1% Retired 8 2.7% -
$61K $75K 30 10.1% 10.0% N/A 31 10.3% -
$76 $100K 40 13.5% 11.9% n 301
Over $100K 50 16.9% 24.9%
Prefer not to answer 28 9.5% - Student Status
n 296 Student (Full-time or Part-time) 70 23.3% 34.2%
Not currently a student 230 76.7% 65.8%
n 300
2011-2015 ACS 5-Year Estimates, Denver County b Age 1st Range is 15 24 for ACS c Income Range for ACS slighly different
30


CHAPTER VI
DRIVER PERSPECTIVE: TRAVEL TIMES, DISTANCES, AND EARNINGS
This chapter focuses on three very important aspects of ridesourcing from the driver perspective: travel times, distances, and earnings. For this study, I used the driver dataset including 416 rides from Lyft, UberX, LyftLine, and UberPool. When driving for Lyft and Uber, travel times were measured in minutes and travel distances in miles starting with the length from app log-in to ride request/acceptance, from ride request/acceptance to pick-up, waiting for passenger (time only), and from passenger pick-up to drop-off. The length from pick-up to drop-off will be referred as with-a-passenger (WP) ride for the rest of the study. These four measurements were recorded for each new ride, and at the end of the shift, lengths from drop-off to app log-out and/or end destination were measured. This involves the commute at the end of the shift. Note that during the period that data was gathered, Uber and Lyft introduced an option to set a destination filter. This option allows the driver to set a destination filtering the ride requests that go along the same route.
I estimated ridesourcing efficiency rates based on WP rides versus total times and distances. Based on the distance efficiency, I also calculated total VMT per 100 with-passenger miles traveled (WPMT), which helps to determine the additional VMT or deadheading experienced in our transportation system due to ridesourcing. Total ridesourcing travel time and distances also allow me to calculate the gross earnings per hour and per mile. Finally, I estimated ridesourcing driving expenses and net earnings per hour and per mile.
This study starts to fill a gap in the literature by studying the effects of ridesourcing on transportation from the driver perspective. My aim is to help cities and regional transportation organizations better account for the impact of technology and evolving
31


transportation services such as Lyft and Uber in their transportation planning and engineering processes and have a clearer picture of the actual gross earnings, expenses, and net earnings for ridesourcing drivers. In this chapters concluding section, I consider improvements to the current ridesourcing services in terms of increasing efficiency to reduce VMT, due to deadheading and wasted time, and provide higher earnings for ridesourcing drivers.
Chapter Related Literature
While most of the studies mentioned on Chapter III (Murphy, 2016; Rayle et al.,
2016) focus mainly on the ridesourcing passengers, there are only a few articles that focus on the driver side.
Ridesourcing has been mainly compared with taxis. There has been a lot of resistance and controversy with the introduction of ridesourcing since they disrupted the industry, competing and taking away many customers from taxis. Both services are similar in the fact that drivers transport passengers for a fee, but there are many differences including technology innovation, labor market differences, and government regulations. In terms of driving and time efficiency of ridesourcing and taxi services, Cramer and Krueger (2016) compared the capacity utilization rate of UberX drivers against taxi drivers in a few U.S. cities. Using the aggregated data across all drivers available for both cities, the findings show that the percent of work hours with a passenger ranges from 32.0% to 49.5% for taxis, and 46.1% to 54.3% for UberX. The mileage-based capacity utilization measure (i.e. percent of miles driven with a passenger) from the same study was calculated at 39.1% to 40.7% for taxis, and 55.2% to 64.2% for UberX. The main limitation of Cramer and Kruegers study was the exclusion of mileage and times drivers have to travel from the point of log-out to the end location (i.e. commute home), which overestimates their capacity utilization rate.
32


The media has put a lot of attention in the income for Lyft and Uber drivers. A Wall Street Journal article in 2013 stated that a typical Uber driver takes in more than $100,000 a year in gross sales (MacMillan, 2013). After this income estimation was questioned, Uber reduced this income characterization and more recently advertise that its drivers earn up to $35 an hour (same as Lyft advertisement). Based on data from October 2014, a study commissioned by Uber found that UberX drivers were grossing around $17.40 an hour for 20 market cities as a whole (Hall & Krueger, 2015). They also reported taxi drivers and chauffeurs wages of around $12.90 an hour based on the Occupational Employment Statistics survey. The main difference is that libers driver-partners, who are independent contractors, are not reimbursed for driving expenses, in contrast to taxi drivers, who are usually employees. The Uber hourly wage calculated in the Hall & Kruegers study was based in 2014, when rates were higher than in 2015 or 2016, and did not include the time drivers have to travel from the point of log-out to the end location, same as previously described for the article by Cramer and Krueger (2016).
A recent online article published by BuzzFeed News based on leaked internal data from Uber reported that Uber drivers earn $12.70 an hour in Detroit, $14.18 an hour in Houston, and $16.89 an hour in Denver before expenses (O'Donovan & Singer-Vine, 2016). The article also estimates drivers expenses, but I find the assumptions and methodology very poor since it underestimates the depreciation cost by using a $16,000 car value, overestimates the lifetime expectancy of an average automobile to 250,000 miles, and uses a low gas cost of $1.75 per gallon. It is also not clear about the insurance, maintenance, and miscellaneous costs associated with driving. It is important to note again that these calculations also do not include the commute time and distance for drivers (from the point of log-out to the end
33


location). By not including this additional time and expenses, the reported earnings per hour could be severely overestimated.
This study is the first research that independently analyzes data from the driver perspective using both Lyft and Uber trips, including all the additional travel distances, additional times, and actual gross, expenses, and net earnings per hour and per mile incurred by Lyft/Uber drivers.
Chapter Data and Analysis
I used a total of 416 rides 108 rides pre-IRB and 308 with IRB approval for this study. For each ride, the information of interest includes: the service the ride was requested from (Lyft, LyftLine, UberX, or UberPool), travel times, travel distances, and earnings including tips. The data analysis process began by calculating the breakdown of travel times and travel distances for each ride (Figure VTI & Figure VTII):
ti = time a driver has to wait until a new ride request
di = travel distance cruising for a ride (if the driver decides to park and wait until a new request, this distance is zero or close to zero)
t2 = travel time from ride request/acceptance to passenger pick-up (i.e. en-route to passenger) or estimated time of arrival (ETA)
d2 = travel distance from ride request/acceptance to passenger pick-up (i.e. en-route to passenger)
t3 = waiting for passenger time once at pick-up location
U = travel time from passenger pick-up to drop-off, or WP time
d3 = travel distance from passenger pick-up to drop-off, or WPMT
34


From Request to
Waiting/Cruising Pirk un fen rnutP Waiting for From Pick-up to
for a ride to passenger Passenger Drop-off (WP ride)
From last Drop-off to End location
Figure VLI. Travel Distances and Times of a Lyft/Uber Driver


f /' 1

\// Mileage and Times
A I I O Cruising/Waiting for a ride (A-B) O En-Route to passenger (B-C)
Jj
i
,§j\ VI 1 East 18th Avenue > Waiting for Passenger (C)
O With-passenger (WP ride) (C-D)

I |

-r-j-- -TT1
ie c East 6th Avenue ; 8
4"'
East 1st Avenue. OpenStreetMap contributors, CC-BY-SA
Figure VI.II. GPS Tracking of a Lyft/Uber Ride
In addition to the previous travel times and distances, drivers have to travel to their end locations and commute home once they drop-off the last passenger and are finished with the shift. The commute at end is also illustrated in Figure VI.I and includes:
ts = travel time from drop-off to app log-out plus travel time from app logout to driver end location
d4 = travel distance from drop-off to app log-out plus travel distance from app log-out to driver end location
Travel Distances and Times
The ridesourcing driving time and distance per shift are calculated by the following equations:
t shift = [^\fl + t2 + + Tj] +
36


d-shift [^"*(^1 + d2 + d3)| + d4
For this study, the total ridesourcing driving time is:
tT = ^ tshift = ^ h + Yj t2 + Z t3 + Z U + Z h
And the total ridesourcing driving distance is:
dT = ^ dshift = ^ d4 + ^ d2 + ^ d3 + ^ d4
In terms of VMT and WPMT, the total ridesourcing driving distance can be expressed as follows:
VMTt = £d1+£i2 + W'PMJY+£i4
vmtt = WPMTt + Edi+Zd2+Zd4l VMTt = WPMTt + Additional VMT
Ridesourcing Efficiency Rate
To determine the time efficiency rate, I compared the sum of WP times (£ t4) against total times (tr):
Time Ridesourcing Efficiency =
1U
And the sum of WPMT travel distances (£ d3) against total travel distances (dT) for the mileage efficiency rate:
£ d3 WPMTt
Mileage Ridesourcing Efficiency = =
Ctj V IVl 1 f
37


Based on the total VMT equation: VMTt = WPMTt + Additional VMT, the
additional percent of WPMT is:
Additional VMT VMTt
WPMTt WPMTt
Finally, I calculated the total driving miles for every 100 miles transporting
passengers (100 WPMT), as follows:
Total Miles per 100 WPMT =
100 VMTt
WPMTt
Ridesourcing Earnings
I calculated driver gross earnings per hour and per mile using total earnings divided by the corresponding travel time or travel distance. For example, the gross earnings for all 416 rides was calculated by adding all driver earnings and divided by total time and total mileage, as per the following equations:
I also calculated three different scenarios to account for the broad range of expenses drivers might incur. The expense rate and calculations are explained in more detail on the results section. After discounting expenses, I estimated the net earnings per hour for all rides, for Lyft-only rides, for Uber-only rides, including before and after tips.
Gross Earnings
£ Driver Earnings (incl. tip)
Gross Earnings (Vmj/e)
£ Driver Earnings (incl. tip)
38


Chapter Results
Using the median travel times and distances summary statistics (Table VI-I) from the dataset, a representative day for a ridesourcing driver would be as the following description. The Lyft/Uber driver logs-on both apps; he/she tries to minimize the cruising distance (0.2 miles) but has to wait 7.5 minutes (mins) until he/she gets a request. Once the driver accepts the request, he/she spends approximately 5.0 minutes traveling 1.0 miles to the passenger pick-up location. Then, the driver has to wait 1.0 minutes for the passenger to board the car and start the actual ride. The median time and distance of the actual WP ride is 11.5 mins and 3.6 miles, traveling at an average speed of 28.8 miles per hour (based on a total of 6,106 minutes and 2929.9 miles). After the passenger is drop-off, the driver starts the process again waiting for a new ride request but minimizing unnecessary driving. When the driver is done for the day, he/she travels to the desired end location, commuting around 12.0 miles in 20.0 minutes (based on median values of 65 commuting trips or shifts). When the sum of all commuting times and distances are equally distributed to all rides, the median total driving time per ride is 32.8 minutes (average of 37.3 mins) and the median total driving distance per ride is 8.3 miles (average of 11.9 miles).
Following this dataset summary statistics, I divided the chapter results section into two subsections covering ridesourcing efficiency rates (time and distance) and earnings (gross and net earnings after expenses).
39


Table VII-I. Travel Times and Distances Summary Statistics
DRIVER LOG-IN di.ti RIDE d2,t2 PICK-UP t3 PICK-UP ^3, ^4 DROP-OFF
OR NEXT RIDE REQUEST LOCATION PASSENGER PASSENGER
DRIVER ^4,^5 END
LOG-OUT LOCATION
Waiting/Cruising for a ride From Request to Pick-up (en-route to passenger) Waiting for Passenger From Pick-up to Drop-off (WP ride) From last Drop-off to End Location Totals (t j & dT)
Total (It) 4,965.00 2,511.00 531.00 6,106.00 1,416.00 15,529.00
p-S St. Dev. 15.46 3.65 2.10 10.04 12.27* 20.30
Median 7.50 5.00 1.00 11.50 20.00* 32.83
Total (Ed) 635.91 600.56 2,929.94 784.29 4,950.69
a ^ Mean 1.53 1.44 7.04 12.07* 11.90
S3 aj GS .22 S St. Dev. 3.94 1.44 8.60 7.43* 10.37
O w Median 0.20 1.00 3.55 12.00* 8.30
Average mph 14.35 28.79 33.23 19.13
n=416 (Lyft: 198, LyftLine: 39, UberX:164, UberPool: 15) Commute based on 65 shifts
o


Ridesourcing Efficiency Rate
The time efficiency rate of a ridesourcing driver based on the time a passenger is in the car and total time from driver log-in to log-out (not accounting for the commute at the end of the shift) is 41.3%, meaning that I, as a driver, during my shift hours spent more time without a passenger than with one in the car. For example, if in a shift, I was working for five hours, I only spent just over two hours with passengers in the car, due to all the time spent waiting for a ride, going to pick-up the passenger, and waiting for the passengers once I was at the pick-up locations. When accounting for commuting time at end of shift, the time efficiency rate drops to 39.3% of total time (tr) (Table VI-II). Based on distance, the ridesourcing mileage efficiency rate without and with commute at end is 65.4% and 59.2%, respectively. The total ridesourcing driving mileage per every 100 WPMT is 169.0.
In other words, Lyft and Uber drivers travel an additional 69.0 miles in deadheading for every 100 miles they are with passengers.
Table VI-II. Time and Distance Efficiency
WP Ride (Hd3 & Et4) Total minus Commute at End Efficiency:! WP/(Total minus j Commute at End)| Totals (t'p & (Ip) Overall Efficiency (WP/Total) Additional Percent of WPMT VMT per 100-WPMT
Time (minutes) 6,106.0 14,767.0 41.3% 15,529.0 39.3%
Distance (miles) 2,929.9 4,482.9 65.4% | 4,950.7 59.2% 69.0% 169.0
41


Ridesourcing Earnings
The rates that passengers pay for Lyft and Uber fluctuates, but traditionally, they have been lowered over time. The percent that Lyft and Uber pay their drivers has also lowered over time going from paying 80% initially (20% commission to Lyft/Uber) to 75% nowadays (25% commission to Lyft/Uber). Table VI-III presents the Lyft/Uber fares and commission rates applicable to this study. Table VI-IV shows the total amount paid by passengers, driver earnings, and the actual Lyft and Uber commission, before and after tips. Earnings include prime and guarantee bonus per hour but does not include initial sign-up bonuses. All monetary values are in 2016 U.S. dollars.
Table VI-III. Lyft/Uber Fares and Driver Commission
Passenger Cost*
Lyft/Uber Service Fee Base Fare Cost per Minute Fare Cost per Mile Fare Minimum Paid by Passenger (Fee + Fare) To Driver** Lyft/Uber Conunision**
Lyft $2.10 $0.50 $0.12 $1.01 $7.10 80% Fare 100% Service Fee
UberX $1.95 $0.75 $0.13 $1.00 $6.95 + 100% Tips + 20% Fare
* Rates as of Fall 2016 in U.S. dollars. Rates varied and have been lowered over time
** 20% Conunision when first signed-up in 2014. Newer drivers pay a higher conunision (25% or more)
42


Table VI-IV. Passenger Cost, Driver Earnings, and Actual Commission
Passenger Cost To Driver To Lyft/Uber
Total Paid (before tip) Total Cost per WP Mile (before tip) Total Earned (before tips) Tips Total Earned (with tips) Actual Conunision (before tip) Actual Comission (after tip)
Lyfl (n=237) $2,934.58 $1.87 $2,059.25 $276.00 $2,335.25 29.8% 27.3%
Uber (n=179) $2,505.62 $1.84 $1,687.83 $39.00 $1,726.83 32.6% 32.1%
All Trips (n=416) $5,440.20 $1.86 $3,747.08 $315.00 $4,062.08 31.1% 29.4%
* Earnings include prime and guarantee bonus per hour but does not include initial sign-up bonus. ** Earnings in Year 2016 U.S. dollars
Gross Earnings
The dataset shows that if only the time and distance drivers spent with a passenger (WP) is taken into account; Lyft/Uber drivers would be making around $40 per hour or $1.39 per mile. However, there is more to account for within the overall work shift. After including all times and travel distances, gross earnings turn out to be $15.69 per hour or $0.82 per mile (Table VI-V).
43


Table VI-V. Gross Earnings
Gross Earnings based on WP Gross Earnings based in Total minus Commute Gross Earnings based in Totals (t| & dr)
$/hr $39.92 $16.50 $15.69
$/mile $1.39 $0.91 $0.82
n=416. Earnings include tips (Year 2016 U.S. dollars)
Disaggregating by ridesourcing company, I found differences between Uber and Lyft earnings (Table VI-VI), with tips playing an important role in the differences. The small amount of Uber tips was from a few passengers giving tips in cash since Uber does not facilitate tipping on its app.
Table VI-VI. Gross Earnings Lyft compared to Uber
Gross Earnings (before tip) ($/hr) Gross Earnings (with tip) ($/hr) Gross Earnings (before tip) ($/mile) Gross Earnings (with tip) ($/mile)
Lyft (n=237) $14.38 $16.31 $0.77 $0.87
Uber (n=179) $14.60 $14.93 $0.75 $0.76
All Trips (n=416) $14.48 $15.69 $0.76 $0.82
* Earnings based in Totals (tT & dT) ** Earnings in Year 2016 U.S. dollars
44


Expenses
There many variables and rates that go into calculating personal car expenses such as ownership costs (e.g. depreciation, finance charges, license, insurance, registration & taxes) and operating costs (e.g. gas, maintenance, miscellaneous upkeep such as car washes and cleaning, mobile device and data fees, parking and traffic violations, and the risk of crash or injury). The expenses also depend on the value of your car, driving mileage, and whether or not you own a car already and/or have already paid for some of these expenses. To account for the broad range of possibilities, I characterize three different expense scenarios (Table VI-VII.) covering all types of drivers, from occasionally part-time drivers to full-time drivers. In the basic added cost, I assume a range of driving hours of 1-15 hrs/week and around 11,000 miles per year. The next scenario included most of the drivers with 16-49hrs/week and around 33,000 miles per year, and the last scenario is based on the U.S. Federal Standard Mileage Rate.
The first cost scenario assumes that a driver already owns a car and has paid off basic ownership expenditures. Ridesourcing drivers are supposed to upgrade their car insurance to be properly insured with ridesourcing but a few drivers probably do, risking that an insurance company would not pay a claim if a person was driving for Lyft/Uber. For this first scenario, I assumed most ownership cost such as insurance as a sunk cost that drivers pay regardless of whether a person drives for a ridesourcing company or not; in other words, it is not considered an additional expense. This scenario also includes conservative values for depreciation, maintenance, and other miscellaneous expenses. The cost expense for this scenario is $0.28 per mile.
45


The next scenario represents the majority of ridesourcing drivers (51% of drivers) based on Uber data published by Hall and Krueger (2015). Since drivers in this scenario experience higher timing and mileage, I included costs associated with owning a car and increased the other values according to the mileage per year. I used assumptions based on AAA rates (AAA, 2015) and other sources but still trend toward the conservative end of the expense spectrum. In this scenario, expenses equal to $0.40 per mile.
In the third scenario, I used the 2016 U.S. standard mileage rate determined by the federal government of 54.0 cents per mile. The average mileage rate based on the previous three scenarios is calculated at $0.41 per mile. The corresponded cost per hour is based on the average of 19.1 mph from Table VI-I.
Table VI-VII. Ridesourcing Expenses
Basic Added Cost Most Drivers U.S. Federal Average
Item l-15hr/week, 16-49hr/week, Standard Mileage Mileage
-Ilk miles/year ~33K miles/year Rate (2016) Rate
Ownership Depreciation $1,320.00 $3,960.00
Finance Charge - $500.00
License, Registration & Tax - $350.00
Insurance - $1,500.00
Operating Gas $1,015.38 $3,046.15
Maintenance $589.60 $1,768.80
Miscellaneous $150.00 $2,000.00
Total $3,074.98 $13,124.95
$/mile $0.28 $0.40 0.54* $0.41
$/hr $5.34 $7.60 $10.31 $7.75
Assumptions: Car value: $18,000; Lifetime mileage: 150,000; Work: 50 weeks/year; Gas price: $2.40/galon (Average in 2015); Gas efficiency: 26 MPG; Maintenance: 5.36 cents/mile; Miscellaneous include car wash & cleaning, mobile device & data fees, parking & traffic violations, risk of crash or injury 2016 U.S. Federal Standard Mileage Rate
46


Net Earnings
Ridesourcing drivers are probably excited to think they are making $40 per hour, or even $16/hr but would be disappointed to learn that, after accounting for expenses, the average hourly rate, including tips, is $7.94 (not even minimum wage in Colorado) as shown in Table VI-VIII. This net earning wage could be even lower because of the higher commission rate of 80% versus 75% of newer driver and relatively conservative expense estimates.
The net earning rate per mileage is between $0.28 and $0.54, with an average of $0.41; meaning drivers gross earnings are cut in half after expenses. With these numbers, if a driver work full-time (40 hours a week, 50 weeks a year) driving over 40,000 miles a year, the annual net income would be around $16,000. These net numbers are all pre-tax earnings.
Table VI-VIII. Net Earnings (Gross Earnings minus Expenses)
Net Earnings
Range (Low to High) Average
$/hr $5.38 $10.36 $7.94
$/mile $0.28 $0.54 $0.41
n=416. Earnings include tips (Year 2016 U.S. dollars)
When disaggregating by ridesourcing company and including tips, the Uber net
earning rate is $7.18 per hour, and Lyft is $8.56/hr (Table VI-IX). Tips makes a significant difference on living wages for Lyft drivers with around $1.93/hr and accounting for a 29.1% increase of net earnings, while is only $0.33/hr (4.9%) for Uber.
47


Table VI-IX. Net Earnings Lyft compared to Uber
Net Earnings (before tip) ($/hr) Net Earnings (with tip) ($/hr) Tip Percent
Lyft (n=237) $6.63 $8.56 29.1%
Uber (n=179) $6.85 $7.18 4.9%
All Trips (n=416) $6.73 $7.94 18.1%
* Earnings based in Totals (tx & dT)
** Earnings in Year 2016 U.S. dollars
Chapter Conclusions
The time efficiency rate without taking into account commuting at the end of the shift is 41.3%. This time efficiency rate is lower than the capacity utilization rate of 46-54% in a previous study (Cramer & Krueger, 2016). Accounting for the commute at the end, the overall time efficiency rate drops to 39.3%, meaning that drivers spent more time without a passenger than with one in their car. The main implication for this result is the reduction on earnings per time ($/hour) since ridesourcing drivers have to spend time waiting for a passenger request, traveling to a pick-up destination, waiting for the passenger once at the pick-up location, and commuting time at the end of the shift.
The efficiency rate in terms of WPMT versus total mileage without including commute distance is 65.4%. The mileage efficiency rate for this study is higher than the 61.0% utilization rate calculated by Cramer and Krueger (2016). I attribute this difference to the research design; which minimized the cruising for a ride request, did not accepting rides when the distance to pick-up a passenger was too long, and used conservative commute
48


distances at end of shifts. When including all distances, the mileage efficiency rate drops to 59.2%, but I believe the real mileage efficiency rate is even lower. Even with this conservative calculation, drivers have to travel 69 extra miles in deadheading for every 100 miles originally from WPMT.
There has been a lot of uncertainty regarding how much money a Lyft/Uber driver makes. What is widely known is the difference between what passenger pay and what Lyft/Uber drivers are paid. The Lyft/Uber fare per mile is around $1, but when we take into account all fees and divided by the WP time, passengers pay around $1.86 per mile (Table VI-3 & Table VI-4). When I became a Lyft and Uber driver, I signed-up with a commission rate of 80-20 (80% of fare for driver and 20% for Lyft/Uber), which is used for this study (newer drivers get even lower rates at 75% of fares). When the booking fee is taking into account, the commission rate before tips for all rides is 31.1%. When tips are taking into account and separated by company, the Uber commission is higher (32.1%) than the Lyft commission (27.3%), suggesting that drivers earn better driving for Lyft.
This is the first study to incorporate commute times and distances into earning calculations, but even without accounting for the commute time of the driver to or from their home, the gross earnings drops to less than $16.50/hour. This is a conservative high number since the commission rate received was at the high-end, I was driving for both Lyft and Uber minimizing waiting for a ride times, and I minimized unnecessary driving whenever possible. As a comparison, a recent Buzzfeed article reported that an Uber driver in Denver makes $16.89/hour (ODonovan & Singer-Vine, 2016), but they overestimated this hourly earning with some of the assumptions used in calculating driver expenses.
49


