You are on page 1of 17

The Professional Geographer

ISSN: 0033-0124 (Print) 1467-9272 (Online) Journal homepage: https://www.tandfonline.com/loi/rtpg20

Uber, Public Transit, and Urban Transportation


Equity: A Case Study in New York City

Scarlett T. Jin, Hui Kong & Daniel Z. Sui

To cite this article: Scarlett T. Jin, Hui Kong & Daniel Z. Sui (2019): Uber, Public Transit, and
Urban Transportation Equity: A Case Study in New York City, The Professional Geographer, DOI:
10.1080/00330124.2018.1531038

To link to this article: https://doi.org/10.1080/00330124.2018.1531038

Published online: 24 Jan 2019.

Submit your article to this journal

View Crossmark data

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=rtpg20
Uber, Public Transit, and Urban Transportation Equity: A Case
Study in New York City
Scarlett T. Jin
The Ohio State University

Hui Kong
JTL Urban Mobility Lab of Singapore–MIT Alliance for Research and Technology (SMART)

Daniel Z. Sui
The Ohio State University

With the explosive growth of Uber and other mobility service providers, their influences on urban mobility have attracted
attention from both researchers and policy circles alike, yet few quantitative studies have been conducted on the topic.
Using Uber pickup data in New York City in 2014, this article investigates the spatiotemporal relationship between Uber
and public transit using buffer analysis and spatial cross-correlation analysis and assesses Uber’s impact on urban transpor-
tation equity with the Gini coefficient and correlation analysis. Our results confirm previous arguments that Uber both
complements and competes with public transit, but competition is more prevalent in New York City. Specifically, Uber
competes with public transit during most hours of the day and in areas with good public transit coverage, whereas it com-
plements public transit at midnight and in places with insufficient public transit services. The distribution of Uber services
is highly unequal, and Uber’s role in improving transport equity is insignificant. Correlation analysis shows that there tend
to be fewer Uber pickups in low-income areas, which diverges from previous studies suggesting that Uber serves low-
income areas well. In addition, a weak negative correlation is detected between the number of Uber pickups and the per-
centage of minorities. Key Words: New York City, public transit, ridesourcing, Uber, urban transportation equity.

随着Uber和其他移动服务提供者的爆炸性成长, 其对城市能动性的影响已同时吸引研究者和相关政策圈的关注, 但该主题


却仍鲜少有量化研究。本文运用2014年Uber在纽约市的载客数据, 使用环域分析与空间交互相关分析, 探讨Uber和公共运
输之间的时空关係, 并以吉尼指数和相关性分析, 评估Uber对城市运输公平的冲击。我们的研究结果, 确认Uber同时补充
公共运输并与其相互竞争的过往主张, 但在纽约市, 竞争则更为普遍。特别是Uber在一日中的多数时间内以及在公共运输
覆盖率高的地区与公共运输相互竞争, 在夜间公共运输与公共运输服务不足之处则与公共运输相互补充。Uber服务的分佈
是高度不均等的, 且Uber在改善运输平等中所扮演的角色是不显着的。相关分析显示, 在低收入的区域, Uber的载客量显
着较少, 因此与过往主张Uber同时服务低收入地区之研究有所分歧。此外, 本研究发现Uber载客数和少数族群比例之间呈
现弱负相关性。 关键词:纽约市, 公共运输, 共乘, Uber, 城市运输公平。