Using all the data from the 416 rides and including all times and distances, the gross earnings for this study equals to $15.69/hour, which might seems like a good hourly rate, but many drivers do not realize the expenses incurred by driving. The expenses varied from our conservative calculations using very basic added cost of $0.28 per mile to the standard 2016 mileage rate of $0.54 per mile, so in reality ridesourcing drivers make between $5.38 and $10.36 per hour, with an average of $7.94/hr before taxes.
Uber net earnings before tips ($6.85/hr) is slightly higher that Lyft earnings ($6.63/hr) but completely change when tips are taken into account. Net earnings with tips included are $8.56/hr for Lyft versus $7.18/hr for Uber. Lyft tips in net earnings equals to a 29.1% increase and plays a critical component in the ridesourcing driving economy. Uber could easily add a tipping option in their app to allow passengers add a tip in their credit card bill if they wish. This choice would help increase drivers earnings, but Uber has thus far refused to implement this option.
Uber and Lyft depend on the driver-partners labor market. They incentivize new drivers with bonuses and referrals, but their retention rate is not very good. According to Hall and Krueger (2015), 89% of Uber driver-partners stay active after one month, 70% after six months, and around 50% after a year. One of the reasons for this may be the realization of driving expenses and costs incurred by driving. For example, a taxi driver who makes $12/hour might think that they can make a lot more driving for Uber ($40/hr or $16/hr). However, they may soon realize that, after accounting for expenses, it is not nearly as profitable as expected and are actually not even making minimum wage at about less than $8/hour.
50


Based on the results from this study, I have several recommendations to create more efficiency by reducing the amount of time and distances ridesourcing drivers have to travel and earn wages that are more decent. Cities authorizing ridesourcing services and companies such as Lyft and Uber should:
Suggest drivers to minimize the amount of miles they drive without a passenger
Balance the driver network better by directing certain drivers to their closest prime rate zones instead of generalizing without specific guidance
Balance the supply of drivers better, especially when passenger demand is not high. This would minimize VMT from drivers circulating around.
Allow drivers to create ride zones so they do not end up far away from their desired location. Also, expand the destination filter option so rides can be matched along certain routes or destinations and not just at the end of the shift (during the study period, Lyft and Uber started an option for drivers to put a destination filter but the option has not been very effective). Lyft and Uber have stated that they want to reduce the inefficiency of empty seats as one of their desired goals so ridesourcing could function like a carpooling app where all drivers set their destination and find passengers along the way.
Not match drivers when the passengers pick-up location is far from the driver location or compensate drivers for these scenarios (I, as a driver, have seen requests from locations more than 30 minutes or 20 miles away).
Concerning earnings, Lyft and Uber could always pay their independent-contractors better by paying drivers on the service fee (which goes 100% to Lyft/Uber), increasing passenger fees, increasing the driver commission fee,
51


providing better incentives, or covering some of the expenses. Thus far, these companies seem to be moving in the other direction by increasing the Lyft/Uber cut from 20% to 25% or higher, lowering passenger rates (mileage and time), and increasing the service fee (which is not shared with the drivers). Uber also has not shown any desire to allow an option to tip in their app, which is the number one request from drivers.
The main limitation to this study is the trip sample size and diversification of drivers. Drivers might have different work strategies such as searching for prime areas, have a desired location in mind, cruising unlimitedly until they get a ride request, or limiting driving without a passenger as much as possible by parking right after a passenger is dropped off. I minimized the distance traveled without a passenger for the results to be conservative. The study is also limited to the Denver metropolitan area so the Lyft/Uber costs and earnings are based on this area.
This is the first independent study to use Lyft and Uber data exclusively to drivers. The results provide insight into the impacts of ridesourcing into travel times, travel distances, and the labor economy of Lyft/Uber independent contractors. This research starts to fill a gap in the academic literature by identifying, measuring, and disentangling the impacts of ridesourcing on very important aspects of transportation. I hope this study helps cities and regional organizations better account for the impacts of ridesourcing on travel time and mileage efficiency, as well as inform the ridesourcing labor market on the complicated issues of earnings and expenses.
52


CHAPTER VII
VMT IMPACTS
Most of my life I have lived in two cities: Cali in Colombia and Denver in the U.S. These cities differ quite dramatically in their economies, demographics, employment, culture, etc. Regarding transportation, they are also very different in terms of land use, transportation services offered, mode share, car ownership, work force, etc. For example, the mode share of private vehicles in Cali is 10% (Cali Como Vamos, 2015) versus around 79% in Denver (U.S. Census Bureau, 2015)
Growing up in Colombia, I lived a completely different transportation experience than my current one. My parents owned one car to be shared by the five members of my family. My travel behavior was truly multimodal; I would take public transportation, carpool, walk, or bike. Occasionally, I would take a taxi to get around the city. Thinking back, one of the things that influenced me the most was the large amount of taxis and their effect in congestion. Still nowadays, Cali experiences many impacts from taxis circulating around (Figure VII.I) and getting in line outside the airport, hospitals, malls, bus terminals, and other public places (Figure VII.II). The taxi impact experienced is very clear since in Cali taxis are yellow and represent 7% of total mode share (Cali Como Vamos, 2015). This suggests a 0.7 (around 7 to 10) relationship when taxi mode share is compared to private vehicles mode share. The estimate of mode share for taxis in Denver (combined with motorcycle or other non-traditional means) is only 1% (U.S. Census Bureau, 2015), representing a 0.01 (around 1 to 79) ratio of taxi versus private vehicles.
53


Figure VII.I. Taxis in Cali, Colombia (Source: ElPais.com.co)


Figure VII.II. Taxi Tracks in Cali, Colombia (Source: ElPais.com.co)
55


This experience is relevant to ridesourcing since very little is known about the contribution that Lyft and Uber provides to the efficiency and impacts of our transportation systems. When people use private vehicles to operate for Lyft and Uber, we might not realize the impacts in city streets since most only carry a barely visible logo sticker.
Cities, regions, and transportation organizations usually set up goals to reduce congestion, environmental impacts, and equity issues. A general term used for housing strategies to aim for more efficient use of transportation resources is Transportation Demand Management (TDM). Some of these goals are in terms of measuring and tracking mode share, passenger miles traveled (PMT), and VMT.
In an effort to contribute to the conversation, this study chapter aims to analyze the mode share replacement occurring with ridesourcing, measure the efficiency ratio of PMT/VMT and VMT/PMT, compare VMT before and after ridesourcing, and estimate the extra VMT generated in the U.S. from Lyft and Uber.
Chapter Related Literature
Transportation organizations across the globe are trying to solve transportation problems by setting strategic goals to reduce SOV or increase the mode share of sustainable modes of transportation including transit, walking, and biking. A few reports and studies have shown that cities have successfully met some of these goals through a variety of strategies, including TDM efforts such as congestion fees, tollways, high occupancy vehicle (HOV) lanes to carpool/increase vehicle occupancy, parking management, transit passes as well as infrastructure investments and policy changes (Bialick, 2015b; Henao et al., 2015; Kaffashi et al., 2016; Steele, 2010).
56


The problem with ridesourcing is that when organizations are trying to set up goals in regards to mode shift and prioritizing certain modes, they do not know the real impacts and efficiency of services like Uber and Lyft. Shall organizations support ridesourcing and encourage its services to a higher use? What modes are they replacing? What is the PMT/VMT or PMT/VMT ratios compared to other modes? How would the transportation system benefit if ridesourcing was replacing modes that are more efficient? For example, we know that it will never be better than biking, walking, or transit since VMT for theses modes is zero or close to zero, but how does it compare to the SOV PMT/VMT ratio of 1.0, or taxis being around 0.40 (Cramer & Krueger, 2016)?
The few studies that look into mode share changes and VMT impacts analyze the data at the aggregate level and do not make a distinction about the magnitude and directional shifts occurring within all modes. This study chapter aims to start filling this gap in the literature by looking in more detail the mode replacement, as well as PMT and VMT changes, and find out the place where ridesourcing stands in terms of efficiency compared to other modes of transportation.
Chapter Data and Analysis
For this research, I used the information containing both the information collected by driving and the corresponding passenger survey information, for a total of 311 passenger surveys during 308 rides. The information gathered by driving is the same as the data collected in chapter VI, except the focus on this chapter is on distance and does not include times nor earnings. I also include information on the number of passenger for each ride. The question of interest from the passenger survey is Q5: For this trip, how would you have
57


traveled if Lyft/Uber wasnt an option? (Figure VII.III). The survey response options to the multiple choice question were:
Wouldnt have traveled
Drive Alone
Carpool (drive)
Carpool (ride)
Public transportation
Bike or Walk
Taxi
Other
After reviewing the other responses, I created three new categories; including two categories for get a ride and car rental. Seventeen passengers responded either Lyft during the Uber ride or Uber during the Lyft ride. These passengers probably did not read the question carefully, or they use Lyft/Uber as their main mode of transportation and did not think of other replacement mode. For these passengers, I created the other ridesourcing category.
58


Q5. For this trip, how would you have traveled if Lyft/Uber wasn't an option?
Carpool (drive)
o no/ Other, 1.6%
Car rental, '10/ Get a ride, 4.;
Other ridesourcing
5.5%
Carpool (ride) 6.1%
Wouldn't have traveled, 12.2%
Figure VII.III. Mode Replacement (Q5)
If the passenger response to question Q5 was carpool, the survey was designed to ask the number of people that the passenger would have carpooled with, with the intent to make a fair comparison (Q6). For this study, I also included the question on whether or not the passenger was using Lyft/Uber for the entire length of the trip (origin to final destination), or he/she was making a connection to another mode of transportation (Q9), and which mode of transportation (Q10). Finally, I included the survey question about car ownership/access
(Q19).
In summary, the information of interest for each ride includes:
Date of ride
Time at request
The service the ride was requested from: Lyft, LyftLine, UberX, or UberPool
59


Travel distances
Number of passengers
Trip Mode replaced (Q5)
Number of people carpooling if passenger would have carpooled (Q6)
Connection with another mode of transportation (Q9 & Q10)
Own or have access to a personal car (Q19).
For the twenty-nine passengers that would have carpooled, the number of people that would have carpooled together is between two and four people with an average of 2.59 people per trip (Q6). Regarding connection with another mode of transportation, 94.50% of the passengers stated that they were using Lyft/Uber for the entire trip, and only 5.50% were using another mode of transportation in connection with Lyft/Uber (Q9). Moreover, 187 people out of 291, or 64.3%, responded to question Q19 by stating that they own or have access to a personal car.
Based on the data previously described and the VMT for each mode replaced, I proceeded to calculate the Replaced VMT (or VMTbefore) and passenger miles traveled (PMT) based on travel behavior prior to Lyft and Uber were in place, as follows:
VMT Replaced for wouldnt have traveled is 0
VMT Replaced for bike or walk is 0
VMT Replaced for car rental is the same as WPMT
VMT Replaced for carpool (drive) is the same as WPMT
VMT Replaced for carpool (ride) is given by the following formula:
VMT = WPMT X
# Passengers for Ride
<# People what would have Carpooled
WPMT for most rides
), which is the same as
60


VMT Replaced for driving is the same as WPMT
VMT Replaced for get a ride is equal to two times (2x) WPMT. This is the case when someone else (e.g. parent or spouse) would have driven the passenger from origin to destination (O-D) and go back to origin incurring in a round-trip doubling the miles of the original O-D trip.
VMT Replaced for other ridesourcing is the same as ridesourcing VMT
VMT Replaced for public transportation is zero for walk to transit (WTT) and 3.4 miles for drive to transit (DTT). The selection of WTT and DTT rides were based on ride distance, answer to connection mode (Q9 & Q10), answer to car access (Q19), percentage of WTT and DTT based on data from a previous study in the Denver area (Marshall & Henao, 2015), and DTT distance based on another paper in the study area (Truong & Marshall, 2014).
VMT Replaced for taxi is equal to 2.5 times (2.5x) WPMT based on the taxi distance efficiency of around 40% (Cramer & Krueger, 2016). I used the same ridesourcing VMT for trips to the airport.
If the ride included a connection, the previous distance replaced is based on total VMT & PMT. For example, if a passenger was dropped-off at a transit station to ride a train to the airport, and the mode replaced was get a ride, the VMT Replaced is equal to two times (2x) the total distance (WPMT plus the train distance) because the person taking the passenger would have travel all the way to the airport and back.
The Ridesourcing VMT (or VMTafter) was calculated using the same methodology as Chapter VI, where all distances with and without a passenger are taken into account.
61


Then, I calculated PMT/VMT and VMT/PMT ratios for before (VMT Replaced) and after (Ridesourcing VMT) to understand the efficiency of PMT/VMT, and how much VMT is put into the transportation system per PMT before and after ridesourcing. Finally, in order to understand the additional VMT put into the system because of ridesourcing, I calculated the ratio of VMTafter versus VMTbefore for every mode replaced and overall.
Chapter Results
The total Ridesourcing VMT in this analysis was 3,617.7 miles while VMT Replaced was 1,959.6 miles, and PMT was 2,200.0 miles. The average passenger travels a mean distance of 7.14 miles (median distance of 3.50 miles) ranging from 0.5 miles to 49.10 miles. See summary statistics in Table VII-I. for more details.
Table VII-I. PMT, VMT Replaced, and Ridesourcing VMT
PMT VMT Replaced or VMTbefore Ridesourcing VMT or VMTafter
Total (Ed) 2200.03 1959.58 3617.68
Mean 7.14 6.30 11.75
St. Dev. 8.83 10.97 10.24
Median 3.50 1.82 7.56
Min. 0.50 0.00 2.54
Max. 49.10 57.00 54.59
Rides=308 (Lyft: 128, LyftLine: 36, UberX:131, UberPool: 13)
PMT/VMT Efficiency
The before travel behavior based on the replaced mode was 112.3% efficient in terms of how much PMT per VMT would have happened if Lyft or Uber were not available, meaning that
62


all the modes replaced were transporting passengers at a rate of 112.3 miles for every 100 VMT. With the introduction of ridesourcing, the PMT/VMT efficiency dropped to 60.8%, meaning that the miles passengers travel is less than the vehicles miles at a rate of only 60.8 PMT for every 100 VMT from Lyft and Uber. This equates to a -51.5% net change or 45.8% percent reduction (Table VII-II.).
VMT/PMT Ratio
Using the same numbers but inverting the numerator and dominator, I calculated the before and after VMT/PMT ratios. If Lyft/Uber were not available, the before travel behavior scenario would have been 0.89 VMT per every 1.0 PMT (Table VII-III), meaning that all the modes replaced were transporting more passengers miles than vehicles miles. With the introduction of ridesourcing, the VMTafter/PMT ratio went up to a value of 1.64 meaning that now with Lyft and Uber the VMT is 1.64 miles for every 1.0 PMT (Table VII-III). This value is very similar to the calculation of 1.69 from chapter VI.
VMT before and after
The total (Ed) and median distances of PMT, VMT Replaced, and Ridesourcing VMT for each mode replaced and total are presented in Table VII-III. The last column of this table shows the VMTafter / VMTbefore ratio for every mode and total, with an overall ratio of 184.6%, meaning an overall increase of 84.6% in VMT.
63


Table VII-II. PMT/VMT, before and after
PMT VMT Replaced Ridesoureing VMT or VM Tbefo ^ or VM TAFIER Efficiency Replaced Ridesoureing Efficiency Net Change Percent Change
Total (2d) PMT PMT Ridesoureing Efficiency Net Change
VMTBEFoRe VMT after minus Replaced Efficiency Efficiency Replaced
2,200.03 1,959.58 3,617.68 112.3% 60.8% -51.5% -45.8%
On


Table VII-III. VMT by Mode Replacement, before and after
Mode Replaced n PMT Total (Ed) Median VMT Replaced or vmtbefore Total (Ed) Median Ride sourcing VMT or VMTafier Total (Ed) Median vmtbefore PMT VM Tapper PMT vmtafter vmtbefore
Public transportation 69 419.6 3.50 27.2 0.00 768.9 7.54 0.065 1.832 2826.7%
Drive alone 59 661.3 5.17 661.2 5.17 935.5 10.97 1.000 1.415 141.5%
Wouldn't have traveled 38 194.0 3.67 0.0 0.00 370.2 8.00 0.000 1.908 GO
Bike or Walk 37 74.3 1.65 0.0 0.00 195.9 4.95 0.000 2.638 GO
Taxi 30 364.2 5.77 639.5 14.41 568.3 10.74 1.756 1.560 88.9%
Carpool (ride) 19 132.1 3.87 82.2 1.82 227.7 7.64 0.622 1.724 277.1%
Other ridesourcing 17 52.8 3.00 143.3 7.58 143.3 7.58 2.713 2.713 100.0%
Get a ride 14 132.6 5.67 265.3 11.33 140.5 9.75 2.001 1.060 53.0%
Car rental 13 54.6 3.71 54.6 3.50 119.7 6.52 1.000 2.191 219.1%
Carpool (drive) 10 77.1 2.74 77.1 2.74 93.6 5.51 1.000 1.215 121.5%
Other 5 37.5 2.55 9.2 2.28 54.1 6.09 0.244 1.441 589.8%
Total 311 2200.0 3.50 1959.6 1.82 3617.7 7.56 0.891 1.644 184.6%
Legend:
Worst VMT Better VMT Overall


Chapter Conclusions
This study chapter shows that, overall, ridesourcing VMT is approximately 184.6% of what would have been without Lyft/Uber, which has significant implications for our cities in terms of congestion and environmental concerns. Ridesourcing provides more mobility -12.2% of passengers stated that they wouldnt have traveled but affects the efficiency of transporting passengers versus vehicles going from a PMT/VMT efficiency of 112.3% to 60.8% (VMT/PMT ratio of 0.89 to 1.64). The worst VMT changes come from passengers that would have used public transportation or active transportation. The problem is that for these modes, the transportation system was previously experiencing zero or close-to-zero VMT, but now with ridesourcing we end up with some of the worst VMTafter/PMT ratios. The main explanatory reason is that these are short distance trips and ridesourcing is less efficient with these trips. For example as shown in Table VII-III the median distance for bike or walk is 1.65 miles and the ridesourcing VMT median is 4.95, which equates to 2.6 times the PMT. The transportation system experiences better VMT efficiency when the replacing mode is taxi or get a ride, since the efficiency rate of ridesourcing is better than taxis or people chauffeuring family or friends. In addition, one of the things to notice on the data is that many of the Lyft/Uber trips that were replacing get a ride were actually connection trips to transit so the VMTafter is minimal compared to the total VMTbefore that a person would have done. For example, if a person rides a Lyft/Uber for 4 miles to a transit station and then, rides the train for 20 miles; the replaced VMT is 48 miles (48 = 2 X [4 + 24]).
In an attempt to have a more general idea of the impacts of Lyft and Uber, I look at some published data on the rides that Lyft and Uber have given so far. This summer, Uber
66


reported that its second billion rides were completed during six months (the first billion was reached in six years) (Somerville, 2016). While Uber operates globally, Lyft is only in the U.S., so I looked at numbers exclusively for the U.S. According to several news articles and based on a statement from Lyft co-founder and president John Zimmer, Lyft does around 17 million rides per month in the U.S., and compared to Uber, their market share of total is around 20% (Buhr, 2016; Kokalitcheva, 2016). Using these numbers, Lyft and Uber gives around 1 billion rides per year in the U.S. If the results for this dissertation held true for the entire country, the impacts of ridesourcing on VMT would be around 5.5 billion extra miles per year in the U.S.; since the VMTbefore (or replaced) would have been 6.4 billion per year, and VMTafter (or Ridesourcing) would be close to 12 billion miles per year from total Lyft/Uber VMT. See Table VII-IV for calculations and more details.
Lyft and Uber have changed travel behavior for many people and they are becoming more popular, but cities and transportation professionals that have expressed concerns over the potential negative implications of ridesourcing have valid points. There should be better policies and regulations in place for ridesourcing companies to operate in a way that would allow our transportation system to function in a more efficient way, but it seems that the opposite is occurring as shown in this chapter study.
Table VII-IV. Extra VMT per year in the U.S. due to Lyft/Uber
Lyft and Uber rides per year in the U. S. 1,000,000,000.00
tT mean= (Ld)/ride (Table IV. 1) 11.90
VMTafter = Rides per year 11.90 11,900,707,268.24
VMTafter/VMTBEFORe (Table V.3) 1.85
VMTbefore = VMTafter / 1.85 6,446,228,741.23
VMTextra = VMTafter VMTbefore 5,454,478,527.02
67


CHAPTER VIII
PARKING IMPACTS
This parking study is one component of a broader research strand that analyzes the impacts of ridesourcing on transportation. As seen in previous chapters, there are negative impacts as it relates to VMT since a ridesourcing driver travels additional miles well above the actual passenger O-D trip. This becomes worse if the passenger is shifting from a mode other than SOV, such as carpooling, transit, biking, or walking. Ridesourcing could also have positive impacts such as parking as more people can access destinations without requiring an accompanying parking space. In turn, this could facilitate reduced parking supply and help alleviate congestion problems that might come from vehicles cruising around searching for parking. On the other hand, parking difficulty and cost is an important TDM strategy that transportation professionals use to deter driving, which might influence peoples decisions to use Lyft/Uber when parking supply is limited and/or expensive. In terms of travel behavior, ridesourcing could also serve as a transition in travel behavior change, car ownership, and long-term modality styles as people shift from an auto-oriented lifestyle towards becoming more of a multimodal traveler. However, this has not yet been adequately studied. Therefore, this research aims to evaluate the impacts of ridesourcing on parking by looking at the trips that otherwise would have needed parking (e.g. if passengers would have driven their own car, renting a vehicle, or carpooling) and investigate specific origins and destinations (e.g. airport, sporting events, special events, bars/night-life, park and ride locations). This study also seek to shed light on understanding parking in terms of issues such as supply and price as a reason of someone using Lyft/Uber instead of driving.
68


Chapter Data and Analysis
For this study, I used the dataset of 311 samples containing both the information collected by driving and the corresponding passenger surveys. The information of interest for this section is on the origins and destinations, trip purposes, and parking-related questions including driving mode replacement and parking as a stated reason to use ridesourcing instead of another mode. I analyzed the dataset for both the specific trip and for general travel behavior questions.
Chapter Results
I divided the results section in three subsections to explore:
Parking demand by exploring the trips where ridesourcing replaced driving
Locations origin & destination (O-D Matrix) with trip purpose, and connection with other modes (e.g. park and rides)
Parking difficulty as a reason to choose ridesourcing over other modes.
Each subsection includes results from the specific Lyft/Uber ride as well as answers from the general travel behavior survey questions.
Parking Demand
Specific Trip
Since parking, theoretically, is only needed for driving trips; I decided to look in more
detail at the mode replacement distribution and pay closer attention to the trips that would have involved driving such as drive alone, carpool (drive), car rental, and carsharing to start understanding potential changes in parking demand.
69