Con el explosivo aumento de Uber y otros proveedores de servicios m oviles, sus influencias sobre la movilidad urbana han
atraıdo la atenci
on tanto de investigadores como de los cırculos polıticos, pero sin que al mismo tiempo se hayan llevado a
cabo estudios cuantitativos al respecto en n umero suficiente. Con datos de servicios de Uber en la ciudad de Nueva York
en 2014, este artıculo investiga la relaci
on espaciotemporal entre Uber y el transito p ublico usando analisis buffer y analisis
de correlaci on espacial cruzada, y evalua el impacto de Uber sobre la equidad del transporte urbano por medio del coefi-
ciente de Gini y el analisis de correlaci
on. Nuestros resultados confirman anteriores argumentos en el sentido de que Uber
a la vez que complementa compite con el transito p ublico, aunque la competencia es lo prevalente en el caso de la ciudad
de Nueva York. Especıficamente, Uber compite con el transito p ublico durante la mayorıa de las horas del dıa y en areas
con buena cobertura de transito p ublico, en tanto que complementa el transito p ublico a la media noche y en lugares en
donde los servicios del transito publico son insuficientes. La distribuci
on de los servicios de Uber es altamente desigual, y
el papel de Uber para mejorar la equidad del transporte es insignificante. El analisis de correlaci on muestra que tienden a
registrarse menos recogidas de pasajeros en areas de bajos ingresos, lo cual se aparta de estudios anteriores que sugerıan
que Uber tambien servıa a areas de ingresos bajos. Ademas, se detecta una correlaci on negativa debil entre el numero de
recogidas de Uber y el porcentaje de minorıas. Palabras clave:: Ciudad de Nueva York, igualdad en el transporte
urbano, solicitud del servicio de taxi, transito p ublico, Uber.

T he explosive growth of Uber and other mobility


service providers, also known as transportation
network companies (TNCs), has put ridesourcing
vehicles with passengers” (Shaheen, Cohen, and
Zohdy 2016, 5). The term ridesourcing is commonly
used by academic transportation researchers, whereas
into the spotlight. Ridesourcing provides popular media often describe it as ride hailing or ride
“prearranged and on-demand transportation services sharing. Because ridesourcing is considered to be an
for compensation, which connect drivers of personal immediate substitute for taxis, there has been

The Professional Geographer, 0(0) 2019, pages 1–16 # 2019 American Association of Geographers.
Initial submission, February 2018; revised submissions, April and June 2018; final acceptance, August 2018.
Published by Taylor & Francis Group, LLC.
2 Volume 0, Number 0, Month 2019

abundant research and media coverage on the com- quantitative evidence has been provided to substan-
parison between ridesourcing and taxis and how the tiate this observation. The arguments on comple-
emergence of ridesourcing changes urban mobility mentarity generally come from two perspectives.
through transforming the taxi industry (Anderson From the temporal perspective, Feigon and
2014; Bialik et al. 2015; Gl€ oss, McGregor, and Murphy’s (2016) surveys find that ridesourcing is
Brown 2016). Furthermore, ridesourcing pulls people frequently used during the evening (8 p.m.–10 p.m.)
not only from taxis but also from public transit. and late-night and early-morning hours (11 p.m.–4
Through passenger surveys, existing research suggests a.m.) and is also more commonly used on weekends
that ridesourcing both complements and competes than weekdays. Spatially, ridesourcing is claimed to
with public transit (Jin et al. 2018). It is not clear, serve as a feeder of public transit to solve the first-
however, to what extent ridesourcing complements or mile/last-mile problem. Evidence supporting this
competes with public transit and how such a relation- claim is mainly provided by the TNCs instead of
ship evolves over time and space. In addition, there is independent research, though. For example, one
a lack of studies on how ridesourcing affects urban recent study conducted by Uber’s policy research
transportation equity in view of the fact that it indu- team finds that people use Uber to connect railway
ces modal shifts (change in transportation mode) stations to their final destinations, based on the sta-
from both taxis and public transit. Against this back- tistics that Uber pickups near the new railway sta-
drop, the objectives of this article are to (1) quantify tions of the Expo Line extension in Los Angeles
the spatiotemporal relationship between Uber and surged one month after the line was put into service
public transit and (2) assess how Uber affects urban (Williams 2017). We cannot tell from the data,
transportation equity. though, whether the increased Uber trips connect
The rest of this article is organized as follows. people from the new railway stations or from the
The next section reviews current research on ride- subsequent commercial and housing developments
sourcing and its impacts on urban mobility, followed near the new stations. Indeed, to figure out whether
by a section that explains the data and methodology Uber serves as a feeder of public transit is an intract-
used in this study. The empirical findings are pre- able task, which is discussed in detail in the method-
sented in the fourth section. Finally, we highlight the ology section of this article.
crucial findings of this article, contrast these results The arguments for competition, on the other
with previous studies, and point out the limitations of hand, are established based on surveys asking riders
this study and directions for future research. questions like, “If ridesourcing is not available, what
other transportation modes would you use?” (Feigon
and Murphy 2016; Rayle et al. 2016; Henao 2017;
Ridesourcing and Urban Mobility: Gehrke, Felix, and Reardon 2018). The results
Research Background reveal that 14 to 42 percent of riders would have
chosen public transit if ridesourcing was not avail-
Along with the growing popularity of ridesourcing able. Also using rider surveys, Clewlow and Mishra
services, there have been an increasing number of (2017) differentiate various types of public transit
studies and debates looking into the impacts of ride- and find that TNCs pull people away from the pub-
sourcing on urban efficiency, equity, and sustainabil- lic bus and light rail but enhance the ridership of
ity (Cramer and Krueger 2016; Schor 2016; heavy rail.
Shaheen and Chan 2016; Frenken and Schor 2017; In terms of the equity issue, limited studies have
Nie 2017). Among these studies and debates, two been done to investigate how ridesourcing affects
issues concerning urban mobility stand out: urban urban transportation equity. There are two general
transportation efficiency and equity. The former types of equity in the transportation equity and pol-
concerns how ridesourcing affects the urban trans- icy literature: horizontal and vertical equity.
portation network: Does it supplement the transpor- Horizontal equity concerns whether the costs and
tation network or worsen urban congestion? The benefits of transportation are allocated equally
latter addresses the role played by ridesourcing in between individuals and groups that are comparable
urban transportation equity. Discussions on these in wealth and ability, whereas vertical equity centers
two issues are convoluted because ridesourcing not on the distribution of costs and benefits between
only generates modal shift from taxis but also from income and social class and between people with dif-
public transit (Rayle et al. 2016). The relationship ferent mobility needs and abilities (Litman 2002;
between ridesourcing and public transit, which is Blumenberg 2017). Vertical equity focuses solely on
the primary focus of this article, has not been rigor- different groups’ levels of access to a city’s transpor-
ously studied. tation system. Inaccessibility to transportation serv-