In terms of replacing driving trips with ridesourcing, we should exercise some caution since in some cases, the trip replaced might have been only a part of the trip with the intent to avoid parking at the final destination. In other words, a passenger might have still driven and parked but ridesourcing allowed him/her to do so in a different location. For example, parking downtown might be limited and expensive so a passenger decides to drive to a location as close as possible to the destination where parking is more abundant and/or cheaper. Then, she/he requested a Lyft/Uber ride to reach the final destination, thus benefiting from cost and time savings of a shorter ridesourcing trip. This has both positive and negative implications for transportation. Within this section, I analyzed the specific trip answers to the question For this trip, how would you have traveled if Lyft/Uber wasnt an option? (Q5) under two conditions: i) the mode replaced is one of the driving option; and ii) driving is not part of the connection trip (Q9 & Q10). Figure VIII.I illustrates the percentage of respondents that otherwise would have driven and needed a parking location, with a result of 26.4% (82 O-D trips) of all respondents.
70


Parking Not Needed
(Percentage of all rides replaced by ridesourcing)
Drive alone Carpool (drive) Car rental, Total ("would have
Carsharing driven")
Q5: "How would you have traveled if Lyft/Uber wasn't an option?"
Figure VIILI. Ridesourcing Replacing Driving Trips (Parking)
General Use
In the general questions section of the survey, I asked passengers about travel behavior changes for different modes (Q25), For the next few questions, complete the sentence based on your travel today compared to the past. The question of interest about parking is Because of ridesourcing, I drive... with the option to respond a lot less, a bit less, about same, a bit more, or a lot more. Figure VIII.II provides the results with a third of participants 14.3% plus 19.9% stating that they drive a lot less, which has implications for reduction in parking demand and slightly higher than the percentage of respondents to the specific trip replacement question. It was not expected that passengers would increase their driving, but seven passengers (2.4%) responded to this question with a bit more and a lot more. Based on my experience and interactions with the passengers, this might be explained under two scenarios. The first one is that passengers share a vehicle with others in their household; the other household drivers begin relying more on
71


ridesourcing, which facilitates greater vehicle availability and more driving for the respondent. The second explanation is that passengers are also ridesourcing drivers, which turned out to be the case for three of the survey respondents.
Q25: Completed the sentence based on your travel today compared to the past. Becuase of ridesourcing, I drive...
A lot less A bit less About same A bit more A lot more
Figure VIILII. Travel Behavior Change (Driving today compared to the past)
Locations, Trip Purpose, and Connectivity to Transit Stations Specific Trip
Data in Chapter V includes the O-D matrix (Table V-I) for all the rides in the study that contains both the driver and passenger datasets. In order to understand the trips that influence parking those that otherwise would have driven as described in the previous subsection -1 created an O-D matrix (Table VIII-I) exclusively for the Lyft/Uber rides that replaced driving showing the locations that could potentially reduce parking supply. The most common origin and destination locations were home and places that people go out for social trips. The third common location was work trips with an even number of origins
72


and destinations. Airport was also a common destination in the dataset, with 14 trips most of them originating at passengers homes.
Table VIII-I. O-D Matrix (Driving Trips Replaced)
\ DESTINATION Home Work School Shopping/ Going Out/ Errands Social Airport Hotel/ Airbnb Family/ Friend Other Totals
ORIGIN
Home 0 5 1 1 19 13 0 1 2 42
Work 2 2 1 0 0 0 4 0 1 10
School 1 0 0 0 0 0 0 1 0 2
Shopping/Errands 1 0 0 0 0 0 0 0 0 1
Going Out/Social 8 0 0 0 2 0 0 2 1 13
Airport 0 0 0 0 0 0 1 0 0 1
Hotel/Airbnb 0 2 0 0 0 0 0 0 2 4
Family /Friend 2 0 0 0 1 0 1 0 0 4
Other 1 1 0 0 0 1 2 0 0 5
Totals 15 10 2 1 22 14 8 4 6 82
General Use
In order to understand travel behavior changes in general, I created Figure VIII.Ill showing the distribution of trip purpose for all respondents, and for the 101 passengers that stated driving either a lot less or a bit less in question Q13. As the figure shows, social trips (i.e. go out) is the number one trip purpose for all respondents as well as those ridesourcing users that are driving less.
73


Q13: In general, what kind of trips do you use ridesourcing for? (Check all that apply)
300
200
100
194


JT
&
o
{S


ep
295
i All Respondents
i Respondents that stated "driving less"
84 H101
| 26 __28 _ |





Sv
&
Figure VIILIII. Ridesourcing Trip Purpose (All respondents and those Driving less)
Connectivity to Transit Stations
Ridesourcing advocates state that these services are solving many of the issues with connecting to public transportation (i.e. the last mile). In order to explore this topic further, I recorded trips with transit stop locations as the origin or destination. The passenger survey also included questions in regards to connectivity Q9 & Q10 as described in Chapter VII for the specific trip, and for general use with Q22: Have you ever used another mode of transportation in addition to Lyft/Uber for a single trip? For example, using Lyft/Uber to/from a bus station, Q23 asks: How often have you used other modes of transportation in addition to Lyft/Uber for a single trip? and Q24: Please write all the modes of transportation you have used in addition to Lyft/Uber for a single trip: Table VIII-II includes the results for the specific trip and general use whenever passengers were replacing driving modes or were connecting to bus or rail at a transit station.
74


Table VIII-II. Connectivity to Transit Stations
Q9. Ride connecting with other mode (n=311) No 294 94.5%
Yes 17 5.5%
If yes, number of rides replacing driving and connecting to transit 3 1.0%
Q22. Have you ever connected with other mode? (n=293) No 233 79.5%
Yes 60 20.5%
If yes, number of passenger that stated driving less and public transportation (e.g. bus, rail) as the connection mode 21 7.2%
Parking as a stated reason to choose Ridesourcing
This subsection analyzes parking as a reason for someone to use ridesourcing over other modes, both for the specific trips and for general use.
Specific Trip
Passengers stated the main reason that led them to choose Lyft/Uber over other options (Q8) for the actual O-D ride. Figure VIII.IV includes the percentage for each reason including for all passengers, as well as those that would have driven if ridesourcing was not available. Parking is difficult/expensive is highlighted in the responses as one of the reasons to choose ridesourcing.
75


General Use
Figure VIII.IV. Main reason for choosing Lyft/Uber for Actual Ride
___ _____r,______________________________,_____________________jver other options?
Going out/drinking Don't have a car available Time (e.g. in a rush)
Parking is difficult/expensive
8.0%
19.5%
Cost
Can't drive (no license, DUI, injury) Weather
Public transportation not available Carrying something Tired/Feeling sick Able to do something while riding
Other
________________ 8.5%
7.4%
0.0%
4.8%
______Z4%
3.9%
0.0%
0.0%
3.9%
2.6%
0.0%
2.6%
1.9%
3.7%
9.8%
8.4%
Passengers that "would have driven"
All rides
0.0%
10.0%
20.0%
30.0%
36.6%
40.0%
On


Similar to the previous subsection, question Q17 ask the main reasons to choose ridesourcing over other modes, allowing respondents to check up to three reasons. In addition to this question, I used the driving frequency question Q21 to create Figure VIII. V. I selected the top five reasons from the dataset per driving frequency. In this graph, I highlighted parking again to show the clear relationship between high frequency of driving and parking. Parking is the second most important reason for frequent drivers to choose ridesourcing.
Driving Frequency and Reasons
Don't have a car available
36
13
13
29
Public transport not available
50
12
12
Time
c
o
gj (e.g. in a rush)
cc
37
13
16
22
29
Parking is difficult
14
27
46
Going Out, Drinking
15
17
28
68
Never
Drive
Rarely Sometimes Regularly
Always
Drive
Drive Frequency
Figure VIII.V. Driving Frequency and Trip Purpose
77


Chapter Conclusions
With evolving transportation services and the promise of autonomous vehicles, changes in the transportation industry are creating a window of opportunity to help dissolve the car-dependency and automobile focus of transportation infrastructure in the U.S. and around the globe. Thus, this study aims at investigating the reciprocal influence of an evolving transportation service ridesourcing in terms of a very important transportation infrastructure parking so we can model, design, and build future transportation infrastructure.
I conducted a cross analysis to compare specific Lyft/Uber rides and general travel behavior for three inter-related parking topics. First, I looked at mode replacement finding that a third of all Lyft/Uber rides were replacing trips that would have needed parking.
Then, I looked in more detail at the origin and destination locations, travel behavior changes, trip purpose, and connectivity to the transit system. The results show that passengers that usually drive (always drive and regularly) are using ridesourcing mostly to go to the airport, when out of town, and for going out/social trips. Thus, parking capacity could be reduced at these types of locations (e.g. airport, bars, and stadiums). With respect to connection to the transit system, very few Lyft/Uber rides were used with this purpose in mind.
Finally, I analyzed parking as a stated reason for using ridesourcing, showing that for the specific trip, parking difficulty/cost is the fourth most cited reason for all trips, and the second most cited reason for trips replacing driving. For general use, the passengers that drive the most (always drive and regularly) stated that parking is difficult/expensive as the second most cited reason why they chose ridesourcing over other modes.
78


The results from this chapter suggest that thanks to ridesourcing, we could easily decrease the parking supply because of two reasons: ridesourcing is replacing driving trips and parking is one of the main reasons people choose not to drive in the first place. Thus, parking can be used as a TDM tool to influence travel behavior. The less parking provided, the less driving desired, especially when people have choices to shift to other modes of transportation.
Continuation of this positive cycle would facilitate decreasing car dependency. In the future, autonomous ridesourcing vehicles would enable us to continue this trend more rapidly. In preparation for this drastic change, we should adapt the current transportation infrastructure to reduce parking stock and make room for more desirable land uses, especially in urban areas where land is costly. For example, we could eliminate parking minimums in the generation rate manual so developers build less parking spaces and more building amenities. They should offer residents different transportation options and plan to reaccommodate parking spaces for future use. Similarly, common destinations such as the airport or hot spot areas for social activity (e.g. restaurants, bars, stadiums, and nightlife) shall remove parking spaces or design future buildings with less parking capacity. In turn, this could facilitate more livable and desirable places to walk, bike, and take public transportation.
79


CHAPTER IX
TRAVEL BEHAVIOR CHANGES
Ideally, mode choice models are run and tested with robust data containing information regarding travel activities (e.g. mode of transportation, frequency, time of days, etc.) and specific individual characteristics (e.g. socio-demographics, transportation resources, land use, travel behavior style, etc.) from a random population sample. While I do not have the necessary data for a traditional mode choice model, I can investigate the most probable mode replacement for each individual, and explore the variables and reasons influencing travel behavior change from previous modes of transportation after the introduction and evolution of ridesourcing. Thus, this study chapter has three main objectives:
1. Collect the data necessary for integrating ridesourcing services into a research framework to help current travel models move past their simplistic focus on traditional modes of transportation (i.e. car, transit, walk, and bike);
2. Use ridesourcing as a shock into the system to evaluate changes in travel behavior as well as relationships between travel behavior, trip purpose, and stated reasons for mode replacement; and
3. Identify and analyze components from the framework such as modality styles.
The term modality style is defined as a certain travel mode or set of travel modes
that an individual habitually uses by Vij, Carrel, and Walker (2013). The concept is derived from empirical studies on higher-level orientations, lifestyles or mobility style, both in short-term decisions and long-term choices, which counters the conventional assumption that people choose a mode independently for every trip.
80


Conceptually, this dissertation builds upon the existing body of literature on the
influence of ridesourcing on modality style, mode replacement, and travel behavior. Methodologically, this work contributes to the understanding of modality styles, changes to modality styles with the introduction of additional transportation options ridesourcing in this case and the influence of modality styles on travel behavior. More generally, this dissertation contributes to the body of knowledge considering transportation planning and future travel demand models. Ultimately, this dissertation sheds light on the travel behavior impacts of ridesourcing with the current trends we are experiencing in our transportation systems to better inform urban planners, policy makers, and transportation engineers.
Chapter Literature Review
Understanding travel behavior and the related decision-making is a very complex area of study. One of the most common ways to rationalize and forecast travel behavior and decisions is through travel demand models. One of the issues with travel demand models is that the outcomes are dependent on the assumptions used in the model, and traditionally, travel demand models assume that an individual is fully aware of the range of transportation options so that the choice is based on utility maximization theory. Utility maximization derived from economic theory assigns a utility value for each transportation alternative and assumes that the mode with the highest utility value will be chosen. While this theory is well established in economics, it contains some limitations. Calculating the actual utility of any good is very complex and typically contains several attributes that are not realistic to measure in many scenarios such as inhibited values, attitudes, perception, and beliefs that could relate to ingrained lifestyles and deeply established habits for certain modes. Individuals adopt different patterns of consumption behavior, not only based on utilitarian needs, but also
81


because they express self-identity, leading to a person making a choice as a result of how to act and who to be (Giddens, 1991).
Transportation researchers have studied links between lifestyles and travel behavior since early 1980s by classifying groups that depend on the time spent at different activities, weekly travel patterns, or travel expenditures (Kitamura, 2009; Pas, 1988; I. Salomon & Ben-Akiva, 1983). More recently, the literature evaluated the effects on transportation choice by the influence of a persons mobility decisions including land use (Handy, Cao, &
Mokhtarian, 2005; Kitamura, Mokhtarian, & Laidet, 1997; Krizek & Waddell, 2002) and mobility resources (e.g. automobile ownership, bicycle ownership, public transportation pass) (Choo & Mokhtarian, 2004; Vredin Johansson, Heldt, & Johansson, 2006). Specific to carsharing, recent studies suggest that carsharing members have different mobility resources compared to the representative population, with fewer cars per households and higher levels of bike ownership and public transportation passes. In general, carsharing seems to broaden the mix of transportation modes and increase the distribution of transportation modes, including intermodality (mixed-modes) and multimodality (Johanna Kopp et al., 2015; Le Vine, Lee-Gosselin, et al., 2014).
In an attempt to capture modality styles and their influence on mode choice decisions, Vij et al. (2013) developed a model framework within the context of travel demand models using the Latent Class Choice Model (LCCMs). LCCM models were first developed in the field of marketing sciences (Kamakura & Russell, 1989) and include two components: a class membership model; and a class-specific choice model. LCCM is seen as a solution to the black box of correlation structure from the continuous mixture distribution in most widely known travel demand models (Walker & Ben-Akiva, 2011).
82


This dissertation uses the framework by Vij et al. (2013) as a foundation to incorporate ridesourcing into modality styles and mode choice modeling. The proposed framework facilitates the analysis of the three objectives in this part of the dissertation by looking at the full datasets and paying special attention to the passengers stated responses regarding travel behavior in general and socio-demographic characteristics. Utilization of the full dataset is needed to disentangle the effects of using ridesourcing on travel behavior by controlling for factors such as modality resources and modality styles auto-oriented (i.e. drivers), multimodal, or non-drivers of passengers prior to start using ridesourcing services.
Chapter Data and Analysis
For this study, I used the dataset containing all 311 passenger surveys same as Chapters VII and VIII and adapted the previously described travel demand framework (Figure IX.I) to use in this dissertation.
Figure IX.I. Travel Demand Framework to Study Ridesourcing
83


The framework shows that for a given origin-destination pair and time, there is a universal transportation set of mode options available. From this universal set, there is an individual subset comprised of the transportation options from the universal set minus the discarded options by the individual. Individuals discard certain transportation options due to perceptions, values, habits, preferences, experience, or because they might not be aware of the existence of a transportation service. Below the individual subset, there is the utility of travel mode, which is the evaluation of each travel mode contained in the subset. The person then chooses a travel mode from the subset based on the utility evaluation. This is done for every given origin-destination pair and time; so every person has a group of individual subsets given the situation.
The group of individual subsets is analogous to what Vij et al. (2013) define as modality style in their model framework. In this dissertation, I identified four different classes of modality styles:
Driver or car-oriented: if the individual stated driving as the main mode of transportation
Multimodal: if individuals use several modes of transportation
Non-driver or multimodal without car: if the individual rarely, if ever, drives
Bi-style: if the individual assumes a combination of any of the three previous modality styles.
To my knowledge, this dissertation is the first study to introduce a bi-style modality classification, which hypothesizes that an individual may adopt two completely different modality styles according to several factors and attributes of travel. For example, a person could normally act as a multimodal traveler, but for trips transporting a child, the person
84


would only consider the car. In this case, the person is bi-style (first: multimodal; then: car-oriented). Similarly, a person that normally only considers the car as the mode of transportation might behave differently for leisure trips (e.g. going out to eat or drink). This persons classification is bi-style (first: car-oriented; then: multimodal without car). Previous studies on carsharing systems provide indirect insights into the bi-modality style topic, with the distribution of journey purposes per mode showing that individuals use personal cars, taxi, and carsharing differently depending on the type of journey (Le Vine, Lee-Gosselin, et al., 2014). Another insight is the potential effect of ridesourcing on drunk driving crash reduction (Jones, 2015), which implies that people may be using these services more often for leisure trips but not necessarily for other purposes. The bi-modality style classification is an important distinction because it allows travel demand modes to explain travel behavior better. Accordingly, we could create better policies and more effective planning to influence modal shifts away from SOV.
The modality style classification is sensitive to the group of individual subsets; therefore, I established time limits to determine modality style classifications for a specific time period. Once a period is determined, I measured changes in modality style due to specific inputs. An individual could potentially change modality style with the introduction of several factors: i) changes in characteristics of the individual and households (e.g. income or expenditures, employment status, marital status, parenthood status); ii) changes in location (e.g. home, work, or school); iii) changes in infrastructure or services (e.g. improvement on certain mode infrastructure, evolving transportation services); or iv) changes in mobility resources (e.g. personal cars, bicycles, public transportation pass, carsharing membership, or ridesourcing membership). For this dissertation, I surveyed passengers once but the survey
85


included questions specific to the current trip as well as more general questions regarding travel behavior changes due to the introduction of ridesourcing. Establishing travel behavior prior to the introduction of Lyft/Uber and controlling for explanatory factors (e.g. the person recently moved from out of town, the passenger cannot drive anymore due to DUI or illness) is key to understanding travel behavior.
The key questions for data analysis in this chapter are:
Mode frequency (Q21), where passengers stated how often they use each mode of transportation. The multiple choice options were never, rarely (special occasions, monthly), sometimes (~ 1-2 times per week), regularly (at least 3 times per week), and always.
Travel Behavior, based on question Q25: For the next few questions, complete the sentence based on your travel today compared to the past, and followed by the following subsections sentences:
o Because of ridesourcing, I go to places... o Because of ridesourcing, I drive... o Because of ridesourcing, I use public transportation... o Because of ridesourcing, I bike or walk... o Because of ridesourcing, I take taxis
With the options to answer: A lot less, A bit less, About same, A bit more, or A lot more.
Change in resources (Q19): Do you (or your household) own fewer cars because of ridesourcing? with Yes or No answer options.
86


Besides questions on mode frequency and travel behavior changes, I reevaluated several other questions such as trip purpose (Q13) and stated reasons (Q17) from the previous chapters.
In the last question, I recorded passengers emails for a possible follow-up study,
Q38: In the future, we would like to do a follow-up study on ridesourcing travel changes. If its OK for us to contact you about a future study, please share your email address.
Chapter Results
I organized this chapter result section in three sub-sections. The first one presents mode frequency and travel behavior changes. The second part investigates relationships between frequency of use, trip purpose, and stated reasons. The last subsection analyzes modality styles.
Mode Frequency and Travel Behavior Changes
Figure IX.II presents frequency of use per mode of transportation. About half of ridesourcing users stated that they always or regularly drive alone. In contrast, only 0.7% of passengers use taxi always or regularly. Carpool, public transportation, and bike/walk modes have a more evenly frequency of use distribution.
87


Q21: Thinking about your typical week travel, how often do you?
Always "Regularly "Sometimes "Rarely "Never
100.0%
80.0%
60.0%
40.0%
20.0%
0.0%
Drive alone Carpool Use public Bike or Walk Taxi
transportation
Mode Frequency (n=292) Never Rarely Sometimes Regularly Always
Drive alone 92 31.5% 26 8.9% 32 11.0% 45 15.4% 97 33.2%
Carpool 84 28.8% 90 30.8% 98 33.6% 18 6.2% 2 0.7%
Use public transportation 102 34.9% 73 25.0% 78 26.7% 33 11.3% 6 2.1%
Bike or Walk 75 25.7% 96 32.9% 77 26.4% 26 8.9% 18 6.2%
Taxi 248 84.9% 33 11.3% 9 3.1% 2 0.7% 0 0.0%
Figure IX.II. Mode Frequency
I collected information on travel behavior changes for five different categories new trips, drive, public transportation, bike/walk, and taxis which allow me to measure and compare the magnitude of change for each mode. For example, Figure IX.III shows driving change versus public transportation change.
88


Full Text
xml version 1.0 encoding ISO-8859-1
DISS_submission publishing_option 1 embargo_code 0 third_party_search Y
DISS_authorship
DISS_author type primary
DISS_name
DISS_surname Henao
DISS_fname Alejandro
DISS_middle
DISS_suffix
DISS_affiliation University of Colorado at Denver
DISS_contact current
DISS_contact_effdt 03/31/2017
DISS_address
DISS_addrline 14000 Winding River Ct. R2
DISS_city Broomfield
DISS_st CO
DISS_pcode 80023
DISS_country US
DISS_email alehenao1@gmail.com
future
03/31/2017
14000 Winding River Ct. R2
Broomfield
CO
80023
US
alehenao1@gmail.com
DISS_citizenship
DISS_description page_count 121 doctoral external_id http:dissertations.umi.comucdenver:10823 apply_for_copyright no
DISS_title IMPACTS OF RIDESOURCING – LYFT AND UBER – ON TRANSPORTATION INCLUDING VMT, MODE REPLACEMENT, PARKING, AND TRAVEL BEHAVIOR
DISS_dates
DISS_comp_date 2017
DISS_accept_date 01/01/2017
DISS_degree Ph.D.
DISS_institution
DISS_inst_code 0765
DISS_inst_name University of Colorado Denver
DISS_inst_contact Civil Engineering
DISS_processing_code D
DISS_advisor
Marshall
Wesley
E
DISS_cmte_member
Marshall
Wesley
E
Janson
Bruce
Silverstein
JoAnn
McAndrews
Carolyn
Main
Debbi
Krizek
Kevin
DISS_categorization
DISS_category
DISS_cat_code 0543
DISS_cat_desc Civil engineering
0999
Urban planning
0602
Behavioral sciences
DISS_keyword Lyft, ride-hailing, ridesourcing, TNCs, Uber, VMT
DISS_language en
DISS_content
DISS_abstract
DISS_para The transportation sector is currently experiencing a disruption with the introduction and evolution of technology and transportation services such as bikesharing, carsharing, on-demand ridesourcing (e.g. Lyft, Uber), and microtransit (e.g. Bridj, Chariot). As these new layers of technology-based transportation options begin to flourish, it is important to understand how they affect our transportation systems and society. This doctoral dissertation analyzes the impacts of ridesourcing on several areas of transportation including: efficiency in terms of distance – Vehicles Miles Traveled (VMT) versus Passenger Miles Traveled (PMT) – and travel times, mode replacement, VMT increase, parking, transportation equity, and travel behavior. Realizing the difficulty in obtaining data directly from Lyft and Uber, this research employs an innovative approach by the author becoming an independent contractor to drive for both companies; this allowed the author to gain access to exclusive data and real-time passenger feedback. The datasets include actual travel attributes – such as times, distances, and earnings – from 416 rides (Lyft, UberX, LyftLine, and UberPool), and travel behavior and socio-demographics from 311 passenger interviews. This dissertation estimates a low ridesourcing efficiency rate compared to other modes, mix of modes replacement, overall increase in VMT, decrease in parking demand, low wages (i.e. net earnings) for drivers, travel behavior changes for users, as well as relationships between modality style, trip purpose, and stated reasons for mode replacement. These results give us insights into the impacts of ridesourcing on several key aspects of transportation. This, in turn, will help cities and transportation organizations better account for ridesourcing in their planning and engineering processes (e.g. travel demand models) as well as policy decisions.
DISS_supp_abstract
DISS_binary PDF Henao_ucdenver_0765D_10823.pdf
DISS_restriction
DISS_repository
DISS_version 2011-11-08 15:37:33
DISS_agreement_decision_date 2017-03-31 18:12:46
DISS_acceptance 1
DISS_delayed_release
DISS_access_option



PAGE 1

IMPACTS OF RIDESOURCING LYFT AND UBER ON TRANSPORTATION INCLUDING VMT, MODE REPLACEMENT, PARKING, AND TRAVEL BEHAVIOR by ALEJANDRO HENAO B.S., University of Colorado Boulder, 2006 M.S., University of Colorado Denver, 2013 A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Doctor of Philosophy Civil Engineering Program 2017

PAGE 2

ii This thesis for the Doctor of Philosophy degree by Alejandro Henao has been approved for the C ivil Engineering Program by Bruce Janson Chair Wesley E. Marshall, Advisor JoAnn Silverstein Carolyn McAndrews Debbi Main Kevin Krizek Date: May 13 2017

PAGE 3

iii Henao, Alejandro (Ph.D., Civil Engineering Program ) Impacts of Ridesourcing Lyft and Uber on Transportation includi ng VMT, Mode Replacement, Parking, and Travel Behavior Thesis directed by Associate Professor Wesley E. Marshall. ABSTRACT The transportation sector is currently experiencing a disruption with the introduction and evolution of technology and transportatio n services such as bikesharing, carsharing, on demand ridesourcing (e.g. Lyft, Uber), and microtransit (e.g. Bridj, Chariot) As these new layers of technology based transportation options begin to flourish, it is important to understand how they affect ou r transportation systems and society. This doctoral dissertation analyzes the impacts of ridesourcing on several areas of transportation including: efficiency in terms of distance Vehicles Miles Traveled (VMT) versus Passenger Miles Traveled (PMT) and travel times, mode replacement, VMT increase parking, transportati on equity, and travel behavior. Realizing the difficulty in obtaining data directly from Lyft and Uber, t his research employ s an innovative approach by the author becoming an independent contractor to drive for both companies ; this allowed the author to gain access to exclusive data and real time passenger feedback. The datasets include actual travel attributes such as times, distances, and earnings from 416 rides (Lyft, UberX, LyftLine, and UberPool), and travel behavior and socio demographics from 311 passenger interviews. This dissertation estimates a low ridesourcing efficiency rate compared to other modes, mix of modes r eplacement, overall increase in VMT, decrease in parking demand, low wages (i.e. net earnings) for drivers, travel behavior changes for users, as well as relationships between

PAGE 4

iv modality style, trip purpose, and stated reasons for mode replacement. These res ults give us insights into the impacts of ridesourcing on several key aspects of transportation. This, in turn, will help cities and transportation organizations better account for ridesourcing in their planning and engineering processes (e.g. travel deman d models) as well as policy decisions. The form and content of this abstract are approved. I recommend its publication. Approved: Wesley E. Marshall

PAGE 5

v DEDICATION I dedicate this work to m y wife Gusty. She has been a constant supporter, contributor and ha s carried a lot of weight in our lives. Without her, this project would not have been possible. To my boys, Toms and Andrs, who are my everyday inspiration To my parents, who instilled in me the unmeasurable value of education and have supported me u nconditional ly in all aspects of my life.