Limited literature that touches on the relation- ices leads to social exclusion, because certain groups
ship between ridesourcing and public transit seems of residents are disconnected from activities and
to suggest that ridesourcing both complements and opportunities in their cities (Church, Frost, and
competes with public transit, although insufficient Sullivan 2000; Lucas 2012). Following this line of
Uber, Public Transit, and Urban Transportation Equity 3
enquiry on transport and social exclusion, existing number of Uber pickups and scheduled public tran-
studies of ridesourcing equity investigate access to sit pickups. To measure the degree of inequality in
ridesourcing services among those of different transportation supply among populations, we adopt
income and social classes. The first group of studies the Gini coefficient and the Lorenz curve. We also
compares ridesourcing with taxis and asserts that calculate the Pearson correlation coefficient to
Uber does a better job than taxis in terms of explore the relationship between the number of
improving the transportation accessibility of low- Uber pickups with population, income, and the per-
income neighborhoods (Bialik et al. 2015; Smart centage of minorities.
et al. 2015). Nevertheless, though ridesourcing
might serve low-income areas better than taxis, it
might still be insufficient to draw the conclusion Data
that ridesourcing can improve transportation equity, We collected five categories of data from various
especially when it starts to merge with public transit. sources, as explained in the following:
Many public transportation agencies in the United
States partner with TNCs, using subsidized or free 1. Uber pickups. Data on Uber pickups in
ridesourcing rides to supplement or even substitute New York City were obtained by
for public transit routes (Shaheen and Chan 2016). FiveThirtyEight from the New York City
People question whether such practices are fair to Taxi and Limousine Commission (TLC)
all stakeholders (Brustein 2016), because public tran- through a Freedom of Information Law
sit services are meant to be available to all commu- request in 2015, which were then shared
nity members, whereas ridesourcing users are more on the GitHub.1 We extracted and ana-
likely to be young, well-educated, and affluent indi- lyzed the data for one weekday (17
viduals (McGrath 2015; Rayle et al. 2016; A. Smith September 2014) and one weekend day
2016; Clewlow and Mishra 2017; Gehrke, Felix, and (20 September 2014) based on the goals of
Reardon 2018). our research. The data contain informa-
The second type of research on the equity of tion on the time, longitude, latitude, and
ridesourcing focuses on ridesourcing itself. base name of 35,531 Uber pickup records
Specifically, it examines the availability of ridesourc- on 17 September and 38,863 records on
ing services to different income, race, gender, and 20 September.
ethnic groups. Using trip data, Hughes and 2. Public transit stops and service schedules. Data
MacKenzie (2016) find that the waiting time for on public transit stops and service sched-
UberX in areas with lower average income is shorter ules were obtained from the Metropolitan
than that in areas with higher average income, and Transportation Authority (MTA) Data
the average association between the percentage of Feeds.2 Using a Python script, we proc-
minorities and the length of waiting time is close to essed the data to extract information on
zero. By analyzing the accessibility measured by public transit stop locations and the
waiting time, Wang and Mu (2018) conclude that vehicle arrival time at each stop.
wealth and race do not have significant associations 3. Taxi data. Taxi data were obtained from
with Uber accessibility, whereas higher road net- the New York City TLC.3 The data
work density, population density, and less commut- include yellow and green taxi trip records
ing time to work do correlate with greater Uber that capture pickup and dropoff dates,
accessibility. Examining the racial and gender dis- times, locations, and other information.
crimination issue of Uber and Lyft, Ge et al.’s 4. Geographic information system layers.
(2016) rider experiment identifies discrimination Geographic information systems (GIS)
against African American riders and excessive fares layers used in this study include shapefiles
targeted at female riders. collected from the U.S. Census Bureau4
and bus, subway, and rail routes (as of
January 2017) obtained from the William
Data and Methodology and Anita Newman Library of
Baruch College.5
Reflecting on the unsolved questions of existing lit- 5. Nonspatial data. We acquired New York
erature, we adopt Uber pickup data in New York City’s socioeconomic data from the
City to analyze the spatiotemporal relationship 2010–2014 American Community Survey
between Uber and public transit through buffer area five-year estimates,6 including total popu-
analysis and spatial cross-correlation analysis. Buffer lation, median household income, and
analysis identifies the relationship between the race at the census block group (CBG)
coverage of public transit stops and the location of level. The data for the year 2014 were
Uber pickups, whereas spatial cross-correlation used to be consistent with Uber
measures the spatial relationship between the pickup data.
4 Volume 0, Number 0, Month 2019

Figure 1 Sample buffer areas. (Color figure available online.)

Table 1 The relationship between Uber and public transit


Relationship
between Uber and
Uber pickup location Scenario Trip origin Trip destination public transit
A. Within 100-m buffer area I Uber pickup location Competition
is the origin
II Uber pickup location is the Final destination Competition
transfer location is inside buffer (Uber as a feeder and a
area (A and B) competitor)
III Uber pickup location is Final destination is Complement
the transfer location outside buffer area (C) (Uber solves the
last-mile problem)
B. 100- to 400-m or 800-m Competition
buffer area
C. Outside buffer area Complement
(Uber improves
transportation accessibility)

Methods A distance of 400 m is commonly assumed to be a


Buffer Analysis. Transit coverage is a measure of comfortable walking distance for most people under
the spatial spread and accessibility of transit services, normal conditions, so it is used as the threshold dis-
and people are considered “covered” by public tran- tance to evaluate public transit coverage in many
sit if they can access a transit stop within an accept- studies (Demetsky and Lin 1982; Murray et al.
able distance (Hawas, Hassan, and Abulibdeh 2016). 1998; Wu and Murray 2005; Hawas, Hassan, and
Uber, Public Transit, and Urban Transportation Equity 5
Abulibdeh 2016). Alternatively, P. N. Smith and Whether Uber, as a feeder, complements or com-
Taylor (1994) use 800 m as the threshold distance to petes with public transit is further contingent,
evaluate the coverage of suburban railway, which is though, on the rider’s final destination, which causes
reasonable because people might be willing to walk the difference between Scenarios II and III.
a longer distance to take the railway. In Scenario II, the rider’s final destination is cov-
In this study, both 400 m (for subway and buses) ered by public transit buffers (Areas A and B) and is
and 800 m (for railway) are used as the threshold thus accessible via public transit. Under this scen-
distances to measure the coverage of public transit. ario, Uber is both a feeder and a competitor of pub-
Additionally, we introduce another threshold of lic transit, because the rider chooses Uber to travel
100 m to assist our analysis of the relationship to an area that can be reached by public transit to
between Uber and public transit. We choose 100 m avoid multiple transfers. In Scenario III, the rider’s
to make our analysis comparable to Williams’s final destination is not well covered by the public
(2017) study so that meaningful discussions can be transit network (Area C), in which case Uber indeed
generated. As such, we define three areas (Figure 1): complements public transit and solves the last-
(A) within the 100-m buffer area, (B) 100-m to mile problem.
400-m or 800-m buffer area, and (C) outside buffer
area. The relationship between Uber and public Spatial Cross-Correlation. As a supplement to buf-
transit is determined by the location of Uber pick- fer analysis, which identifies the relationship
ups in reference to the three buffer areas (Table 1). between public transit and Uber based on service
First, when Uber pickups are located in the coverage, spatial cross-correlation measures the spa-
100-m to 400-m or 800-m buffer area (Area B), it is tial relationship based on service frequency. Instead
less likely that riders use Uber to connect public of simply using Pearson correlation coefficients to
transit stations with their final destinations. Hence, characterize the relationship, which ignores the
we estimate that Uber competes with public transit effects of spatial distances, this article adopts spatial
under this condition. Second, when Uber pickups cross-correlation to measure the spatial relationship
take place outside the coverage of public transit buf- between the number of Uber pickups and scheduled
fers (Area C), Uber is considered to complement public transit pickups at the scale of CBG in view of
public transit and improve the transportation acces- distance decay. The spatial cross-correlation index
sibility of areas with relatively insufficient public (SCI) is defined by two size measurements (number
transit services. Third, when Uber pickups are of Uber pickups and schedule public transit pickups)
within the 100-m buffer area (Area A), we cannot and a distance weights matrix (Chen 2015). Detailed
ascertain whether Uber complements or competes formulas to calculate SCI are illustrated in Appendix
with public transit without additional information A. Scheduled public transit pickups are calculated by
on riders’ origins and final destinations. There are the following formula:
three possible scenarios in this case.
In Scenario I, the Uber pickup location is a X
n
Pi ¼ Dk ; (1)
rider’s trip origin, meaning that the rider takes
k¼1
Uber instead of public transit when a public tran-
sit station is available within 100 m. In this scen- where Pi is the scheduled public transit pickups in
ario, Uber obviously competes with public transit. CBG i, Dk is the total number of public transit
This scenario is highly possible because many housing, vehicles that arrive at stop k during a certain time
business, retail, and entertainment facilities are built period, and n is the number of stops in CBG i.
within close range of public transit stops. The SCI could be interpreted as the correlation
In scenarios II and III, the pickup location is a between the scheduled public transit pickups in each
transfer location, where a rider uses Uber to travel CBG and the number of Uber pickups in the nearby
from the public transit station to his or her final des- CBGs, or vice versa (Table 2). A positive SCI means
tination, making Uber a public transit feeder. that a CBG with more public transit pickups tends to

Table 2 Spatial cross-correlation index (SCI) interpretation


Relationship
between Uber and
Sign of SCI Interpretation public transit
Positive (+) Census block groups with Competition
more public transit pickups are
surrounded by more Uber pickups and vice versa
Negative (−) Census block groups Complement
with more public transit pickups
are surrounded by fewer Uber pickups and vice versa

Note: SCI ¼ spatial cross-correlation index.