PAGE 6

vi ACKNOWLEDGMENTS I would like to thank Dr. We sley Marshall for mentoring me and providing extensive support and advice by sharing his time, knowledge, and funding for t his doctoral dissertation I thank Dr. Bruce Janson Dr. Joann Silverstein, Dr. Carey McAndrews Dr. Debbi Main, and Dr. Kevin Krizek for contributing to my education for providing feedback and mentorship, and for serving on my Thesis Committee I would like to thank all members of the Active Communities Transportation (ACT) Research Group and the Integrative Graduate Education and Research Traineeship (IGERT) Program for their con tributions to my research work. I also thank the transportation org anizations (e.g. the Institute of Transportation Engineers CO/WY Section) and local professionals for their advice and financial support; as well as the Mountain Plains Consortium, the Eisenhower Fellowship Program, and the National Science Foundation th rough their IGERT Award and the Bridge to the Doctorate Pro gram for providing funding for this work. Finally, I am grateful to all the Lyft and Uber pass engers Without their patience, time, and willingness t o share, this project would not have been feasible.

PAGE 7

vii TABLE OF CONTENTS CHAPTER I. INTRODUCTION ................................ ................................ ................................ ........... 1 Specific Aims ................................ ................................ ................................ ........... 4 Study Organizati on ................................ ................................ ................................ ... 5 II. BACKGROUND ................................ ................................ ................................ ............. 6 III. LITERATURE REVIEW ................................ ................................ .............................. 10 IV. RESEARCH METHODS ................................ ................................ .............................. 19 Driving for Lyft/Uber and Driver Dataset ................................ ............................. 20 Driving Strategy and Passenger Survey ................................ ................................ 23 Study Area ................................ ................................ ................................ .............. 25 V. DATA ................................ ................................ ................................ ............................ 27 Driver Dataset ................................ ................................ ................................ ........ 27 Passenger Dataset ................................ ................................ ................................ ... 28 VI. DRIVER PERSPECTIVE: TRAVEL TIMES, DISTAN CES, AND EARNINGS ...... 31 Chapter Related L iterature ................................ ................................ ..................... 32 Chapter Data and Analysis ................................ ................................ ..................... 34 Travel Distances and Times ................................ ................................ ............ 36 Ridesourcing Efficiency Rate ................................ ................................ ......... 37 Ridesourcing Earnings ................................ ................................ .................... 38 Chapte r Results ................................ ................................ ................................ ...... 39 Ridesourcing Efficiency Rate ................................ ................................ ......... 41 Ridesourcing Earnings ................................ ................................ .................... 42 Chapter Conclusions ................................ ................................ .............................. 48 VII. VMT IMPACTS ................................ ................................ ................................ ............ 53

PAGE 8

viii Chapter Related L iterature ................................ ................................ ..................... 56 Chapter Data and Analysis ................................ ................................ ..................... 57 Chapter Results ................................ ................................ ................................ ...... 62 PMT/VMT Effi ciency ................................ ................................ ..................... 62 VMT/PMT Ratio ................................ ................................ ............................. 63 VMT before and after ................................ ................................ ..................... 63 Chapter Conclusions ................................ ................................ .............................. 66 VIII. PARKING IMPACTS ................................ ................................ ................................ ... 68 Chapter Data and Analysis ................................ ................................ ..................... 69 Chapter Results ................................ ................................ ................................ ...... 69 Parking Demand ................................ ................................ .............................. 69 Locations, Trip Purpose, and Connectivity to Transit Stations ...................... 72 Parking as a stated reason to choose Ridesourcing ................................ ......... 75 Chapter Conclusions ................................ ................................ .............................. 78 IX. TRAVEL BEHAVIOR CHAN GES ................................ ................................ .............. 80 Chapter Literature Review ................................ ................................ ..................... 81 Ch apter Data and Analysis ................................ ................................ ..................... 83 Chapter Results ................................ ................................ ................................ ...... 87 Mode Frequency and Travel Behavior Changes ................................ ............. 87 Relationships between Drive Frequency and Other Variables ....................... 90 Modality Styles ................................ ................................ ............................... 92 Chapter Conclusions ................................ ................................ .............................. 94 X. OVERALL RESULTS ................................ ................................ ................................ .. 96 Driver Dataset ................................ ................................ ................................ ........ 96 Ridesourcing T imes and Distances ................................ ................................ 96

PAGE 9

ix Ridesourcing Earnings ................................ ................................ .................... 96 VMT ................................ ................................ ................................ ....................... 97 Parking ................................ ................................ ................................ ................... 97 Travel Behavior ................................ ................................ ................................ ...... 97 XI. SUMMARY CONCLUSIONS AND FUTURE WORK ................................ .............. 99 REFERENCES ................................ ................................ ................................ ..................... 103

PAGE 10

x LIST OF TABLES Table V I. Origin Destination (O D) Matrix ................................ ................................ ........ 28 Table V II. Demographics of Ridesourcing Passengers ................................ ......................... 30 Table VI I. Travel Times and Distances Summary Statistics ................................ ................. 40 Table VI II. Time and Distance Efficiency ................................ ................................ ............ 41 Table VI III. Lyft/Uber Fares and Driver Commission ................................ .......................... 42 Table VI IV. Passenger Cost, Driver Earnings, and Actual Commission .............................. 43 Table VI V. Gross Earnings ................................ ................................ ................................ ... 44 Table VI VI. Gross Earnings Lyft compared to Uber ................................ ......................... 44 Table VI VII. Ridesourcing Expense s ................................ ................................ .................... 46 Table VI VIII. Net Earnings (Gross Earnings minus Expenses) ................................ ............ 47 Table VI IX. Net Earnings Lyft compared to Uber ................................ ............................. 48 Table VII I. PMT, VMT Replaced, and Ridesourcing VMT ................................ ................. 62 Table VII II. PMT/VMT, before and after ................................ ................................ ............. 64 Table VII III. VMT by Mode Replacement, before and after ................................ ................ 65 Table VII IV. Extra VMT per year in the U.S. due to Lyft/Uber ................................ ........... 67 Table VIII I. O D Matrix (Driving Trips Replaced) ................................ .............................. 73 Table VIII II. Connectivity to Transit Stations ................................ ................................ ....... 75 Table IX I. Bi modality Style Classification ................................ ................................ .......... 94

PAGE 11

xi LIST OF FIGURES Figure II.I. Lyft and Uber Timeline ................................ ................................ .......................... 7 Figure II.II. LyftLine serving cities ................................ ................................ .......................... 7 Figure IV.I. Lyft and Uber Driver Profiles ................................ ................................ ............. 21 Figure IV.II. Smartphone Apps ................................ ................................ .............................. 21 Figure IV.III. Driver Data Collection Form ................................ ................................ ............ 22 Figure IV.IV. Car Sign for Passenger Survey ................................ ................................ ......... 24 Figure V.I. Ridesourcing Data ................................ ................................ ................................ 27 Figure VI.I. Travel Distances and Times of a Lyft/Uber Driver ................................ ............ 35 Figure VI.II. GPS Tracking of a Lyft/Uber Ride ................................ ................................ .... 36 Figure VII.I. Taxis in Cali, Colombia (Source: ElPais.com.co) ................................ ............. 54 Figure VII.II. Taxi Tracks in Cali, Colombia (Source: ElPais.com.co) ................................ 55 Figure VII.III. Mode Replaceme nt (Q5) ................................ ................................ ................. 59 Figure VIII.I. Ridesourcing Replacing Driving Trips (Parking) ................................ ............ 71 Figure VIII.II. Travel Behavior Change (Driving today compared to the past) ..................... 72 ....... 74 Figure VIII.IV. Main reason for choosing Lyft/Uber for Actual Ride ................................ ... 76 Figure VIII.V. Driving Frequency and Trip Purpose ................................ ............................. 77 Figure IX.I. Travel Demand Framework to Study Ridesourcing ................................ ........... 83 Figure IX.II. Mode Frequency ................................ ................................ ................................ 88 Figure IX.III. Travel Behavior Changes, Driving & Public Transportat ion ........................... 89 Figure IX.IV. Driving Frequency and Trip Purpose ................................ ............................... 90 Figure IX.V. Driving Frequency and Stated Reasons ................................ ............................. 91 Figure IX.VI. Modality Style Classification ................................ ................................ ........... 93

PAGE 12

xii LIST OF ABBREVIATIONS DTT Drive to Transit ETA Estimated Time of A rrival HOV High Occupancy Vehicles Hr Hour IRB Institutional Review Board MAX Maximum MIN Minimum MINS Minutes MPH Miles per Hour O D Origin Destination PMT Passenger Miles Traveled Q# Question Number S OV S ingle Occupancy Vehicle St. Dev. Standard Deviation TDM Transportation Demand Management T NC Transportation Network Company U.S. United States of America VM T Vehicle Miles Traveled WP With Passenger WPMT With Passenger Miles Traveled WTT W alk to Transit

PAGE 13

1 INTRODUCTION Ev olving transportation services such as bikesharing, carsharing, ridesharing on demand ridesourcing (e.g. Lyft, Uber), and microtransit (e.g. Bridj) are becoming increasingly popular all over the world. Many factors including social networks, real time information, and mobile technology allow passengers and drivers to connect through mobile smartphone applications (i.e. apps). In turn, this has led to the creat ion and popularization of technology companies offering app based on demand transportation platforms. As these new layers of technology based transportation options begin to flourish, it is important to understand how they compete and interact with more tr aditional modes. Beyond travel behavior, these tools and evolving transportation services can also significantly impact our transportation systems, society, and the environment; yet, very little data is known and the academic research is minimum to underst and and measure the impacts of these services regarding outcomes such as vehicle miles traveled (VMT), mode replacement, parking, equity, and travel behavior. Providing a more diverse array of travel options should theoretically reduce car dependence and l ower parking demand; however, there remain unresolved questions about what cities actually gain (or lose) in terms of sustainability related outcomes including efficiency, congestion, carbon emissions, and transportation equity issues. Even when replacing single occupancy vehicle (SOV) trips, there are negative effects. For example with VMT, there are additional miles traveled by the ridesourcing driver before passenger pick up or after passenger drop off over and above the actual trip the passenger wo uld have driven in the first place ( Cramer & Krueger, 2016 ; Henao & Marshall, in press ) There is also

PAGE 14

2 a theoretical saturation point where higher ridesourcing supply than demand leaves many drivers circulating without riders, which can cause unnecessary VMT, congestion, environmental issues, and other problems that are not yet documented with these new technology based modal options. While there is widespread information online regarding companies such as Uber and Lyft, the academic literature on ridesourcing is extremely limited due to the lack of open data on these services. Obtaining data for independent academic research from Lyft and Uber is extremely difficult ( Bialick, 2015a ; Levitt, 2016 ) and even when these companies agree to share data, the data is often not adequate for research purposes ( Vacca ro, 2016 ) These lack of data sharing, but perhaps they do to avoid showing the potential negative impacts in our transportation system. C ity officials and transit advocates have expressed concerns about th e lack of open data and potential problems with ridesourcing such as congestion competition with public transportation and equity issues ( Flegenhe imer & Fitzsimmons, 2015 ; Grabar, 2016 ; Rodriguez, 2016 ) Without appropriate data, measuring impac ts is not possible; and even when such data is available, investigating short term and long term impacts of ridesourcing on travel behavior such as the travel modes replaced by ridesourcing and why people shifted from a previous mode remains extremely difficult. There are still limitations with regard to measuring new trips that may not have occurred before (i.e. induced travel), modality resources (e.g. car ownership), and modality style (e.g. car oriented) of users as well as multimodality (i.e. avail ability of several modes) and intermodality (i.e. combination of various modes for a single trip or mixed modes). This combination of problems makes

PAGE 15

3 analyzing the impact of these services on the overall transportation system exceedingly difficult. Due to t he complexity of this topic, this dissertation first proposes a comprehensive framework aimed at starting the conversation on the type of data that needs to be collected, the questions that researchers need to be asking, and pointing out issues that might arise with conventional research methods. For examp le, if we ask someone that does not own a car what they would have done without Lyft/Uber for a specific trip, they might answer transit. In theory, the ridesourcing trip is classified as a negative enviro nmental impact. However, a more comprehensive research framework might reveal that the decision not to own a car in the first place was made in part due to the availability of Lyft/Uber. Considering such long term car ownership decisions would now expose t he ridesourcing trip as a positive environmental benefit. Beyond looking at the travel modes replaced by ridesourcing, the framework also includes insights from individuals on the process of why a specific mode was selected over the alternatives. For examp le, what is the role of travel time, travel cost, parking, and other factors in the decision making process? Such insight would help provide researchers with the ability to investigate the impact of ridesourcing on a region or city in terms of VMT and park ing demand. It may also facilitate studies across different geographical areas (e.g. urban vs. suburban, city size, density, etc.) where we could find differing impacts in different contexts. In other words, could ridesourcing have, for example, positive i mpacts in more suburban areas and negative impacts in more urban areas? Or could the contrary be true? The intent is to provide a framework that will allow such questions to be explored, and then to carry out the research.

PAGE 16

4 The overall goal of this disserta tion is to start fill ing the gap in the academic literature and help researchers study the effects of evolving services such as ridesourcing and start measuring these impacts on transportation. This, in turn, will help cities and transportation organizations better account for th e impacts of evolving transportation services in their policies, transportation planning, and engineering processes. Specific Aims The specific aims and key contributions of this research are that I will build upon the existing literature on evolving trans portation services by: 1. Developing a comprehensive research framework to study ridesourcing 2. Collecting unique and interrelated datasets of ridesourcing drivers and passengers 3. Developing a ride sourcing survey for passengers seeking Institutional Review Board (IRB) approval 4. Measuring travel distances, times, earnings, and its efficiencies from the driver perspective 5. Measuring the VMT and parking demand impacts of ridesourcing services 6. Investigating travel behavior changes by assessing what travel modes are replaced by these evolving transportation services; and evaluating the factors associated with why people shifted from their previous travel modes and for what trip purposes. 7. Developing a framework for a mode choice model that would allow for integrating r idesourcing services into regional travel models.

PAGE 17

5 Study Organization This dissertation is organized into eleven chapters. Chapter II provides a background for ridesourcing including a history and overview of L yft and Uber. Chapter II I (Liter ature Review) o verviews the topic of evolving transportation services and covers the limited research in this area. In order to better understand how to do re search on ridesourcing services, the first step is to develop a comprehensive research framework Thus, Chapter IV is devoted to this, and includes research methods, city choice, and data collected for its application in this dissertation. Chapter V presents the data. The first three objectives are addressed in Chapter IV and V Objective four is add ressed in Chapter V I (Driver Perspective: Travel Times, Distances, and Earnings), the fifth objective in Chapters V II (VMT Impacts) and VI II (Parking Impacts ), and the sixth in Chapter IX (Travel Behavior Changes). Chapter I X is a summary of results and Ch apter X1 finalizes this dissertation with overall conclusions, recommendations, and future research. Assisting the reader and for better organization, each of the four paper chapters (Chapters V I through IX ) includes its own detail section on literature re view, specific data and analysis, chapter results, and chapter conclusions for each detail topic

PAGE 18

6 BACKGROUND While Lyft and Uber in their curre nt form are mostly known for their regular Lyft and UberX services, and carpool options: Lyf t Line and UberPool, they offer other options and have evolved from a variety of services in their history (Figure II. I ) For example, Uber started as a black car limousine service called UberCa b launched in San Francisco in 2010 ( McAlone, 2015 ) while Lyft co founders Logan Green and John Zimmer previously co founded Zimride, a tru e rideshare platform created to connect drivers and passengers through soc ial networking. Green and Zimmer started Zimride in 2007 and sold it to Enterprise Holding i n July 2013 ( Lawler, 2014 ) While Lyft was launched in June 2012 with it s original regular Lyft service, Uber did not unveil its regular UberX service until July 2012, a couple of yea rs after it started with UberCab LyftLine and UberPool services started in 2014 but are only available in certain metropolitan cities ( Lyft Blog, 2016 ; Uber Newsroom, 2014 2016 ) For example, Figure II II shows the cities where L yftLine was in servic e or about to launch as of April 2016 ( including Denver ) As of the summer 2016, Uber was already in 450 cities globally, and completed two billion trips in its life span. One billion rides were completed in six years, while the same number of rides were completed in six months ( Somerville, 2016 ) mated valuation continues to grow and currently is at $62.6 billion, making it the most valuable transportation company in the world; and currently, without owning any vehicle, infrastructure, or having to hire drivers as employees. Lyft operates exclusive ly in the U.S. and is valued at approximately $5.5. billion dollars ( B. Salomon, 2016 )

PAGE 19

7 Figure II I Lyft and Uber Timeline Figure II II LyftLine serving cities (Source: Lyft Blog, April 5, 2016)

PAGE 20

8 One of the latest news releases shows that Lyft is giving rides at a rate of 17 million U.S. rides per month. It i s estimated that Lyft has around 20% of the market share, making Uber the ridesourcing company with the highest volume in the U.S. These numbers show the magnitude of Lyft and Uber and their influence on the way people get around Uber and Lyft path has not been worry free. They have to constantly deal with different situations such as regulations, protests, and lawsuits from taxi companies, city officials, and drivers claiming employment rights. They als o have taken advantages of the terminology in their marketing strategies. The terminology of new and evolving transportation services can be confusing and sometimes ill defined by the transportation sector. Intentionally or unintentionally, many accredited people and companies use the terminology incorrectly, which can mislead public perception and general use of the services. A recent example is the misused word ( Goddin, 2014 ) Ride hailing services such as Uber or Lyft let people use smartphone apps to book and pay for a private car service or in some cases, a taxi. They may also be called ride booking services. Do not use ride sharing ( Warzel, 2015 ) While there seems to be a consensus that these services are not ridesharing, there is still no clearly a defined term. Some of the names demand n an attempt to be consistent with previous academic research ( Rayle, Dai, Chan, Cervero, & Shaheen, 2016 ) and to allow for possible future variations of such schemes to be housed under the same header, this study

PAGE 21

9 sourcing of rides from a for fare driver pool accessible through an app based platform.