6 Volume 0, Number 0, Month 2019

Figure 2 Uber pickups and public transit scheduled trips per hour, 15–21 September 2014.

Table 3 Public transit coverage (as percentage of area) by time slot and borough in New York City
Time slots Bronx (%) Brooklyn (%) Manhattan (%) Queens (%) Staten Island (%) Whole city (%)
1. Weekday morning (7 a.m.–8 a.m.) 85.2 88.4 96.5 62.8 74.7 76.7
2. Weekend morning (7 a.m.–8 a.m.) 83.2 87.8 96.1 59.8 66.8 73.6
3. Weekday noon (11 a.m.–12 p.m.) 84.6 87.9 96.2 61.1 73.7 75.6
4. Weekend noon (11 a.m.–12 p.m.) 84.5 87.9 96.2 60.1 66.8 73.9
5. Weekday afternoon (6 p.m.–7 p.m.) 84.5 88.3 96.6 62.2 74.0 76.3
6. Weekend afternoon (6 p.m.–7 p.m.) 84.5 87.9 96.2 60.2 66.8 74.0
7. Weekday evening (9 p.m.–10 p.m.) 84.6 88.3 96.2 60.5 70.6 74.9
8. Weekend evening (9 p.m.–10 p.m.) 84.5 87.8 96.2 60.1 66.5 73.8
9. Weekday midnight (12 a.m.–1 a.m.) 57.5 52.6 78.0 33.8 46.9 47.3
10. Weekend midnight (12 a.m.–1 a.m.) 54.0 55.6 78.4 34.4 42.2 46.9

have more Uber pickups nearby, which implies a com- the population, the median household income, and
petitive relationship between Uber and public transit. the percentage of minorities at the CBG level.
A negative SCI indicates a complementary relation-
ship, because Uber pickups concentrate in places with
insufficient public transit pickups, whereas places with Results
ample public transit supplies have fewer Uber pickups.

Transportation Equity Assessment. The Gini Relationship between Uber and Public Transit
coefficient has traditionally been used to calculate Buffer Analysis. First we look at the diurnal and
the distribution of income among populations. It is weekly variation in public transit and Uber services.
used in this study to express the overall degree of Figure 2 shows the hourly supply of public transit
inequality of transportation provision by calculating (including bus, subway, and railway) and Uber serv-
the distribution of transportation supply among ices during a week (15–21 September 2014) in New
populations. Following Delbosc and Currie (2011) York City. Several observations can be detected
and Kaplan et al. (2014), the Gini coefficient is from Figure 2 to assist buffer analysis. First, New
approximated as: York City has an exceptionally large supply of public
transit services. Second, from Monday to Thursday,
X
n
both public transit trips and Uber pickups decline
G¼1 ðXk Xk1 ÞðYk þ Yk1 Þ; (2)
k
steeply during late-night and early-morning hours
(around 11 p.m. to 3 a.m.). Third, although the
where k is the ranked order of each CBG’s transpor- number of public transit trips declines significantly
tation supply, and k ¼ 1, … , n; Xk is the cumulative during weekends, the number of Uber trips is rather
proportion of population, with X0 ¼ 0 and Xn ¼ 1; consistent throughout the week and notably high on
Yk is the cumulative proportion of transportation Friday and Saturday nights. These results imply that
supply, with Yo ¼ 0 and Yn ¼ 1. The Gini coefficient Uber might complement public transit during late-
ranges from zero to one, with zero representing per- night and early-morning hours on Friday
fect equality and one representing perfect inequality. and Saturday.
The Lorenz curve is used to graphically represent Regarding public transit coverage and Uber pick-
the distribution of transportation supply across the ups, in view of the fluctuating supply of public tran-
population. In addition, the Pearson correlation coef- sit and Uber services over time, ten time slots are
ficient is calculated to investigate the relationship extracted to conduct buffer analysis. As a first step,
between the supply of Uber and public transit serv- we calculate the public transit coverage of New
ices and various socioeconomic indicators, including York City boroughs during the ten time slots (Table
Uber, Public Transit, and Urban Transportation Equity 7

Figure 3 Public transit coverage in New York City (time slots 1 and 10). (Color figure available online.)