PAGE 22

10 LITERATURE REVIEW Lyft and Uber are disrupting urban transportation systems and competing with more traditional modes (i.e. car, taxi, transit, walk, and bike), but a minimal number of U.S. cities ha s been able to account the impacts of ridesourcing ( DuPuis, Martin, & Rainwater, 2015 ) The introduction of these services has implications for travel behavior and mode shift, as well as impacts on the ov erall transportation system. Other services such as b ik esharing and carsharing are continuously evolving and increasing users in cities across the globe ( S. Shaheen & Cohen, 2012 ; S. Shaheen, Guzman, & Zhang, 2010 ) In addition, while the academic literature on carsharing and bikesharing systems has provided ins characteri stics (e.g. socio economic demographics, preferences etc. ) and transportation impacts (e.g. car ownership, car use, VMT, reduction s of cars on the network, and mode share), the literature on ridesourcing remains very limited. While there is abundant infor mation online regarding companies such as Lyft and Uber, the academic literature on ridesourcing is very limited, in part due to the ir novelty and lack of open data on these services. Due to the ridesourcing history evolution, and similarity to other service s the few academic studies on this topic compared ridesourcing mostly to the taxi industry and ridesharing services ( Anderson, 2014 ; Cramer & Krueger, 2016 ; Rayle et al., 2016 ) Rayle et al. (2016) did a research study comparing ridesourcing and tradit ional taxis in San Francisco using an intercept survey in spring 2014. The findings from this study indicated that compared to the overall San Francisco population, ridesourcing users tend to

PAGE 23

11 be a lot younger, have higher incomes, h ave lower car ownership, and frequently travel with companions. This study also show s that compared to taxis, ridesourcing customers experienced shorter waiting times Participants in this study said that ridesourcing both substitute and complement public transit, walking, and biking ; and 8% of survey respondents s tated that they would not have traveled (i.e. induced travel effect ) if ridesourcing services were not available. More recently, the S hared Use Mobility Center investigated the relationship betwee n public transportation and shared modes, including bikesharing, carsharing, and ridesourcing in seven U.S. cities. This report found that the higher the use of shared modes, the more likely people use public transportation, own fewer cars, and spent less on transportation. It also shows that shared modes complement public transportation ( Murphy, 2016 ) Regarding literature not currently published in academia, the website FiveThirtyEight h as published a few articles reg a r ding ridesourcing companies using data acquired via a Freedom of Information Act request. The articles show that in New York, Uber is taking rides away from taxis and generally covers a larger area ( Bialik, Flowers, Fischer Baum, & Mehta, 2015 ; Fischer Baum & Bialik, 2015 ) In another article, FiveThirtyEight argues that for Uber to be worth its $50 billion valuation, it has to complement and attract customers that normally use public transportation. This last article also used data on median income levels by census tract and residential pick up rates showing that lower incomes experienced fewer pickups ( Silver & Fischer Baum, 2015 ) The article compare d general travel cost (using basic assumptions) of public transit, Uber, and the cost to own a car; arguing that Uber in combination with high use (around 65% to 85%) of public transportation can be significantly

PAGE 24

12 cheaper than car ownership. Overall, the articles suggest ed that Uber is affecting mode choice, intermodality, and travel costs (that co uld in turn affect mode choice). Since the literature in ridesourcing is extremely limited, is important to review the literature on a similar service that has evolved over the last few years and contains more in depth studies This is useful in helping understand ridesourcing and f or helping design this newer strand of transportation research Carsharing systems provide a fleet of shared vehicles for short term use where members pay in time increments of minutes or hours. Currently, there are several ca rsharing models including the following variations: round trip or one way (i.e. point to point), station based or free floating, and peer to peer. Round trip station based carsharing is the oldest and most established system, where users need to return th e vehicle at the same fixed station it was checked out. Round trip carsharing started in Europe as early as the 1940s but more successful programs did not began operating until the mid 1990s ( S. Shaheen & Cohen, 2007 ) While most carsharing research is based on the traditional station based round trip carsharing system, the last few years ha ve seen a surge in one way carsharing research. The first services without any fixed vehicle stations Car2go by Daimler and DriveNow by BMW started in 2009 and 2011, respectively ( Firnkorn, 2012 ) As of October of 2014, approximately 4.8 million individuals are members of carsharing programs worldwide with a total fleet of 104,000 vehicles (Shaheen and Cohen, 2014). There have been a number of studies aiming to evaluate carsharing impacts, but the results are not clear with respect to the effects resulting from changes in the launch of a carsharing system. This is probably due to difficulties with respect to data availability,

PAGE 25

13 t imelines, confounding effects, as well as research design and methodologies ( Firnkorn, 2012 ; Graham Rowe, Skippon, Gardn er, & Abraham, 2011 ; J Kopp, Gerike, & Axhausen, 2013 ; Johanna Kopp, Gerike, & Axhausen, 2015 ; Le Vine, Ad amou, & Polak, 2014 ; Stopher & Greaves, 2007 ) Numerous carsharing studies focus on determining impacts on transportation, land use, environmental, and social benefits with some mixed results in certain areas and clear evidence on others. As regards to this dissertation, carsharing research on travel behavior c an be classified and quantified in the following areas: Socio demographics for carsharing users and non users: Studies suggest that carsharing users do not usually represent the overall population with regard to socio economics, demographics, and travel be havior characteristics. Carsharing users tend to be younger, with higher levels of education and income, and live in denser areas with better access to public transportation. Carsharing users also tend to have higher public transit, walking, and biking mod e shares and lower car usage compared to the general population ( Cervero & Tsai, 2004 ; J Kopp et al., 2013 ; Johanna Kopp et al., 2015 ; Martin, Shaheen, & Lidicker, 2010 ; Sioui, Morency, & Trpanier, 2012 ) Car ownership: Studies revealed that car ownership for carsharing members is lower than the general population and non members. Empirical evidence has also shown a reduction in private vehicle ownership after joining a carsharing program by gettin g rid of a vehicle owned or foregoing vehicle purchase ( Cervero & Tsai, 2004 ; Meijkamp, 1998 ; S. A. Shaheen, Cohen, & Chung, 2009 ; Steininger, Vogl, & Zettl, 1996 ) For example, a study on City Carshare in San Francisco indicated that a higher share of members reduced car ownership as compared to a control group of non members, approximately 2 9% versus 8%. Two thirds of members also said they refrain from purchasing a vehicle as compared to 39%

PAGE 26

14 of non members ( Cervero & Tsai, 2004 ) Another study based on a survey in 2010 of members of Communauto, a Montreal carsharing company, concluded that members of the carsharing service have approximately 30% lower car usage compared to the level of those that own a vehicle ( Sioui et al., 2012 ) Another study showed that the average number of vehicles per household dropped from 0.47 to 0.24 ( Martin et al., 2010 ) Car use and vehicle miles traveled (VMT) : A large study across North America on round trip car share subscribers revealed that while most members drive more with carsharing, the minority that drive less are driving less by a higher order of magnitude, which leads to less driving overall. In this study, VMT declined by 27%, and when including those that decided not purchase a vehicle in the first place, it was a 43% reduction ( Martin et al., 2010 ; S. A. Shaheen et al., 2009 ) The first year of City Carshare operation in San Francisco suggested an increase in motorized travel for membe rs ( Cervero, 2003 ) ; however, in the second year of operation, the daily VMT reduced slightly for members and increased for non members ( Cervero & Tsai, 2004 ) Reduction of cars on the transportation network: Based on several carsharing reports in the U.S., carsharing helps remove an aggregate of 9 to 23 vehicles from the road (including both shed aut os and foregone car purchases) per shared use vehicle from the transportation network ( Lane, 2005 ; S. A. Shaheen et al., 2009 ) For example, Cervero and Tsai (2004) estimated that a carsharing fleet of 74 in San Francisco removed approximately 500 vehicles from the streets, equivalent to 6.8 private vehicle per carsharing vehicle. Similarly, a study from Philadelphia fo und that each PhillyCarShare vehicle replaced an average of 23 private vehicles, 11 vehicles from members giving up a car and 12 vehicles from not acquiring one in the first place ( L ane, 2005 )

PAGE 27

15 Mode share : Studies on station based carsharing suggest that some of its members change travel behavior towards public transportation and non motorized modes, while others do the opposite by reducing transit, walking and biking usage; overal l, however, most people tend to increase public transit and non motorized modal use ( E. Martin & S. Shaheen, 2011 ) A study of Ulm, Germany using two different methods reported that after the introduction of a point to point cars haring service, members shift modes and reduce the usage of all other modes of transportation including private cars, public transportation, and non motorized travel ( Firnkorn, 2012 ) Carsharing research on both round trip and point to point carsharing concluded that point to point is a substitute for public transport while round trip carsharing is a complement ( Le Vine, Lee Gosselin, Sivakumar, & Polak, 2014 ) Many of the studies on carsharing research rely on sample surveys to gathe r information on members demographics, current usage of the carsharing service, and prior to joining carsharing travel behavior information ( Lane, 2005 ; E. Martin & S. Shaheen, 2011 ; Martin et al., 2010 ; E. W. Martin & S. A. Shaheen, 2011 ) While these studies provide a basic idea on socio economic demographics and travel behavior patterns at the aggregate level, they are inconclusive on the effects of carsharing because they fail to control for several factors that could affect the results (such as predisposition characteristics of people joining a carsharing system) or by not comparing the study population with a control group. From all the carsharing studies, only a few include a statistical control group in their methodology. Control groups, either on longitudinal or cross sectional research, allo w to correct for some confounding effects that otherwise would be difficult to distinguish from effect results. The best example of the use of control groups is the study over time by Cervero, Golub, and Nee (2007) on City Carshare in San Francisco. After two years of

PAGE 28

16 service, VMT for carsharing members decreased, but it decreased even further for non members; so relativ e to the control group, VMT for members increased. Another example is the research study by Johanna Kopp et al. (2015) where they used a reference group of non carsharing users using an online and app based travel da iry, MyMobility, to collect individual trips over a 7 day period. This was a relatively well designed study (with respect to survey instruments, methodology, and clearly stated limitations) of a free floating carsharing service. The study also implemented a multimodal index by analyzing the distribution of transportation modes of carsharing for users and non users, and stating future research needs to disentangle the effects of joining a carsharing service on mobility behavior, which this dissertation aims to find. Although studies that use control groups are considered to have a better statistical methodological research design, there are still some problems to overcome such as selection and arbitra ry choice of non member sampling that could potentially misrepresent the population. Concerning this dissertation, using latent classes will help understand the modality style of individuals using carsharing in relation to the same classes from the general population. Per the literature review, carsharing members tend to have a more sustainable modality style as compared to the general population, including higher use of non motorized transportation and lower frequency of private car use. In this case, the fair comparison would be to calculate the difference against non members that have a multimodal travel behavior. Another way to compare, track, and measure the impacts of carsharing is using the research design implemented by Firnkorn (2012) using Car2go data. Firnkorn used the following two approaches to triangulate toward the impact of carsharing on travel behavio r:

PAGE 29

17 i) hypothetical travel behavior at present without Car2go; and ii) past mobility travel behavior on top of current behavior with Car2go. Details on the survey methods and meth odology from this study are applicable to this dissertation. However, the auth or states that the two measurement techniques should theoretically have produced the exact same carsharing could easily be different to what that person would do today without carsharing. The results from the few ridesourcing studies were similar to carsharing studies suggesting that carsharing users do not usually represent the overall population with regard to socio economics, demographics, and travel behavior characteristics and users tend to be younger, with higher levels of education and income, and live in denser areas with better access to public transportation. Members also have different mobility resources with fewer cars per households, higher levels of bike ownership and public transportation passes, as well as higher transit, walking, and biking mode shares compared to the general population. The current carsharing, and ridesourcing literature offers a general idea of the socio economic demographics and insights into travel behavior impacts at the aggregate level, but there is no clear understanding at the individual level on the actual motivations why a user chooses a mode over the alternatives. For example from the previous studies, there is no investigation on the r ole of travel time, travel cost, or convenience (e.g. parking) on the utility and mode choice of travel demand models. There is also no implementation of modality style on the effects of carsharing on travel behavior. The changes cannot clearly be attribut ed to carsharing or ridesourcing without knowing the members behavior prior to joining a new service (e.g. car oriented or multimodal) and controlling for the factors that influence travel behavior over time such as individual and household characteristics location

PAGE 30

18 choice, or transportation resources. This dissertation aims to address these problems by implementing a methodology that focuses on a more comprehensive examination of ridesourcing effects on individual travel behavior and overall impacts on the transportation system. As seen in this overall literature review section, independent research on ridesourcing remains very limited. Each chapter cov ering specific topics (Chapter V I through Chapter IX ) includes a more detail review and related literature to each theme.

PAGE 31

19 RESEARCH METHODS The first step in understanding the impacts of ridesourcing is to develop a framework to guide the research and fill the important gaps in the literature. With Dr. Wesley Marshall, we co authored the book chapter amework for Understanding the Impacts of Disrupting Mobility: Impacts of Sharing Economy and Innovative Transportation on Cities ( Henao & Marshall, 2017 ) This study lays out the research framework needed to investigate ridesourcing impacts in t ransportation, emphasizing the need to employ a combination of travel attributes (e.g. travel times), revealed behavior data, and stated response data structures. Many transportation planners and engineers dream of having ridesourcing data to analyze and m ake transportation decisions. While it would be nice to have access to this data, we still have not seen any examples of data sharing from these companies for independent academic research. Realizing the difficulty obtaining data directly from Lyft and Ube r, I decided to become an independent contractor and drive for both companies; this allowed me to gain access to exclusive data and real time passenger feedback. I signed up to drive for both companies in early 2015, initially doing exploratory analysis to determine how viable this methodology would be for collecting data. After the initial test rides, I decided to continue in this direction by developing the research framework and the passenger survey. I then sought IRB approval and applied for research fu nding. attributes

PAGE 32

20 from actual trips including date, time of the day, origin and destination (O D) locations, travel times, travel distances, passenger cost, and driver earnings. The second dataset is the information gathered by surveying passengers during the ac this research. In the spring of 2016, I submitted a research proposal to the Colorado Multiple Institutional Review Board (COMIRB), obt aining IRB approval to interview passengers (COMIRB Protocol 16 0773, Exception APP001 3). Driving for Lyft/Uber and Driver Dataset I conducted my data collection using a sedan vehicle 2015 Honda Civic and a smartphone iPhone 5s to drive as an inde pendent contract for both Lyft and Uber (Figure IV. I driver II ). GoogleMaps and MyTracks helped me to track and record ridesourci ng travel data. Passengers completed the online survey using their own smartphone or via a tablet device, Samsung Galaxy Tab A, that I provided.

PAGE 33

21 Figure IV I Lyft and Uber Driver Profiles Figure IV II Smartphone Apps I used the data collection form presented in Figure IV. III to help guide the travel The ridesourcing driver data includes in formation for each ride such as date and time of the day, weather, pick up and drop off locations, driver earnings, and times and distances broken down by

PAGE 34

22 and d istance, as shown in Figure IV.III Additionally, I collected information about parking; at destination Chapter VI includes a more detail description of each segment for the driving travel times and distances. For the origin and destination locations, I collected the closest cross streets, rather than the address, to maintain confidentiality to double check the data recorded. Figure IV III Driver Data Collection Form

PAGE 35

23 Driving Strategy and Passenger Survey On a typical driving day, I turned on both Lyft and Uber apps and waited until a passenger requested a ride. To be conservative, I generally minimized unnecessary driving; thus, I accepted most of the requests unless there were problems with the app or the pick up location was more than 15 miles away from the driver location (again, this is to minimize driving without a passenger). Once the ride was accepted, I turned off the driving mode for the other service. For example, if it was a Lyft request, the Ube r driver mode was turned off; or vice versa. Then, I traveled to the pick up passenger location and waited until the passenger got into the car to travel to the desired destination. I, as a driver invited passengers to participate in a short survey about ridesourcing both verbally and with signs in the car (Figure IV. IV grad student doing research on transportation. Would you help me by doing a short survey (~6 minutes) about this ride? You can use my tablet or go t o this link www.ride survey.com As the sign indicates, passengers had the option to take the survey on a tablet provided by me, the driver, or use their own device by going to a pre defined website. I n some cases, I conducted a verbal interview with the passenger that covered all the questions included in the survey. I waited until the ride was over to take notes and record the interview. Once the ride ended at the destination location, I turned on the other app and waited for a new passenger request. Once the passenger got out of the car, I tried to find the closest parking space available with the intent to record parking data, and again, to minimize cruising distance without a passenger in the car. D riving for both Lyft and Uber helped minimize the waiting times and cruising distance. For example, there were occasions where new requests came in even before I finished parking. I did all of the data collection by myself

PAGE 36

24 to eliminate bias between drivers to control travel without a passenger (i.e. deadheading minimization ), to reduce surveyor errors and to ensure data quality. Figure IV IV Car Sign for Passenger Survey The passenger survey included three group s of questions: Specific Trip Questions (Q1 Q10): The first section asks passengers questions regarding the specific Lyft/Uber ride and includes questions such as trip purpose, travel mode replacement, and reasons to shift from a previous mode General Use Questions (Q11 Q25): The second part of the survey covers broader questions about travel behavior in general such as modality resources (e.g. car ownership, transit pass, etc.), general ridesourcing use, frequency of use for different modes travel behavior changes, and more general trip purposes and reasons.

PAGE 37

25 Demographic Questions (Q26 Q37): The third section of the survey includes questions regarding characteristics of the individual and household (i.e. socio economic demographics). All s urvey questions are included in Appendix A. Chapter V, about data, as well as Chapters VII, VIII, and IX include a more detailed description of the survey questions in this dissertation. Study Area While Lyft and Uber originated in what they considered an unregulated space, Colorado was the first state in the U.S. to legislatively authorize Lyft and Uber services to operate with a bill signed by Governor John Hickenlooper in June 2014 ( Vuong, 2014 ) This helped make Denver and the surrounding cities an innovative and welcoming location for these evolving transportation services. The Denver metropolitan region comprises a variety of places, covering both urban and suburban areas. For example, it contains very urb an places like Union Station in downtown Denver, as well as low density areas such as those surrounding the Denver International Airport (DIA), located about 24 miles north east of Union Station. This metropolitan area also includes a college town like Bou lder and suburban cities like Westminster or Broomfield in between Denver and Boulder. This diversity of characteristics (e.g. density, race diversity, income levels) makes the Denver region an ideal place to study ridesourcing. Another positive factor in the research design was the randomness of the passenger destinations. As the driver, I did not know w here each ride would end up; thu s, I drove all over the study area and visited many of the places previously described. The only location

PAGE 38

26 that I had contro l over is where I turned on t he app at the beginning of the shift. Thus, I varied my starting location.

PAGE 39

27 DATA Since I signed up for Lyft and Uber in 2015 including the rides in exploratory analysis I gave ar ound 500 rides, transporting over 650 passengers. This dissertation includes 416 collected over a period of 14 weeks mostly during the fall 2016. The flowchart in Figure V. I shows t he types of interconnected datasets. Figure V I Ridesourcing Data Driver Dataset The distribution of the 416 rides for the different services was: 198 regular Lyft rides 164 UberX rides 39 LyftLine rides 15 UberPool rides For this dissertation, I drove a total of 4,950.7 miles, spent a total of 15,529 minutes (or 258 hours and 49 minutes) working as a driver, and earned a total of $4,062.08, including

PAGE 40

28 tips. More detail s on the summary statistics for travel times, travel distances, and earnings can be found in Chapter VI (Table VI I & Table VI IV ). Passenger Dataset As stated before, the passenger dataset from 311 surveys include three types of questions. I analyz ed responses to specific trip questions and general ridesourcing usage in Chapters VII through IX. To give the reader an idea of the origin and destination (O D) combinations, I created the O D matrix shown in Table V I. Of all O D combinations, the three Home ith common origin and Table V I Origin Destination (O D) Matrix Table V II provides description statistics from all 311 passengers surveyed. Comparing the summary statistics to the Denver population, the sample seems very representative of the population. Previous studies have shown that the ridesourcing

PAGE 41

29 population (and carshari ng) does not usually replicat e the area they represent with higher incomes, low min ority representation, and younger users ( Murphy, 2016 ; Rayle et al., 2016 ) The authors from these research papers suggest that these services mostly serve certain populations but I believe is mostly due to the location of the intercept surveys. My research has the advantage of being random by design since I did not know the destination location. Thus, allowing this study to cover a larger area and include populations that are usually not represe nted in this type of studies. The sample has a very close split of male female population Passengers were mostly younger adults but compared to other studies, I had higher participation from persons of ages 55 to 64 and 65 + years old people While two th irds of the sample stated being of white race I obtain ed representation from different races and ethnicities. In contrast to previous studies, income is better distributed between different ranges, and not very far from the Denver population.

PAGE 42

30 Table V II Demographics of Ridesourcing Passengers

PAGE 43

31 DRIVER PERSPECTIVE: TRAVEL TIMES, DISTANCES, AND EARNINGS This chapter focuses on three very important aspects of ridesourcing from the driver perspective: travel times, distances, and earnings. For this study, I used the driver dataset including 416 rides from Lyft, UberX, LyftLine, and UberPool. When driving fo r Lyft and Uber, travel times were measured in minutes and travel distances in miles starting with the pick a rest of the study. These four measurements were recorded for each new ride, and at the end of the shift, pp log measured. This involves the commute at the end of the shift. Note that during the period that data was gathered, Uber and Lyft introduced an option to set a destination filter. This option allows the driver to set a destination filtering the ride requests that go along the same route. I estimated ridesourcing efficiency rates based on WP rides versus total times and distances. Based on the distance efficiency, I also calculated total VMT per 100 with passenger mile s traveled (WPMT), which helps to determine the additional VMT or deadheading experienced in our transportation system due to ridesourcing. Total ridesourcing travel time and distances also allow me to calculate the gross earnings per hour an d per mile Finally, I estimate d ridesourcing driving expenses and net earnings per hour and per mile. This study starts to fill a gap in the literature by studying the effects of ridesourcing on transportation from the driver perspective. My aim is to help cities and regional transportation organizations better account for the impact of technology and evolving

PAGE 44

32 transportation services such as Lyft and Uber in their transportation planning and engineering processes and have a clearer picture of the actual gross earnings expenses, and net earnings current ridesourcing services in terms of i ncreasing efficiency to reduce VMT, due to deadheading and wasted time, and provide highe r earnings for ridesourcing drivers. Chapter Related L iterature While most of the studies mentioned on Chapter III ( Murphy, 2016 ; Rayle et al., 2016 ) focus mainly on the ridesourcing passengers, there are only a few articles that focus on the driver side. Ridesourcing has been mainly compared with taxis. There has been a lot of resistance and controversy with the introduction of ridesourcing since they disrupted the industry, competing and taking away many customers from taxis. Both services are similar in the fact that drivers transport passengers for a fee, but there are many differences including technology innovation, labo r market differences, and government regulations. In terms of driving and time efficiency of ridesourcing and taxi services, C ramer and Krueger (2016) compared the capacity utilization rate of UberX drivers against taxi drivers in a few U.S. cities. Using the aggregated data across all drivers available for both cities, the findings show that the percent of work hours with a passenger ranges from 32.0% to 49.5% for taxis, and 46.1% to 54.3% for UberX. The mileage based capacity utilization measure (i.e. percent of miles driven with a passenger) from the same study was calculated at 39.1% to 40.7% for taxis, and 55.2% to 64.2% for Ub erX. The main limitati was the exclusion of mileage and times drivers have to travel from the point of log out to the end location (i.e. commute home), which overestimates their capacity utilization rate.

PAGE 45

33 The media has put a lot of attention in the income for Lyft and Uber drivers. A Wall Street Journal article in 2013 stated that a typical Uber driver takes in more than $100,000 a year in gross sales ( MacMillan, 2013 ) After this income estimation was questioned, Uber reduced this income characterization and more recently advertise that its drivers earn up to $35 an hour (same as Lyft advertisement). Based on data from October 2014, a study commissioned by Uber found that UberX dri vers were grossing around $17.40 an hour for 20 market cities as a whole ( Hall & Krueger, 2015 ) They also reported taxi drivers and chauffeurs wages of around $12.90 an hour based on the Occupational Employment Statistics partners, who are independent contractors, are not reimbursed for driving expenses, in co ntrast to taxi drivers, who are usually employees. The Ub study was based in 2014, when rates were higher than in 2015 or 2016, and did not include the time drivers have to travel from the point of log out t o the end location, same as previously described for the article by Cramer and Krueger (2016) A recent online article published by BuzzFeed News based on l eaked internal data from Uber reported that Uber drivers earn $12.70 an hour in Detroit, $14.18 an hour in Houston, and $16.89 an hour in Denver before expenses ( O'Donovan & Singer Vine, 2016 ) poor since it underestimates the depreciation cost by using a $16,000 car v alue, overestimates the lifetime expectancy of an average automobile to 250,000 miles, and uses a low gas cost of $1.75 per gallon. It i s also not clear about the insurance, maintenance, and miscellaneous costs associated with driving. It is important to n ote again that these calculations also do not include the commute time and distance for drivers (from the point of log out to the end

PAGE 46

34 location). By not including this additional time and expenses, the reported earnings per hour could be severely overestima ted. This study is the first research that independently analyzes data from the driver perspective using both Lyft and Uber trips, including all the additional travel distances, additional times, and actual gross, expenses, and net earnings per hour and pe r mile incurred by Lyft/Uber drivers. Chapter Data and Analysis I used a total of 416 rides 108 rides pre IRB and 308 with IRB approval for this study. For each ride, the information of interest includes: the service the ride was requested from (Lyft, LyftLine, UberX, or UberPool), travel times, travel distances, and earnings including tips. The data analysis process began by calculating th e break down of travel times and travel distances for each ride (Figure VI. I & Figure VI. II ): t 1 = time a driver has to wait until a new ride request d 1 = travel distance cruising for a ride (if the driver decides to park and wait until a new request, this distance is zero or close to zero) t 2 route to passenger) or estimated time of arrival (ETA) d 2 en route to passenger) t 3 = waiting for passenger time once at pick up location t 4 = travel time from passenger d 3

PAGE 47

35 Figure VI I Travel Distances and Times of a Lyft/Uber Driver

PAGE 48

36 Figure VI II GPS Tracking of a Lyft/Uber Ride In addition to the previous travel times and distances, drivers have to travel to their end locations and commute home once they drop off the last passenger and are finished with the shift. The commute at end is also illustrated in Figure VI.I and includes: t 5 d 4 of Travel Distances and Times The ridesourcing driving time and distance per shift are calculated by the following equations:

PAGE 49

37 For this study, the total ridesourcing driving time is: And the total ridesourcing driving distance is: In terms of VMT and WPMT, the total ridesourcing driving distance can be expressed as follows: Ridesourcing Efficiency Rate To determine the time efficiency rate, I compared the sum of WP times against total times : And the sum of WPMT travel distances against total travel distances for the mileage efficiency rate:

PAGE 50

38 Based on the total VMT equation: the additional percent of WPMT is: Finally, I calculated the total driving miles for every 100 miles transporting passengers (100 WPMT), as follows: Ridesourcing Earnings I calculated driver gross earnings per ho ur and per mile using total earnings divided by the corresponding travel time or travel distance. For example, the gross earnings for all 416 rides was calculated by adding all driver earnings and divided by total time and total mileage, as per the following equations: I also calculated three different scenarios to account for the broad r ange of expenses drivers might incur. The expense rate and calculations are explained in more detail on the results section. After discounting expenses, I estimated the net earnings per hour for all rides, for Lyft only rides, for Uber only rides, includin g before and after tips.