Figure 4 Uber pickups in New York City (time slots 1 and 10). (Color figure available online.)

3). Public transit coverage refers to the percentage transit coverage (around 74–77 percent for the
of areas covered by public transit buffers. In general, whole city), whereas time slots 9 and 10 present a
New York City has a very high level of public transit lower coverage level (about 47 percent). This result
coverage. Comparing across boroughs, Manhattan indicates that public transit coverage declines
has the highest level of coverage and Queens has the remarkably at midnight (12 a.m.–1 a.m.). The week-
lowest coverage. Comparing over time, time slots 1 day–weekend difference in public transit coverage is
through 8 demonstrate a higher level of public not obvious, probably because buffer analysis only
8 Volume 0, Number 0, Month 2019

Table 4 Buffer analysis result by time slot in New York City


B. 100- to 400-m or C. Outside
A. Within 100-m buffer area 800-m buffer area buffer area

Time slots Uber pickups % Uber pickups % Uber pickups %


1. Weekday morning (7 a.m.–8 a.m.) 1,303 67.9 593 30.9 24 1.3
2. Weekend morning (7 a.m.–8 a.m.) 436 59.6 278 38.0 18 2.5
3. Weekday noon (11 a.m.–12 p.m.) 804 67.4 371 31.1 18 1.5
4. Weekend noon (11 a.m.–12 p.m.) 936 66.5 444 31.5 28 2.0
5. Weekday afternoon (6 p.m.–7 p.m.) 1,896 68.3 822 29.6 59 2.1
6. Weekend afternoon (6 p.m.–7 p.m.) 1,758 66.7 793 30.1 84 3.2
7. Weekday evening (9 p.m.–10 p.m.) 1,407 66.6 680 32.2 26 1.2
8. Weekend evening (9 p.m.–10 p.m.) 1,691 69.1 725 29.6 32 1.3
9. Weekday midnight (12 a.m.–1 a.m.) 104 29.3 207 58.3 44 12.4
10. Weekend midnight (12 a.m.–1 a.m.) 464 28.2 973 59.2 207 12.6
Relationship Competition or complement Competition Complement

Table 5 Buffer analysis result by borough, weekday morning (7 a.m.–8 a.m., 17 September)
Buffer area Bronx Brooklyn Manhattan Queens Staten Island Whole city
A. Within 100-m buffer area Uber pickups 6 139 1,127 30 1 1,303
% 54.5 54.9 73.0 26.8 100 67.9
Uber pickups/km2 0.18 2.01 33.72 0.66 0.03 6.14
B. 100- to 400-m or 800-m buffer area Uber pickups 5 113 416 59 0 593
% 45.5 44.7 27.0 52.7 0 30.9
Uber pickups/km2 0.08 1.25 17.64 0.45 0 1.53
C. Outside buffer area Uber pickups 0 1 0 23 0 24
% 0 0.4 0 20.5 0 1.3
Uber pickups/km2 0 0.05 0 0.22 0 0.13

Table 6 Buffer analysis result by borough, weekend midnight (12 a.m.–1 a.m., 20 September)
Buffer area Bronx Brooklyn Manhattan Queens Staten Island Whole city
A. Within 100-m buffer area Uber pickups 1 102 356 5 0 464
% 14.3 29.7 28.8 8.8 28.2
Uber pickups/km2 0.10 5.16 32.24 0.33 0 6.98
B. 100- to 400-m or 800-m buffer area Uber pickups 6 205 739 23 0 973
% 85.7 59.6 59.8 40.4 59.2
Uber pickups/km2 0.12 2.55 20.93 0.28 0 3.24
C. Outside buffer area Uber pickups 0 37 141 29 0 207
% 0 10.8 11.4 50.9 12.6
Uber pickups/km2 0 0.46 11.05 0.16 0 0.50