PAGE 51

39 Chapter Results Using the median travel times and distances summary statistics (Table VI I ) from the dataset, a representative day for a ridesourcing driver would be as the following description The Lyft/Uber driver logs on both a pps; he/she tries to minimize the cruising distance (0.2 miles) but has to wait 7.5 minutes (mins) until he/she gets a request. Once the driver accepts the request, he/she spends approximately 5.0 minutes traveling 1.0 miles to the passenger pick up locati on. Then, the driver has to wait 1.0 minutes for the passenger to board the car and start the actual ride. The median time and distance of the actual WP ride is 11.5 mins and 3.6 miles, traveling at an average speed of 28.8 miles per hour (based on a total of 6 106 minutes and 2929.9 miles). After the passenger is drop off, the driver starts the process again waiting for a new ride request but minimizing unnecessary driving. When the driver is done for the day, he/she travels to the desired end location, co mmuting around 12.0 miles in 20.0 minutes (based on median values of 65 commuting trips or shifts). When the sum of all commuting times and distances are equally distributed to all rides, the median total driving time per ride is 32.8 minutes (average of 3 7.3 mins) and the median total driving distance per ride is 8.3 miles (average of 11.9 miles). Foll owing this dataset summary statistics, I divided the chapter results section into two subsections covering rides ourcing efficiency rates (time and distance) and earnings (gross and net earnings after expens es).

PAGE 52

40 Table VI I I Travel Times and Distances Summary Statistics

PAGE 53

41 Ridesourcing Efficiency Rate The time efficiency rate of a ridesourcing driver based on the time a passenger is in the car and total time from driver log in to log out (not accounting for the commute at the end of the shift) is 41.3%, meaning that I, as a driver, during my shift hours spe nt more time without a passenger than with one in the car. For example, if in a shift, I was working for five hours, I only spent just over two hours with passengers in the car, due to all the time spent waiting for a ride, going to pick up the passenger, and waiting for the passengers once I was at the pick up locations. When accounting for commuting time at end of shift, the time efficiency rate drops to 39.3% of total time (t T ) (Table VI II ). Based on distance, the ridesourcing mileage efficiency rate without and with commute at end is 65.4% and 59.2%, respectively. The total ridesourcing driving mileage per every 100 WPMT is 169.0. In other words, Lyft and Uber drivers travel an additional 69.0 miles in deadheading for every 100 miles they are with p assengers. Table VI II Time and Distance Efficiency

PAGE 54

42 Ridesourcing Earnings The rates that passengers pay for Lyft and Uber fluctuates, but traditionally, they have been lowered over time. The percent that Lyft and Uber pay their drivers has also lowered over time going from paying 80% initially (20% commission to Lyft/Uber) to 75% nowadays (25% comm ission to Lyft/Uber). Table VI III presents the Lyft/Uber fares and commission rates appl icable t o this study. Table VI IV shows the total amount paid by passengers, driver earnings, and the actual Lyft and Uber commission, before and after tips. Earnings include prime and guarantee bonus per hour but does not include initial sign up bonuses. All mone tary values are in 2016 U.S. dollars. Table VI III Lyft/Uber Fares and Driver Commission

PAGE 55

43 Table VI IV Passenger Cost, Driver Earnings, and Actual Commission Gross Earnin gs The dataset shows that if only the time and distance drivers spent with a passenger (WP) is taken into account; Lyft/Uber drivers would be making around $40 per hour or $1.39 per mile. However, there is more to account for within the overal l work shift. After including all times and travel distances, gross earnings turn out to be $15.69 per ho ur or $0.82 per mile (Table VI V ).

PAGE 56

44 Table VI V Gross Earnings Dis aggregating by ridesourci ng company, I found differences between U ber and Lyft earnings (Table VI VI ), with tips playing an important role in the differences. The small amount of Uber tips was from a few passengers giving tips in cash since Uber does not facilitate tipping on its app. Table VI VI Gross Earnings Lyft compared to Uber

PAGE 57

45 Expenses There many variables and rates that go into calculating personal car expenses such as ownership costs (e.g. depreciation, finance charges license, insurance, registration & taxes) and operating costs (e.g. gas, maintenance, miscellaneous upkeep such as car washes and cleaning, mobile device and data fees, parking and traffic violations, and the risk of crash or injury). The expenses also d epend on the value of you r car, driving mileage, and whether or not you own a car already and/or have already paid for some of these expenses. To account for the broad range of possibilities, I characterize three differ ent expense scenarios (Table VI VII .) covering all types of drivers, from occasionally part time drivers to full time drivers. In the basic added cost, I assume a range of driving hours of 1 15 hrs/week and around 11,000 miles per year. The next scenario included most of the drivers with 16 4 9hrs/week and around 33,000 miles per year, and the last scenario is based on the U.S. Federal Standard Mileage Rate. The first cost scenario assumes that a driver already owns a car and has paid off basic ownership expenditures. Ridesourcing drivers are supposed to upgrade their car insurance to be properly insure d with ridesourcing but a few drivers probably do, risking that an insurance company would not pay a claim if a p erson was driving for Lyft/Uber For this first scenario, I assumed most ownership cost such as insurance as a sunk cost that drivers pay regardless of whether a person drives for a ridesourcing company or not; in other words, it is not considered an additional expense. This scenario also includes conservative values for depreciatio n, maintenance, and other miscellaneous expenses. The cost expense for this scenario is $0.28 per mile.

PAGE 58

46 The next scenario represents the majority of ridesourcing drivers (51% of drivers) based on Uber data published by Hall and Krueger (2015) Since drivers in this scenario experience higher timing and mileage, I included cost s associated with own ing a car and increase d the other values according to the mileage per year. I used assumptions based on AAA rates ( AAA, 2015 ) and other sources but still trend toward the conservative end of the expense spectrum. In this scenario, expenses equal to $0.40 per mile. In the third scenario, I used the 2016 U.S. standard mileage rate determined by the federal government of 54.0 cents pe r mile. The average mileage rate based on the previous three scenarios is calculated at $0.41 per mile. The corresponded cost per hour is based on the average of 19.1 mph from Table VI I Table VI VII Ridesourcing Expenses

PAGE 59

47 Net E arnin gs Ridesourcing drivers are probably excited to think they are making $40 per hour, or even $16/hr but would be disappointed to learn that, after accounting for expenses, the average hourly rate, including tips, is $7.94 (not even minimum wage in Colorado) as shown in Table VI VIII. This net earning wage could be even lower because of the higher commiss ion rate of 80% versus 75% of newer driver and relatively conservative expense estimates The net earning rate per mileage is between $0.28 and $0.54, with an average of numbers, if a driver work full time (40 hours a week, 50 weeks a year) driving over 40,000 miles a year, the annual net income would be around $16,000. These net nu mbers are all pre tax earnings. Table VI VIII Net Earnings (Gross Earnings minus Expenses) When disaggregating by ridesourcing company an d including tips, the Uber net earning rate is $7.18 per hour, and Lyft is $8.56/hr (Table VI IX) Tips makes a significant difference on living wages for Lyft d rivers with around $1.93/hr and accounting for a 29.1% increase of net earnings, while is only $0.33/hr (4.9%) for Uber.

PAGE 60

48 Table VI IX Net Earnings Lyft compared to Uber Chapter Conclusions The time efficiency rate without taking into account commuting at the end of the shift is 41.3%. This time efficiency rate is lower than the capacity utilization rate of 46 54% in a previous study ( Cramer & Krueger, 2016 ) Accounting for the commute at the end, the overall time efficiency rate drops to 39.3%, meaning that drivers spent more time without a passenger than with one in their car. The main implication for this result is the reduction on earnings per time ($/hour) since ridesourcing d rivers have to spend time waiting for a passenger request, traveling to a pick up destination, wait ing for the passenger once a t the pick up location, and commuting time at the end of the shift The efficiency rate in terms of WPMT versus total mileage without including commute distance is 65.4%. The mileage eff iciency rate for this study is higher than the 61.0% utilization rate calculated by Cramer and Krueger (2016) I attribute this difference to the research design ; wh ich minimized the cruising for a ride request, did not accepting rides when the di stance to pick up a passenger was too long, and used conservative commute

PAGE 61

49 distances at end of shifts. When including all distances, the mileage efficiency rate drops to 59.2%, but I believe the real mileage efficiency rate is even lower. Even with this conservative calculation, drivers have to travel 69 extra miles in deadheading for every 100 miles originally from WPMT. There has been a lot of uncertainty regarding how m uch money a Lyft/Uber driver makes. What is widely known i s the difference between what pas senger pay and what Lyft/Uber driver s are paid. The Lyft/Uber fare per mile is around $1, but when we take into account all fees and divided by the WP time, passenge rs pay around $1.86 per mile (Table VI 3 & Table VI 4 ) When I became a Lyft and Uber driver, I signed up with a commission rate of 80 20 (80% of fare for driver and 20% for Lyft/Uber ), which is used for this study (newer drivers get even lower rates at 75 % of fares). When the booking fee is taking into account, the commission rate before tips for all rides is 31.1%. When tips are taking into account and separated by company, the Uber commission is higher (32.1%) than the Lyft commission (27.3%), suggesting that drivers earn better driving for Lyft. This is the first study to incorporate commute times and distances into earning calculations, but even without accounting for the commute time of the driver to or from their home, the gross earnings drops to less than $16.50/hour. This is a conservative high number since the commission rate received was at the high end, I was driving for both Lyft and Uber minimizing waiting for a ride times, and I minimized unnecessary driving whenever possible. As a comparison, a recent Buzzfeed article reported that an Uber driver in Denver makes $16.89/hour ( O'Donovan & Singer Vine, 2016 ) but they overestimated this hourly earning with some of the assumptions used in calculating driver expenses.

PAGE 62

50 Using all the data from the 416 rides and including all times and distances, the gross earnings for this study equals to $15.69/hour, which might seems like a good hourly rate, but many drivers do not realize the expenses incurred by driving. The expenses varied from our conservative calculations using very basic added cost of $0.28 per mile to the standard 2016 mileage rate of $0.54 per mile, so in reality ridesourcing dr ivers make between $5.38 and $10.36 per hour, with an average of $7.94/hr before taxes. Uber net earnings before tips ($6.85/hr) is slightly higher that Lyft earnings ($6.63/hr) but completely change when tips are taken into account. Net earnings with tips included are $8.56/hr for Lyft versus $7.18/hr for Uber. Lyft tips in net earnings equals to a 29.1% increase and plays a critical component in the ridesourcing driving economy. Uber could easily add a tipping option in their app to allow passengers add a tip in their credit card bill if implement this option. Uber and Lyft depend on the driver partners labor market. They incentivize new drivers with bonuses and referrals, but their retention rate is not very good. According to Hall and Krueger (2015) 89 % of Uber driver partners stay active after one month, 70% after six months, and around 50% after a year. One of the reasons for this may be the realization of driving expenses and costs incurred by driving. For example, a taxi driver who makes $12/hour mi ght think that they can make a lot more driving for Uber ($40/hr or $16/hr). However, they may soon realize that, after accounting for expenses, it is not nearly as profitable as expected and are actually not even making minimum wage at about less than $8 /hour.

PAGE 63

51 Based on the results from this study, I have several recommendations to create more efficiency by reducing the amount of time and distances ridesourcing drivers have to travel and earn wages that are more decent Cities authorizing ridesourcing se rvices and c ompanies such as Lyft and Uber should: Suggest drivers to minimize the amount of miles they drive without a passenger Balance the driver network better by directing certain drivers to their closest prime rate zones instead of generalizing without specific guidance Balance the supply of drivers better, especially when passenger demand is not high. This would minimize VMT from drivers circulating around. A llow drivers to create ride zones so they do not end up far away from their desired loca tion. Also, expand the destination filter option so rides can be matched along certain routes or destinations and not just at the end of the shift (during the study period, Lyft and Uber started an option for drivers to put a destination filter but the opt ion has not been very effective). Lyft and Uber have stated that they want to reduce the inefficiency of empty seats as one of their desired goals so ridesourcing could function like a carpooling app where all drivers set their destination and find passeng ers along the way. N ot match drivers when the passengers pick up location is far from the driver location or compensate drivers for th ese scenarios ( I, as a driver, have seen requests from locations more than 30 minutes or 20 miles away). Concerning earnin gs, Lyft and Uber could always pay their independent contractors better by paying drivers on the service fee (which goes 100% to Lyft/Uber), increasing passenger fees, increasing the driver commission fee,

PAGE 64

52 providing better incentives, or covering some of t he expenses. Thus far, these companies seem to be moving in the other direction by increasing the Lyft/Uber cut from 20% to 25% or higher, lowering passenger rates (mileage and time), and increasing the service fee (which is not shared with the drivers). U ber also has not shown any desire to allow an option to tip in their app, which is the number one request from drivers. The main limitation to this study is the trip sample size and diversification of drivers. Drivers might have different work strategies such as searching for prime areas, have a desired location in mind, cruising unlimitedly until they get a ride request, or limiting driving without a passenger as much as possible by parking right after a passenger is dropped off. I minimized the distance traveled without a passenger for the results to be conservative. The study is also limited to the Denver metropolitan area so the Lyft/Uber costs and earnings are based on this area. This is the first independent study to use Lyft and Uber data exclusively to drivers. The results provide insight into the impacts of ridesourcing into travel times, travel distances, and the labor economy of Lyft/Uber independent contractors. This research starts to fill a gap in the academic literature by identify ing, measuring, and disentangling the im pacts of ridesourcing on very importa nt aspects of transportation. I hope this study helps cities and regional organizations better account for the impacts of ridesourcing on travel time and mileage efficiency, as we ll as inform the ridesourcing labor market on the complicated issues of earnings and expenses.

PAGE 65

53 VMT IMPACTS Most of my life I have lived in two cities: Cali in Colombia and Denver in the U.S. These cities differ quite dramatically in their economies, demog raphics, employment, culture, etc. Regarding transportation, they are also very different in terms of land use, transportation services offered, mode share, car ownership, work force, etc. For example, the mode share of private vehicles in Cali is 10% ( Cali Cmo Vamos, 2015 ) versus around 79% in Denver ( U.S. Census Bureau, 2015 ) Growing up in Colombia, I lived a completely different transportation experience than my current one. My parents owned one car to be shared by the five members of my family. My travel behavior was truly multimodal; I would take public transportation, carpool, walk, or bike. Occasionally, I would tak e a taxi to get around the city. Thinking back, one of the things that influenced me the most was the large amount of taxis and their effect in congestion. Still nowadays, Cali experiences many impacts from taxis circulating around (Figure VII. I ) and getti ng in line outside the airport, hospitals, malls, bus terminals, and other public places (Figure VII. II ) The taxi impact experienced is very clear since in Cali taxis are yellow and represent 7% of total mode share ( Cali Cmo Vamos, 2015 ) This suggests a 0.7 (around 7 to 10) relationship when taxi mode share is compared to private vehicles mode share. The estimate of mode share for taxis in Denver (combined with motorcycle or other non traditional means) is only 1% ( U.S. Census Bureau, 2015 ) representing a 0.01 (around 1 to 79) ratio of taxi versus private vehicles.

PAGE 66

54 Figure VII I Taxis in Cali, Colombia (Source: ElPais.com.co)

PAGE 67

55 Figure VII II Taxi Tracks in Cali, Colombia (Source: ElPais.com.co)

PAGE 68

56 This experience is relevant to ridesourcing since very little is known about the contribution that Lyft and Uber provides to the efficiency and impacts of our transportation systems. When p eople use private vehicles to operate for Lyft and Uber, we might not realize the impacts in city streets since most only carry a barely v isible logo sticker. Cities, regions, and transportation organizations usually set up goals to reduce congestion, environmental impacts, and equity issues. A general term used for h ousing strategies to aim for more efficient use of transportation resources is Transportation Demand Management (TDM). Some of these goals are in terms of measuring and tracking mode share, passenger miles traveled (PMT), and VMT. In an effort to contribute to the conversation, this study chapter aims to analyze the mode share r eplacement occurring with ridesourcing, measure the efficiency ratio of PMT/VMT and VMT/PMT, compare VMT before and after ridesourcing, and estimate the extra VMT generated in the U.S. from Lyft and Uber. Chapter Related L iterature Transportation organizat ions across the globe are trying to solve transportation problems by setting strategic goals to reduce SOV or increase the mode share of sustainable modes of transportation including transit, walking, and biking. A few reports and studies have shown that c ities have successfully met some of these goals through a variety of strategies, including TDM efforts such as congestion fees, tollways, high occupancy vehicle (HOV) lanes to carpool/increase vehicle occupancy, parking management, transit passes as we ll as infrastructure investments and policy changes ( Bialick, 2015b ; Henao et al., 2015 ; Kaffashi et al., 2016 ; Steele, 2010 )

PAGE 69

57 The problem with ridesourcing is that when organizations are trying to set up goals in regards to mode shift and prioriti zing certain modes, they do not know the real impacts and efficiency of services like Uber and Lyft. Shall organizations support ridesourcing and encourage its services to a higher use? What modes are they replacing? What is the PMT/VMT or PMT/VMT ratios c ompared to other modes? How would the transportation system benefit if ridesourcing was replacing modes that are more efficient? For example, we know that it will never be better than biking, walking, or transit since VMT for theses modes is zero or close to zero, but how does it compare to the SOV PMT/VMT ratio of 1.0, or taxis being around 0.40 ( Cramer & Krueger, 2016 ) ? The few studies that look into mode share changes and VMT impacts analyze the data at the aggregate level and do not make a distinction about the magnitude and directional shifts occurring within all modes. This study chapter aims to start filling this gap in the literat ure by looking in more detail the mode replacement, as well as PMT and VMT changes, and find out the place where ridesourcing stands in terms of efficiency compared to other modes of transportation. Chapter Data and Analysis For this research, I used the i nformation containing both the information collected by driving and the corresponding passenger survey information, for a total of 311 passenger surveys during 308 rides. The information gathered by driving is the same as the data collected in chapter VI, except the focus on this chapter is on distance and does not include times nor earnings. I also include information on the number of passenger for each ride. The

PAGE 70

58 trave III ). The survey response options to the multiple choice question were: Drive Alone Carpool (drive) Carpool (ride) Public transportation Bike or Walk Taxi Other the question carefully, or they use Lyft/Uber as their main mode of transportation and did not category.

PAGE 71

59 Figure VII III Mode Re placement (Q5) If the passenger response to question Q5 was carpool, the survey was designed to ask the number of people that the passenger would have carpooled with, with the intent to make a fair comparison (Q6). For this study, I also included the ques tion on whether or not the passenger was using Lyft/Uber for the entire length of the trip (origin to final destination), or he/she was making a connection to another mode of transportation (Q9), and which mode of transportation (Q10). Finally, I included the survey question about car ownership/access (Q19). In summary, the information of interest for each ride includes: Date of ride Time at request The service the ride was requested from: Lyft, LyftLine, UberX, or UberPool

PAGE 72

60 Travel distances Number of passengers Trip Mode replaced (Q5) Number of people carpooling if passenger would have carpooled (Q6) Connection with another mode of transportation (Q9 & Q10) Own or have access to a personal car (Q19). For the twenty nine passengers that would have carp ooled, the number of people that would have carpooled together is between two and four people with an average of 2.59 people per trip (Q6). Regarding connection with another mode of transportation, 94.50% of the passengers stated that they were using Lyft/ Uber for the entire trip, and only 5.50% were using another mode of transportation in connection with Lyft/Uber (Q9). Moreover, 187 people out of 291, or 64.3%, responded to question Q19 by stating that they own or have access to a personal car Base d on t he data previously described and the VMT for each mode replaced, I proceeded to calculate the Replaced VMT (or VMT BEFORE ) and passenger miles traveled (PMT) based on travel behavior prior to Lyft and Uber were in place, as follows: which is the same as WPMT for most rides

PAGE 73

61 is the case when someone else (e.g. parent or spouse) would have driven the passenger from origin to destination (O D) and go back to origin incurring in a round trip doubling the miles of the original O D trip. e same as ridesourcing VMT zero for walk to transit (WTT) and 3.4 miles for drive to transit (DTT). The selection of WTT and DTT rides were based on ride distance, answer to connection mode (Q9 & Q10), answer to car access (Q19), percentage of WTT and DTT based on data from a previous study in the Denver area ( Marshall & Henao, 2015 ) and DTT distance based on another paper in the study area ( Truong & Marshall, 2014 ) distance efficiency of around 40% ( Cramer & Krueger, 2016 ) I used the same ridesourcing VMT for trips to the airport. If the ride included a connection, the previous distance replaced is based o n total VMT & PMT. For example, if a passenger was dropped off at a transit station to Replaced is equal to two times (2x) the total distance (WPMT plus the train distance) becaus e the person taking the passenger would have travel all the way to the airport and back. The Ridesourcing VMT (or VMT AFTER ) was calculated using the same methodology as Chapter VI, where all distances with and without a passenger are taken into account

PAGE 74

62 Then, I calculated PMT/VMT and VMT/PMT ratios for before (VMT Replaced) and after (Ridesourcing VMT) to understand the efficiency of PMT/VMT, and how much VMT is put into the transportation system per PMT before and after ridesourcing. Finally, in order to understand the additional VMT put into the system because of ridesourcing, I calculated the ratio of VMT AFTER versus VMT BEFORE for every mode replaced and overall. Chapter Results The total Ridesourcing VMT in this analysis was 3,617.7 miles while VMT Replaced was 1,959.6 miles, and PMT was 2,200.0 miles. The average passenger travels a mean distance of 7.14 miles (median distance of 3.50 miles) ranging from 0.5 miles to 49.10 miles. See summary statistics in Table VII I for more details. Table VII I PMT, VMT Replaced, and Ridesourcing VMT PMT/VMT Efficiency The before travel behavior based on the replaced mode was 112.3% efficient in terms of how much PMT per VMT would have happened if Lyft or Uber were not available, meaning that

PAGE 75

63 all the modes replaced were transporting passengers at a rate of 112.3 miles fo r every 100 VMT. With the introduction of ridesourcing, the PMT/VMT efficiency dropped to 60.8%, meaning that the miles passengers travel is less than the vehicles miles at a rate of only 60.8 PMT for every 100 VMT from Lyft and Uber. This equates to a 51 .5% net change or 45.8 % percent reduction (Table VII II .). VMT/PMT Ratio Using the same numbers but inverting the numerator and dominator, I calculated the before and after VMT/PMT ratios. If Lyft/Uber were not available, the before travel behavior scenari o would have been 0.89 VMT per ever y 1 .0 PMT (Table VII III), meaning that all the modes replaced were transporting more passengers miles than vehicles miles. With the introduction of ridesourcing, the VMT AFTER /PMT ratio went up to a value of 1.64 meaning that now with Lyft and Uber the VMT is 1.64 miles for every 1 .0 PMT (Table VII III). This value is very similar to the calc ulation of 1.69 from chapter VI. VMT before and after The total ( d) and median distances of PMT, VMT Repla ced, and Ridesourcing VMT for each mode replaced and total are presented in Table VII III The last column of this table shows the VMT AFTER / VMT BEFORE ratio for every mode and total, with an overall ratio of 184.6%, meaning an overall increase of 84.6% in VMT.

PAGE 76

64 Table VII II PMT/VMT, before and after

PAGE 77

65 Table VII III VMT by Mode Replacement, before and after

PAGE 78

66 Chapter Conclusions This study chapter shows that, overall, ridesourcing VMT is approximately 1 84.6% of what would have been without Lyft/Uber which has significant implications for our cities in terms of congestion and environmental concerns. Ridesourcing provides more mobility 12.2 but affects the efficiency of transporting passengers versus vehicles going from a PMT/VMT efficiency of 112.3% to 60.8% (VMT/PMT ratio of 0.89 to 1.64). The worst VMT changes come from passengers that would have used public transportation or active transportation. The problem is that for these modes, the transportation system was previously experiencing zero or close to zero VMT, but now with ridesourcing we end up with some of the worst VMT AFTER / PMT ratios. The main explanatory reason is that these are short distance trips and ridesourcing is less efficient with these trips. For example as shown in Table VII III the median distance for an is 4.95, which equates to 2.6 times the PMT. The transportation system experiences better VMT efficiency when the since the efficiency rate of ridesourcing is better than taxis or people chauffeuring family or friends. In addition one of the things to notice on the connection tri ps to transit so the VMT AFTER is minimal compared to the total VMT BEFORE that a person would have done. For example, if a person r i de s a Lyft/Uber for 4 m iles to a transit station and then, ri de s the train for 20 miles ; the replaced VMT i s 48 miles ( ). In an attempt to have a more general idea of the impacts of Lyft and Uber, I look at some published data on the rides that Lyft and Uber have given so far. This summer, Uber

PAGE 79

67 reported that its second billion rides were completed during six months (the first billion was reached in six years) ( Somerville, 2016 ) While Uber operates globally, Lyft is only in the U.S., so I looked at numbers exclusively for the U.S. According to several news articles and based on a statement from Lyft co founder and president John Zimmer Lyft does around 17 million rides per month in the U.S., and compared to Uber, their market share of total is around 20% ( Buhr, 2016 ; Kokalitcheva, 2016 ) Using these numbers, Lyft and Uber gives around 1 billion rides p er year in the U.S. I f the results for this dissertation h e ld true for the entire country, the impacts of ridesourcing on VMT would be around 5.5 billion extra miles per year in the U.S. ; since the VMT BEFORE (or replaced) would have been 6.4 billion per year and VMT AFTER (or Ridesourcing) would be c lose to 12 billion miles per year from total Lyft/Uber VMT. See Table VII IV for calculations and more details. Lyft and Uber have changed travel behavior for many people and they are becoming more popular but cit ies and transportation professionals that have expressed concerns over the potential negative imp lications of ridesourcing have valid point s There should be better policies and regulations in place for ridesourcing companies to operate in a way that would allow our transportation system to function in a more efficient way, but it seems that the opposite is occurring as shown in this chapter study. Table VII IV Extra VMT per year in the U.S. due to Lyft/Uber

PAGE 80

68 PARKING IMPACTS This parking study is one component of a broader research strand that analyzes the impacts of ridesourcing on transportation. As se en in previous chapters, there are negative impacts as it relates to VMT since a ridesourcing driver travels additional miles well above the actual passenger O D trip. This becomes worse if the passenger is shifting from a mode other than SOV, such as carp ooling, transit, biking, or walking. Ridesourcing could also have positive impacts such as parking as more people can access destinations without requiring an accompanying parking space. In turn, this could facilitate reduced parking supply and help allevi ate congestion problems that might come from vehicles cruising around searching for parking. On the other hand, parking difficulty and cost is an important TDM strategy that decisions to use Lyft/Uber when parking supply is limited and/or expensive. In terms of travel behavior, ridesourcing could also serve as a transition in travel behavior change, car ownership, and long term modality styles as people shift from an auto orie nted lifestyle towards becoming more of a multimodal traveler. However, this has not yet been adequately studied. Therefore, this research aims to evaluate the impacts of ridesourcing on parking by looking at the trips that otherwise would have needed park ing (e.g. if passengers would have driven their own car, renting a vehicle, or carpooling) and investigate specific origins and destinations (e.g. airport, sporting events, special events, bars/night life, park and ride locations). This study also seek to shed light on understanding parking in terms of issues such as supply and price as a reason of someone using Lyft/Uber instead of driving.

PAGE 81

69 Chapter Data and Analysis For this study, I used the dataset of 311 samples containing both the information col lected by driving and the corresponding passenger surveys. The information of interest for this section is on the origins and destinations, trip purposes, and parking related questions including driving mode replacement and parking as a stated reason to us e ridesourcing instead of another mode. I analyzed the dataset for both the specific trip and for general travel behavior questions. Chapter Results I divided the results section in three subsections to explore: Parking demand by exploring the trips where ridesourcing replaced driving Locations origin & destination (O D Matrix) with trip purpose, and connection with other modes (e.g. park and rides) Parking difficulty as a reason to choose ridesourcing over other modes. Each subsection includes results from the specific Lyft/Uber ride as well as answers from the general travel behavior survey questions. Parking Demand Specific Trip Since parking, theoretically, is only needed for driving trips; I decided to look in more detail at the mode replacement distribution and pay closer attention to the trips that would have involved driving such as drive alone, carpool (drive), car rental, and carsharing to start understanding potential changes in parking demand.

PAGE 82

70 In terms of replacing driving trips with ri desourcing, we should exercise some caution since in some cases, the trip replaced might have been only a part of the trip with the intent to avoid parking at the final destination. In other words, a passenger might have still driven and parked but ridesou rcing allowed him/her to do so in a different location. For example, parking downtown might be limited and expensive so a passenger decide s to drive to a location as close as possible to the destination where parking is more abundant and/or cheaper. Th en, she/he requested a Lyft/Uber ride to reach the final destination, thus benefiting from cost and time savings of a shorter ridesourcing trip. This has both positive and negative implications for transportation. Within this section, I analyzed the specif ic trip driving is not part of the connection trip (Q9 & Q10). Figure VIII. I illustrates the percentage of respondents that otherwise would have driven and needed a parking location, with a result of 26.4% (82 O D trips) of all respondents.

PAGE 83

71 Figure VIII I Ridesourcing Replaci ng Driving Trips (Parking) Gen era l Use In the general questions section of the survey, I asked passengers about travel about II provides the results with a third of participants 14.3% plus 19.9% stating that they drive a lot less, which has implications for reduction in parking demand and slightly higher than the percentage of respondents to the specific trip replacement question. It was not expected that passengers would increase their driving, but seven passengers (2. this might be explained under two scenarios. The first one is that passengers share a vehicle with others in their household; th e other household drivers begin relying more on

PAGE 84

72 ridesourcing, which facilitates greater vehicle availability and more driving for the respondent. The second explanation is that passengers are also ridesourcing drivers, which turned out to be the case for t hree of the survey respondents. Figure VIII II Travel Behavior Change (Driving today compared to the past) Locations, Trip Purpose, and Connectivity to Transit Stations Spec ifi c Trip Data in Chapter V i n cludes the O D matrix (Table V I ) for all the rides in the study that contains both the driver and passenger datasets. In order to understand the trips that influence parking those that otherwise would have driven as described in the previous subsection I created an O D matrix (Table VIII I ) exclusively for the Lyft/Uber rides that replaced driving showing the locations that could potentially reduce parking supply. The ut

PAGE 85

73 14 trips most of Table VIII I O D Matrix (Driving Trips Replaced) General Use In order to understand travel behavior changes in general, I created Figure VIII. III showing the distribution of trip purpose for all respondents, and for the 101 passenge rs that figure shows, social trips (i.e. go out) is the number one trip purpose for all respondents as well as those ridesourcing users that are driving less.

PAGE 86

74 Figure VIII III Connec tivity to Transit Stations Ridesourcing advocates state that these services are solving many of the issues with connecting to public t ransportation (i.e. the last mile). In order to explore this topic further, I recorded trips with transit stop locations as the origin or destination. The passenger survey also included questions in regards to connectivity Q9 & Q10 as described in Chapte r VII transportation in addition to Lyft/Uber for a single trip? For example, using Lyft/Uber transportation in transportation you have used in addition to Lyft/Uber for a single trip: Table VIII II includes the results for the specific trip and general use whenever passengers were replacing driving modes or were connecting to bus or rail at a transit station.

PAGE 87

75 Table VIII II Connectivity to Transit Stations Parking as a stated reason to choose Ridesourcing This subsection analyzes parking as a reason for someone to use ridesourcing over other modes, both for the specific trips and for general use. Specific Trip Passengers stated the main reason that led them to choose Lyft/Uber over other options (Q8) for the ac tual O D ride. Figure VIII. IV include s the percentage for each reason including for all passengers reasons to choose ridesourcing.

PAGE 88

76 General Use Figure VIII IV Main reason for choosing Lyft/Uber for Actual Ride

PAGE 89

77 Similar to the previous subsection, question Q17 ask the main reasons to choose ridesourcing over other modes, allowing respondents to check up to three reasons. In addition to this question, I used the driving frequency question Q21 to create Figure VIII. V I selected the top five reasons from the dataset per driving frequency. In this graph, I highlighted parking again to show the clear relationship between high frequency of driving and parking. Parking is the second most important reason for frequent drivers to choose ridesourcing. Figure VIII V Driving Frequency and Trip Purpose

PAGE 90

78 Chapter Conclusions With evolving transportation services and the promise of autonomous vehicles, changes in the transportation industry are creating a window of opportunity to help dissolve the car dependency and automobile focus of transportation infrastructure in the U.S. and around the globe. Thus, this study aims at investigating the reciprocal influence of an evolving transportation service ridesourcing in terms of a very im portant transportation infrastructure parking so we can model, design, and build future transportation i nfrastructure. I conducted a cross analysis to compare sp ecific Lyft/Uber rides and general travel behavior for three inter related parking topics. First, I looked at mode replacement finding that a third of all Lyft/Uber rides were replacing trips that would have needed parking. Then, I looked in more detail at the origin and destination location s travel behavior change s, t rip purpose, and connectiv ity to the transit system. The results show that passengers that usually drive are using ridesourcing mostly Thus, parking capacity could be reduced at these types of locat ions (e.g. airport, bars, and stadiums ). With respect to connection to the transit system, v ery few Lyft/Uber rides were used with this purpose in mind. Finally, I analyzed parking as a stated reason for using ridesourci ng, showing that for the specific trip, parking difficulty/cost is the fourth most cited reason for all trips and the second most cited reason for trips replacing driving. For general use, the passengers that the second most cited reason why they chose ridesourcing over other modes

PAGE 91

79 The results from this chapter suggest that thanks to ridesourcing we could easily decrease the parking supply because of two reaso ns: ridesourcing is replac ing driving trip s and parking is one of the main reasons people choose not to drive in the first place Thus, parking can be used as a TDM tool to influence travel behavior The less parking provided, the less driving desired, esp ecially when people have choices to shift t o other modes of transportation Continuation of this positive cycle would facilitate decreasing car dependency. In the future, autonomous ridesourcing vehicles would enable us to continue this trend more rapidly. In preparation for this drastic change, we should adapt the current transportation infrastructure to reduce parking stock and make room for more desirable land uses, especially in urban areas where land is costly. For example, we could eliminate parking m inimums in the generation rate manual so developers build less parking spaces and more building amenities They should offer residents different transportation options and plan to re accommodate parking spaces for future use. Similarly, common destinations such as the airport or hot spot ar eas for social activity (e.g. restaurants, bars, stadiums, and nightlife ) shall remove parkin g spaces or design future buildings with less parking capacity In turn, this could facilitate more livable and desirable places to walk, bike, and take public transportation.

PAGE 92

80 TRAVEL BEHAVIOR CHANGES Ideally, mode choice models are run and tested with robust data containing information regarding travel activities (e.g. mode of transportation, frequency, time of days, etc.) and specific individual characteristics (e.g. socio demographics, transportation resources, land use, travel behavior style, etc.) from a random population sample. While I do not have the necessary data for a traditional mode choice mo del, I can investigate the most probable mode replacement for each individual, and explore the variables and reasons influencing travel behavior change from previous modes of transportation after the introduction and evolution of ridesourcing. Thus, this s tudy chapter has three main objectives: 1. Collect the data necessary for integrating ridesourcing services into a research framework to help current travel models move past their simplistic focus on traditional modes of transportation (i.e. car, transit, wal k, and bike); 2. Use ridesourcing as a shock into the system to evaluate changes in travel behavior as well as relationships between travel behavior, trip purpose, and stated reasons for mode replacement; and 3. Identify and analyze components from the framework such as modality styles. a certain travel mode or set of travel modes that an individual habitually uses Vij, Carrel, and Walker (2013) The concept is derived from empirical studies on higher short term decisions and long term choices, which counters th e conventional assumption that people choose a mode independently for every trip.

PAGE 93

81 Conceptually, this dissertation builds upon the existing body of literature on the influence of ridesourcing on modality style, mode replacement, and travel behavior. Methodo logically, this work contributes to the understanding of modality styles, changes to modality styles with the introduction of additional transportation options ridesourcing in this case and the influence of modality styles on travel behavior. More gene rally, this dissertation contributes to the body of knowledge considering transportation planning and future travel demand models. Ultimately, this dissertation sheds light on the travel behavior impacts of ridesourcing with the current trends we are exper iencing in our transportation systems to better inform urban planners, policy makers, and transportation engineers. Chapter Literature Review Understanding travel behavior and the related decision making is a very complex area of study. One of the most com mon ways to rationalize and forecast travel behavior and decisions is through travel demand models. One of the issues with travel demand models is that the outcomes are dependent on the assumptions used in the model, and traditionally, travel demand models assume that an individual is fully aware of the range of transportation options so that the choice is based on utility maximization theory. Utility maximization derived from economic theory assigns a utility value for each transportation alternative and a ssumes that the mode with the highest utility value will be chosen. While this theory is well established in economics, it contains some limitations. Calculating the actual utility of any good is very complex and typically contains several attributes that are not realistic to measure in many scenarios such as inhibited values, attitudes, perception, and beliefs that could relate to ingrained lifestyles and deeply established habits for certain modes. Individuals adopt different patterns of consumption behav ior, not only based on utilitarian needs, but also

PAGE 94

82 because they express self identity, leading to a person making a choice as a result of how to act and who to be ( Giddens, 1991 ) Transportation researchers have studied links between lifestyles and travel behavior since early 1980s by classifying groups that depend on the time spent at different activities, weekly travel patterns, or travel expenditures ( Kitamura, 2009 ; Pas, 1 988 ; I. Salomon & Ben Akiva, 1983 ) More recently, the literature evaluated the effects on transportation choice by ( Handy, Cao, & Mokhtarian, 2005 ; Kitamura, Mokhtarian, & Laidet, 1997 ; Krizek & Waddell, 2002 ) and mobility resources (e.g. automobile ownership, bicycle ownership, public tr ansportation pass) ( Choo & Mokhtarian, 2004 ; Vredin Johansson, Heldt, & Johansson, 2006 ) Speci fic to carsharing, recent studies suggest that carsharing members have different mobility resources compared to the representative population, with fewer cars per households and higher levels of bike ownership and public transportation passes. In general, carsharing seems to broaden the mix of transportation modes and increase the distribution of transportation modes, including intermodality (mixed modes) and multimodality ( Johanna Kopp et al., 2015 ; Le Vine, Lee Gosselin, et al., 2014 ) In an attempt to capture modality styles and their influence on mode choice decisions, Vij et al. (2013) developed a model framework within the context of travel demand models using the Latent Class Choice Model (LCCMs ). LCCM models were first developed in the field of marketing sciences ( Kamakura & Russell, 1989 ) and include two components: a class membership model; and a clas s specific choice model. LCCM is seen as a solution to the black box of correlation structure from the continuous mixture distribution in most widely known travel demand models ( Walker & Ben Akiva, 20 11 )

PAGE 95

83 This dissertation uses the framework by Vij et al. (2013) as a foundation to incorporate ridesourcing into mo dality styles and mode choice modeling. The proposed framework facilitates the analysis of the three objectives in this part of the dissertation by regarding trav el behavior in general and socio demographic characteristics. Utilization of the full dataset is needed to disentangle the effects of using ridesourcing on travel behavior by controlling for factors such as modality resources and modality styles auto ori ented (i.e. drivers), multimodal, or non drivers of passengers prior to start using ridesourcing services. Chapter Data and Analysis For this study, I used the dataset containing all 311 passenger surveys same as Chapters VII and VIII and adapted the previously described travel demand framework (Figure IX.I) to use in this dissertation. Figure IX I Travel Demand Framework to Study Ridesourcing

PAGE 96

84 The framework shows that for a given origin destination pair and time, there is a discarded options by the individual. Indiv iduals discard certain transportation options due to perceptions, values, habits, preferences, experience, or because they might not be aware of which is the evaluation of each travel mode contained in the subset. The person then chooses a travel mode from the subset based on the utility evaluation. This is done for every given origin destination pair and time; so every person has a group of indivi dual subsets given the situation. The group of individual subsets is analogous to what Vij et al. (2013) define as mo dality style in their model framework. In this dissertation, I identified four different classes of modality styles: Driver or car oriented: if the individual stated driving as the main mode of transportation Multimodal: if individuals use several modes of transportation Non driver or multimodal without car: if the individual rarely, if ever, drives Bi style: if the individual assumes a combination of any of the three previous modality styles. To my knowledge, this dissertation is the first study to introdu ce a bi style modality classification, which hypothesizes that an individual may adopt two completely different modality styles according to several factors and attributes of travel. For example, a person could normally act as a multimodal traveler, but fo r trips transporting a child, the person

PAGE 97

85 would only consider the car. In this case, the person is bi style (first: multimodal; then: car oriented). Similarly, a person that normally only considers the car as the mode of transportation might behave differen tly for leisure trips (e.g. going out to eat or drink). This style (first: car oriented; then: multimodal without car). Previous studies on carsharing systems provide indirect insights into the bi modality style topic, with th e distribution of journey purposes per mode showing that individuals use personal cars, taxi, and carsharing differently depending on the type of journey ( Le Vine, Lee Gosselin, et al., 2014 ) Another insight is the potential effect of ridesourcing on drunk driving crash reduction ( Jones, 2015 ) which implies that people may be using these services more often for leisure trips but not necessarily for other purposes. The bi modality style classification is an important distinction because it allows travel demand modes to explain travel behavior better. Accordingly, we could create better policies and more effective pla nning to influence modal shifts away from SOV. The modality style classification is sensitive to the group of individual subsets; therefore, I established time limits to determine modality style classifications for a specific time period. Once a period is determined, I measured changes in modality style due to specific inputs. An individual could potentially change modality style with the introduction of several factors: i) changes in characteristics of the individual and households (e.g. income or expendit ures, employment status, marital status, parenthood status); ii) changes in location (e.g. home, work, or school); iii) changes in infrastructure or services (e.g. improvement on certain mode infrastructure, evolving transportation services); or iv) change s in mobility resources (e.g. personal cars, bicycles, public transportation pass, carsharing membership, or ridesourcing membership). For this dissertation, I surveyed passengers once but the survey

PAGE 98

86 included questions specific to the current trip as well as more general questions regarding travel behavior changes due to the introduction of ridesourcing. Establishing travel behavior prior to the introduction of Lyft/Uber and controlling for explanatory factors (e.g. the person recently moved from out of tow n, the passenger cannot drive anymore due to DUI or illness) is key to understanding travel behavior. The key questions for data analysis in this chapter are: Mode frequency (Q21), where passengers stated how often they use each mode of transportation. The multiple choice options were never, rarely (special occasions, monthly), sometimes (~ 1 2 times per week), regularly (at least 3 times per week), and always. the sentence based the following subsections sentences: o o o o Because of ridesourcing, I b o Because of ridesourcing, I take taxis

PAGE 99

87 Besides questions on mode frequency and travel behavior changes, I reevaluated several other questions such as trip purpose (Q13) and stated reasons (Q17) from the previous chapters. emails for a possible follow up study, up study on ridesourcing travel changes. If Chapter Results I organized this chap ter result section in three sub sections. The first one presents mode frequency and travel behavior changes. The second part investigates relationships between frequency of use, trip purpose, and stated reasons. The last subsection analyzes modality styles Mode Frequency and Travel Behavior Changes Figure IX.II presents frequency of use per mode of transportation. About half of of passengers use taxi always or r egularly. Carpool, public transportation, and bike/walk modes have a more evenly frequency of use distribution.

PAGE 100

88 Figure IX II Mode Frequency I collected information on travel behavior changes for five d ifferent categories new trips, drive, public transportation, bike/walk, and taxis which allow me to measure and compare the magnitude of change for each mode. For example, Figure IX.III shows driving change versus public transportation change.

PAGE 101

89 Figure IX III Travel Behavior Changes, Driving & Public Transportation The three yellow circles in Figure IX.III have the same magnitude of change for driving and public transportation. The triangle coverin g the area in quadrant IV is labeled as less sustainable more sustainable e occurs (blue zone). For example, a positive change occurred when public transportation use increased and driving use decreased due to ridesourcing. The

PAGE 102

90 figure shows that the majority of users did not change travel behavior, and there are more negative ch anges (red) than positive changes (blue). Relationships between Drive Frequency and Other Variables Drive Frequency and Trip Purpose There are some clear relationships between drive frequency and trip purpose. Passengers that drive the most regularly and always are using ridesourcing to go out for social trips, going to the airport, and/or when they are out of town (Figure IX.IV). In contrast, passengers that never drive are using ridesourcing to go to work and/or school. Figure IX IV Driving Frequency and Trip Purpose

PAGE 103

91 Driving Frequency and Stated Reasons I first explored stated reason on why passengers use ridesourcing in chapter VIII on parking. Further analysis on the topic shows that car dependents are replacing driving trips mainly because they are consuming alcohol and/or because they find parking difficult and/or expensive. On the other hand, non drivers use ridesourcing because they do not have a car available, public transportation is not great, and/or time constraints (e.g. passengers are in a rush to get somewhere), as shown in Figure IX.V. Figure IX V Driving Frequency and Stated Reasons

PAGE 104

92 Modality Styles Based on use frequency for different tr ansportation modes (Q21), I was able to identify three modality style classes 111 drivers, 64 multimodal passengers, and 117 non drivers as the research methodology suggested (Figure IX.VI). Then, I explored the the introduction of ridesourcing (Q11), frequency of ridesourcing use (Q14), and travel behavior change due to ridesourcing (Q15). I was able to identify a fourth modality style classification bi style comprised of typical drivers that use ridesourcing only for rarely or special occasions such as going out to eat or drink and/or when out of town. Table IX.I presents summary results for questions Q11, Q14, and Q15, showing the bi modality group comprised of 49 passengers.

PAGE 105

93 Figure IX VI Modality Style Classification

PAGE 106

94 Table IX I Bi modality Style Classification Chapter Conclusions In this study chapter, I identified mode use profiles of ridesourcing passengers, showing that most passengers investigated the magnitude of change for driving and public transportation modes to dete rmine if the direction in which travel behavior is occurring is more or less sustainable The data shows that most travel behavior stays unchanged, but for those changing modes, more people are replacing public transportation than driving (i.e. the magnitu de of change for public transportation is more negative than the magnitude of change for driving). With regards to trip purpose, this study finds that most drivers use ridesourcing to go out for social trips, going to/from the airport, and/or when out of t own with the main reasons being

PAGE 107

95 drivers are using ridesourcing rush). Finally, I conclude this chapter study identifying four modality style groups. The existing literature identified three classifications of ridesourcing passengers: drivers, multimodals, and non drivers. With further analysis, I identified a fourth class, the bi modality style, including those that vary between two or more different modality systems depending on the type of trip.

PAGE 108

96 OVERALL RESULTS The overall result section fo r this dissertation is divided in four main parts: results from the driver dataset, VMT, parking, and travel behavior. Driver Dataset Ridesourcing Times and Distances The overall time efficiency rate for this study, accounting for commute time at the end o f the shift, is 39.3%. The efficiency rate in terms of WPMT versus total mileage is 59.2%. The overall efficiency rate could be even lower since, by research design, I minimized the cruising for a ride request, did not accept rides required long travel di stances for passenger pick up, and used conservative commute distances at the end of shifts. Even with this conservative rate, drivers have to travel 69 extra miles in deadheading for every 100 miles originally from WPMT. Ridesourcing Earnings Including al l times and distances, the gross earnings for this study equals to $15.69/hour, which might seems like a good hourly rate, but this does not include expenses. I estimated expenses to be between $0.28 per mile to the U.S. Federal Standard 2016 mileage rate of $0.54 per mile, so in reality, ridesourcing drivers make between $5.38 and $10.36 per hour, with an average of $7.94/hr (including tips). Uber net earnings is slightly higher that Lyft earnings but completely change when tips are taking into account. Ly ft tips in net earning equates to a 29.1% increase. Uber could

PAGE 109

97 easily add a tipping option on their app to allow passengers add a tip in their credit card bill if imp lement this option. VMT from a PMT/VMT efficiency of 112.3% to 60.8%. Overall, ri desourcing increases VMT by 185%, which has significant implications for our cities in terms of congestion and environmental concerns. If the results for this dissertation held true for the entire country, the VMT impact of ridesourcing would be around 5.5 billion extra miles per year in the U.S We need more empirical studies that would help understand the magnitude of VMT impact. Parking Ridesourcing allows parking supply to decrease. At the same time, many passengers stated that parking is one of the main reasons for passengers with high drive frequency to use ridesourcing instead of driving. Continuing with this cycle of reducing parking supply will help decrease car dependency. Travel Behavior Based on modality style, I identified four groups of moda lity styles for ridesourcing users: drivers, multimodals, non drivers, and bi modal style based on trip purpose. For typical drivers, ridesourcing is mostly replacing social trips (e.g. go out), to/from airport, and when out of town. Drivers stated drinkin g/avoid driving and parking as the main

PAGE 110

98 reasons not to drive. For typical non drivers, ridesourcing is replacing work and school trips with the main reason being that public transportation is not available.

PAGE 111

99 SUMMARY CONCLU SIONS AND FUTURE WORK Ridesourcing has quickly become a very popular service that is successfully competing and interacting with other modes of transportation. This dissertation had several objectives in mind. First, this dissertation established a data collection methodol ogy and gathered ridesourcing data to investigate ridesourcing impacts. Then, the dissertation measured ridesourcing efficiency in terms of times and distances (VMT and PMT), and evaluated VMT impacts of ridesourcing by analyzing before and after scenarios After investigating the labor market economy for ridesourcing drivers, I investigate d parking impacts and opportunities to calibrate parking generation rates and reduce car dependency. Finally, this dissertation explore d travel behavior and its implicati ons on future travel demand models. I hope this research helps cities and transportation organizations better understand the impacts of ridesourcing on several aspects of transportation. For example, transit agencies that are contemplating removal of bus s ervices in certain areas can use these results in terms of ridership and PMT/VMT efficiency to help them look at the issue in more detail. This will help inform them as to what they gain or lose in terms of replacing or connecting transit with ridesou rcing. As this dissertation shows, transportation officials also need to be cautious regarding the positive and negative outcomes of ridesourcing. One of the most important results from this dissertation is the realization that ridesourcing in the current form is only more efficient in terms of transportation passengers with VMT than two other modes, taxis and if getting a ride. We might think that shifting all drivers to ridesourcing would be a positive change, but in reality, that might not be the case. F or example, a driver needing to go

PAGE 112

100 five miles to his/her destination would put in the transportation system five miles (5 VMT); if that same person is taking a Lyft or Uber instead, he/she would be still move five miles, but the ridesourcing driver would b e adding approximately nine miles (9 VMT) to the transportation system. This is even worse if walking or bicycling are the replaced modes since such modes do not add VMT to the transportation system. Ridesourcing efficiency is even worse with shorter trips since a driver might be circulating and driving a few miles for example 3 or 4 miles only to be transporting a passenger for a shorter distance for example 1 or 2 miles The same occurs with public transportation and carpooling, modes that are mor e efficient than ridesourcing. Looking at the positive aspects, ridesourcing provides a positive utility to many costumers by increasing their mobility and/or ease of mind when it comes to avoid drinking and driving and resolving parking. Concerning parkin g, ridesourcing decreases the need for parking supply. At the same time, many passengers stated that parking is one of the main reasons to use ridesourcing instead of driving. Continuing this positive cycle would help decrease car dependency. Potential pos itive effects into reducing VMT and increasing ridesourcing efficiency may come with newer services such as LyftLine and UberPool, which are true ridesharing platforms; thus far, these services has not been proven to be effective. Passengers that use these services are usually less car dependent and multimodal compared to the general population. New members willing to try these service might be only doing so for two reasons; one, that the ride is way below the regular cost; and two, that the probability of finding a match during the ride is low. Academic research has mathematically proven using algorithms that on demand high capacity ride sharing is doable and provides positive effects

PAGE 113

101 in terms of efficiency ( Alonso Mora, Samaranayake, Wallar, Frazzoli, & Rus, 2017 ) but we need more research on travel behavior and social sciences to determine the willingness to share in our cities. This conversation will become more relevant with autonomous vehicles coming in the near future. The other issue with LyftLine and UberPool is that drivers make even lo wer wages with these services since they spent more time picking up and dropping off several customers. This study does not come without limitations. The main limitation is the trip sample size relative to the overall number of rides these companies provid e. Doing the data collection by myself is also both a limitation and an advantage. It is a limitation because I am not taking into account different driver strategies but an advantage since I was able to control the amount of driving and design the researc h with conservative estimations. However, we need more empirical studies like this one to gather independent datasets and help narrow down the numbers calculated in this dissertation At the same time, the transportat ion sector should demand that ridesourc ing companies share data in order to operate in our cities. I plan to continue doing ridesourcing research as it relates to travel demand models. Modelers need these types of data to include newer services in their travel demand plans, as well as calibrate better inputs for newer services such as autonomous vehicles. This research starts to fill several gaps in the literature regarding ridesourcing services. My ultimate goal is to help cities and transportation organizations better account for the impacts o f technology and evolving transportation services in their policies, planning, and engineering processes. I would also hope to contribute to the conversation on how ridesourcing companies can help better achieve sustainable transportation goals including

PAGE 114

102 m ore VMT efficiency, better interconnectivity and integration with active modes of transportation, equity, and safety for both users and drivers.

PAGE 115

103 REFERENCES AAA. (2015). Your Driving Costs: How much are you really paying to drive? http://exchange.aaa.com/wp content/uploads/2015/04/Your Driving Costs 2015.pdf Alonso Mora, J., Samaranayake, S., Wallar, A., Frazzoli, E., & Rus, D. (2017). On demand high capacity ride sharing via dynamic trip vehicle assignment. Proceedings of the Natio nal Academy of Sciences doi: 10.1073/pnas.1611675114 profit ridesharing, driver strategies, and VMT. Transportation, 41 (5), 1099 1117. doi: 10.1007/s11116 014 9531 8 elease Data to Shed Light on How They Affect Traffic. StreetsBlogSF Bialick, A. (2015b). New Data Shows Most Trips in SF Are Made Without a Private Automobile. StreetsBlobSF Retrieved from http://sf.streetsblog.org/2015/02/03/new data shows most trips in sf are made without a private automobile/ Bialik, C., Flowers, A., Fischer Outer Boroughs More Than Taxis Are. FiveThirtyEight Retri eved from FiveThirtyEight website: https://fivethirtyeight.com/features/uber is serving new yorks outer boroughs more than taxis are/ Buhr, S. (2016). Lyft picked up 17 million rides in October, up more than twice that amount from last year. Techcrunch Ca li Cmo Vamos. (2015). Encuesta de Percepcin Ciudadana Movilidad. Cervero, R. (2003). City CarShare: First Year Travel Demand Impacts. Transportation Research Record: Journal of the Transportation Research Board, 1839 159 166. doi: doi:10.3141/1839 18 Cervero, R., Golub, A., & Nee, B. (2007). City CarShare: Longer Term Travel Demand and Car Ownership Impacts. Transportation Research Record: Journal of the Transportation Research Board, 1992 70 80. doi: doi:10.3141/1992 09 Cervero, R., & Tsai, Y. (2004) City CarShare in San Francisco, California: Second Year Travel Demand and Car Ownership Impacts. Transportation Research Record: Journal of the Transportation Research Board, 1887 117 127. doi: doi:10.3141/1887 14

PAGE 116

104 Choo, S., & Mokhtarian, P. L. (2004). What type of vehicle do people drive? The role of attitude and lifestyle in influencing vehicle type choice. Transportation Research Part A: Policy and Practice, 38 (3), 201 222. doi: http://dx.doi.org/10.1016/j.tra.2003.10.005 Cramer, J., & Krueger, A. B. (2016). Disruptive change in the taxi business: The case of Uber. The American Economic Review, 106 (5), 177 182. DuPuis, N., Martin, C., & Rainwater, B. (2015). City of the Future: Technology & Mobility: National League of Cities. Firnkorn, J. (2012). Tr iangulation of two methods measuring the impacts of a free floating carsharing system in Germany. Transportation Research Part A: Policy and Practice, 46 (10), 1654 1672. doi: http://dx.doi.org/10.1016/j.tra.2012.08.003 Fischer Baum, R., & Bialik, C. (2015) Uber Is Taking Millions Of Manhattan Rides Away From Taxis. FiveThirtyEight Retrieved from FiveThirtyEight website: http://fivethirtyeight.com/features/uber is taking millions of manhattan rides away from taxis/ Flegenheimer, M., & Fitzsimmons, E. (2015 ). City Hall and Uber Clash in Struggle Over New York Streets. The New York Times Retrieved from http://www.nytimes.com/2015/07/17/nyregion/city hall and uber clash in struggle over new york streets.html Giddens, A. (1991). Modernity and self identity: Se lf and society in the late modern age : Stanford University Press. Mobility Lab website: http://mobilitylab.org/2014/04/17/redefining uber why the term rideshare doesnt f it/ Grabar, H. (2016). "They Can Just Take an Uber". Slate Graham Rowe, E., Skippon, S., Gardner, B., & Abraham, C. (2011). Can we reduce car use and, if so, how? A review of available evidence. Transportation Research Part A: Policy and Practice, 45 (5), 401 418. doi: http://dx.doi.org/10.1016/j.tra.2011.02.001 Partners in the United States. Princeton University Industrial Relations Section Working Paper, 587 Handy, S ., Cao, X., & Mokhtarian, P. (2005). Correlation or causality between the built environment and travel behavior? Evidence from Northern California. Transportation Research Part D: Transport and Environment, 10 (6), 427 444. doi: http://dx.doi.org/10.1016/j. trd.2005.05.002

PAGE 117

105 Henao, A., & Marshall, W. (2017). A Framework for Understanding the Impacts of Ridesourcing on Transportation. In G. Meyer & S. Shaheen (Eds.), Disrupting Mobility: Impacts of Sharing Economy and Innovative Transportation on Cities (pp. 197 209). Cham: Springer International Publishing. Henao, A., & Marshall, W. E. (in press). Lyft/Uber Driver Perspective: Travel Distances, Times, and Earnings Transportation Research Board, 2016 Henao, A., Piatkowski, D., Luckey, K. S., Nordback, K., Marsha ll, W. E., & Krizek, K. J. (2015). Sustainable transportation infrastructure investments and mode share changes: A 20 year background of Boulder, Colorado. Transport Policy, 37 64 71. doi: http://dx.doi.org/10.1016/j.tranpol.2014.09.012 Jones, R. G. (2015 ). Uber Can't Really Prove That It's Caused A Reduction In Drunk Driving Accidents. Huffington Post Retrieved from Huffington Post website: http://www.huffingtonpost.com/2015/02/04/uber drunk driving study_n_6612106.html Kaffashi, S., Shamsudin, M. N., Cl ark, M. S., Sidique, S. F., Bazrbachi, A., Radam, A., Rahim, K. A. (2016). Are Malaysians eager to use their cars less? Forecasting mode choice behaviors under new policies. Land Use Policy, 56 274 290. doi: http://dx.doi.org/10.1016/j.landusepol.20 16.05.005 Kamakura, W. A., & Russell, G. J. (1989). A Probabilistic Choice Model for Market Segmentation and Elasticity Structure. Journal of Marketing Research, 26 (4), 379 390. doi: 10.2307/3172759 Kitamura, R. (2009). Life style and travel demand. Transportation, 36 (6), 679 710. doi: 10.1007/s11116 009 9244 6 Kitamura, R., Mokhtarian, P., & Laidet, L. (1997). A micro analysis of land use and travel in five neighborhoods in the San Francisco Bay Area. Transportation, 24 (2), 125 158. doi: 10.1023/A:1017959825565 Kokalitcheva, K. (2016). Lyft Will Complete 17 Million U.S. Rides This Month. Fortune Retrieved from http://fortune.com/2016/10/25/lyft 17 million rides/ Kopp, J., Gerike, R., & Axhausen, K. (2013). Status quo and perspectives for car sharing systems: the example of DriveNow. Strategies for Sustainable Mobilities: Opportunities and Challenges 207 226. Kopp, J., Gerike, R., & Axhausen, K. (2015). Do sharing people behave differently? An empirical evaluation of the distinctive mobil ity patterns of free floating car sharing members. Transportation, 42 (3), 449 469. doi: 10.1007/s11116 015 9606 1

PAGE 118

106 Krizek, K., & Waddell, P. (2002). Analysis of Lifestyle Choices: Neighborhood Type, Travel Patterns, and Activity Participation. Transportati on Research Record: Journal of the Transportation Research Board, 1807 119 128. doi: doi:10.3141/1807 15 Lane, C. (2005). PhillyCarShare: First Year Social and Mobility Impacts of Carsharing in Philadelphia, Pennsylvania. Transportation Research Record: J ournal of the Transportation Research Board, 1927 158 166. doi: doi:10.3141/1927 18 Lawler, R. (2014). Lyft TechCrunch Retrieved from https://techcrunch.com/2014/08/29/6000 words about a pink mustache/ Le Vi ne, S., Adamou, O., & Polak, J. (2014). Predicting new forms of activity/mobility patterns enabled by shared mobility services through a needs based stated response method: Case study of grocery shopping. Transport Policy, 32 60 68. doi: http://dx.doi.org /10.1016/j.tranpol.2013.12.008 Le Vine, S., Lee Gosselin, M., Sivakumar, A., & Polak, J. (2014). A new approach to predict the market and impacts of round trip and point to point carsharing systems: Case study of London. Transportation Research Part D: Tra nsport and Environment, 32 218 229. doi: http://dx.doi.org/10.1016/j.trd.2014.07.005 Levitt, S. (2016, September 07). H. Huggins. (Producer). Dream [Podcast]. Retrieved from http://freakonomics.com/podcast Lyft Blog. (2016). Fiv e Days. Six Cities. A Lyft Line First. Retrieved from http://blog.lyft.com/posts/six cities five days MacMillan, D. (2013, Nov 25, 2013). Uber Cuts Deals to Lower Car Costs. The Wall Street Journal Retrieved from http://blogs.wsj.com/digits/2013/11/25/ub er cuts deals to lower car costs/ Marshall, W. E., & Henao, A. (2015). The Shock Heard Round the Suburbs. Transportation Research Record: Journal of the Transportation Research Board, 2531 63 75. doi: 10.3141/2531 08 Martin, E., & Shaheen, S. (2011). The Impact of Carsharing on Public Transit and Non Motorized Travel: An Exploration of North American Carsharing Survey Data. Energies, 4 (11), 2094. Martin, E., Shaheen, S., & Lidicker, J. (2010). Impact of Carsharing on Household Vehicle Holdings. Transporta tion Research Record: Journal of the Transportation Research Board, 2143 150 158. doi: doi:10.3141/2143 19 Martin, E. W., & Shaheen, S. A. (2011). Greenhouse Gas Emission Impacts of Carsharing in North America. Intelligent Transportation Systems, IEEE Tra nsactions on, 12 (4), 1074 1086. doi: 10.1109/TITS.2011.2158539

PAGE 119

107 McAlone, N. (2015). Here's how Uber got its start and grew to become the most valuable startup in the world. Business Insider Retrieved from http://www.businessinsider.com/history of uber and its rise to become the most valuable startup in the world 2015 9 Meijkamp, R. (1998). Changing consumer behaviour through eco efficient services: an empirical study of car sharing in the Netherlands. Business Strategy and the Environment, 7 (4), 234 244. do i: 10.1002/(SICI)1099 0836(199809)7:4<234::AID BSE159>3.0.CO;2 A Murphy, C. (2016). Shared Mobility and the Transformation of Public Transit (pp. 39p): Shared Use Mobility Center. O'Donovan, C., & Singer Vine, J. (2016). How Much Uber Drivers Actually Make Per Hour. BuzzFeedNews Retrieved from https://www.buzzfeed.com/carolineodonovan/internal uber driver pay numbers?utm_term=.pl7eWrK0w#.km2W9aYR3 Pas, E. (1988). Weekly travel activity behavior. Transportation, 15 (1 2), 89 109. doi: 10.1007/BF00167982 Rayl e, L., Dai, D., Chan, N., Cervero, R., & Shaheen, S. (2016). Just a better taxi? A survey based comparison of taxis, transit, and ridesourcing services in San Francisco. Transport Policy, 45 168 178. doi: http://dx.doi.org/10.1016/j.tranpol.2015.10.004 Ro driguez, J. F. (2016). SF blasts Uber, Lyft for downtown traffic congestion Retrieved from: http://www.sfexaminer.com/sf blasts uber lyft downtown traffic congestion/ Salomon, B. (2016). Lyft: We're Closing In On Uber With A 'Path To Profitability'. Forbe s Salomon, I., & Ben Akiva, M. (1983). The Use of the Life Style Concept in Travel Demand Models. Environment and Planning A, 15 (5), 623 638. doi: 10.1068/a150623 Salonen, M., & Toivonen, T. (2013). Modelling travel time in urban networks: comparable meas ures for private car and public transport. Journal of Transport Geography, 31 (0), 143 153. doi: http://dx.doi.org/10.1016/j.jtrangeo.2013.06.011 Shaheen, S., & Cohen, A. (2007). Growth in Worldwide Carsharing: An International Comparison. Transportation Re search Record: Journal of the Transportation Research Board, 1992 81 89. doi: doi:10.3141/1992 10 Shaheen, S., & Cohen, A. (2012). Carsharing and Personal Vehicle Services: Worldwide Market Developments and Emerging Trends. International Journal of Sustainable Transportation, 7 (1), 5 34. doi: 10.1080/15568318.2012.660103

PAGE 120

108 Shaheen, S., Guzman, S., & Zhang, H. (2010). Bikesharing in Europe, the Americas, and Asia. Transportation Research Record: Journal of the Transportation Research Board, 2143 159 16 7. doi: doi:10.3141/2143 20 Shaheen, S. A., Cohen, A. P., & Chung, M. S. (2009). North American Carsharing: 10 Year Retrospective. Transportation Research Record: Journal of the Transportation Research Board (2110), pp 25 44. Shoup, D. C. (2006). Cruising for parking. Transport Policy, 13 (6), 479 486. doi: http://dx.doi.org/10.1016/j.tranpol.2006.05.005 Silver, N., & Fischer FiveThirtyEight Retrieved from FiveThirtyEight website: http://fivethirtyeight.com/features/public transit should be ubers new best friend/ Sioui, L., Morency, C., & Trpanier, M. (2012). How Carsharing Affects the Travel Behavior of Households: A Case Study of Montral, Canada. International Journal of Sustain able Transportation, 7 (1), 52 69. doi: 10.1080/15568318.2012.660109 Somerville, H. (2016). Here's How Many Rides Uber Has Given to Date. Money Steele, K. (2010). Bicycling and Walking in the United States: 2010 Benchmarking Report (pp. 196p). Washington, D.C.: Alliance for Biking & Walking. Steininger, K., Vogl, C., & Zettl, R. (1996). Car sharing organizations: The size of the market segment and revealed change in mobility behavior. Transport Policy, 3 (4), 177 185. doi: http://dx.doi.org/10.1016/S0967 070 X(96)00024 8 Stopher, P. R., & Greaves, S. P. (2007). Household travel surveys: Where are we going? Transportation Research Part A: Policy and Practice, 41 (5), 367 381. doi: http://dx.doi.org/10.1016/j.tra.2006.09.005 Truong, L., & Marshall, W. (2014). Are Park and Rides Saving the Environment or Just Saving Parking Costs? Transportation Research Record: Journal of the Transportation Research Board, 2419 109 117. doi: 10.3141/2419 11 U.S. Census Bureau. (2015). 2011 2015 American Community Survey 5 Year Es timates. Retrieved Jan 01, 2017 Uber Newsroom. (2014). Announcing UberPool. Retrieved from https://newsroom.uber.com/announcing uberpool/ Uber Newsroom. (2016). uberPOOL: Share your ride and save. Retrieved from https://www.uber.com/ride/uberpool/

PAGE 121

109 Vacc aro, A. (2016, December 05). Highly touted Boston Uber partnership has not lived up to hype so far. Boston.com Retrieved from http://www.boston.com/news/business/2016/06/16/bostons uber partnership has not lived up to promise Vij, A., Carrel, A., & Walker J. L. (2013). Incorporating the influence of latent modal preferences on travel mode choice behavior. Transportation Research Part A: Policy and Practice, 54 164 178. doi: http://dx.doi.org/10.1016/j.tra.2013.07.008 Vredin Johansson, M., Heldt, T., & Jo hansson, P. (2006). The effects of attitudes and personality traits on mode choice. Transportation Research Part A: Policy and Practice, 40 (6), 507 525. doi: http://dx.doi.org/10.1016/j.tra.2005.09.001 Vuong, A. (2014). Colorado first to authorize Lyft and DenverPost Retrieved from http://www.denverpost.com/2014/06/05/colorado first to authorize lyft and ubers ridesharing services/ Walker, J. L., & Ben Akiva, M. (2011). Advances in discrete choice: mixture models. A Handbook of Transport Economics 160. Retrieved from Buzzfeed News website: http://www.buzzfeed.com/charliewarzel/lets all join the ap stylebook in killing the term ride shar#.chR y8o4py