considers service availability, not frequency. This Table 7 Global spatial cross-correlation index
implies that the majority of public transit stations in
Borough Weekday Weekend
New York City are still operative during weekends
Bronx 0.0027 0.0026
but with lower service frequencies. Compared to Brooklyn 0.0010 0.0031
the extensive coverage of public transit stations Manhattan 0.0129 0.0097
(Figure 3), Uber services are more concentrated in Queens 0.0038 0.0083
Staten Island 0.0051 0.0012
Manhattan and the northern part of Brooklyn Whole city 0.0208 0.0186
(Figure 4). Staten Island, on the other hand, has so
few Uber pickups that few significant results can be Note: Uber data are for 17 September (weekday) and 20
September (weekend). Public transit includes bus, subway,
generated for it. and rail.
The total number of Uber pickups in each of the
three buffer areas during the ten time slots are improves the transportation accessibility of areas
counted and present in Table 4. Comparing across with poor public transportation coverage in New
time slots, the distribution patterns of Uber pickups York City. The competing effect is evident, on the
in time slots 1 through 8 are similar, even though other hand, because one third of Uber pickups take
the absolute number of pickups fluctuates. The place in Area B. For Uber pickups located in Area
small proportion of Uber pickups in Area C indi- A, as explained later, the relationship is uncertain
cates that, in time slots 1 through 8, Uber hardly because we have no idea whether the pickup location
Uber, Public Transit, and Urban Transportation Equity 9

Figure 5 Hourly SCI of New York City boroughs on (A) weekdays and (B) weekends. SCI ¼ spatial cross-correl-
ation index.

is the rider’s origin or transfer location or where the unlikely to happen because if a rider uses Uber to
rider’s final destination is. That said, we can still connect a public transit stop to his or her final des-
make some conjecture by looking into the borough tination, the final destination is most likely to be
level, which is discussed later in this section. located inside buffer areas (Areas A and B).
The distribution patterns of Uber pickups in Therefore, we argue that in Manhattan, Uber com-
time slots 9 and 10 are significantly different from petes with public transit most of the time, except for
the other time slots, with a higher percentage of midnight. At midnight, as public transit coverage
pickups located in Areas B and C and a lower per- mildly declines, Uber slightly improves transporta-
centage in Area A. The absolute number of Uber tion accessibility in those poorly covered areas.
pickups in time slots 9 and 10 diverges, though their Second, in Queens, patterns of Uber pickups are
patterns are similar. There are considerably more different from those of other boroughs. Most not-
Uber pickups on weekends than on weekdays. able, the percentage of Uber pickups taking place in
Given that the distribution pattern of Uber pick- Area C is much higher in Queens (20.5 percent on
ups at midnight diverges significantly from the other weekday mornings and 50.9 percent on weekends at
time periods, we choose time slots 1 (as representa- midnight). This result implies that Uber has a much
tive of time slots 1–8) and time slot 10 (representing higher complementary effect on the public transpor-
time slots 9 and 10) to conduct further analysis on tation network in Queens, where public transit
the borough level (Tables 5 and 6), which generates coverage is relatively low.
two interesting observations.
First, Manhattan stands out from the five bor- Spatial Cross-Correlation. To supplement buffer
oughs, because Uber pickups are highly concen- analysis, which only accounts for public transit
trated in Manhattan. More important, given that coverage, not service frequency, the SCI is calcu-
Manhattan has extremely high public transit cover- lated to analyze the relationship between the num-
age (over 96 percent during the daytime and over 78 ber of scheduled public transit pickups and Uber
percent at midnight), we can speculate on the rela- pickups (Table 7). As explained later, a positive SCI
tionship between Uber and public transit when indicates a competitive relationship, whereas a nega-
Uber pickups are located in Area A. Recalling the tive SCI indicates a complementary relationship. In
scenarios in Table 1, with such a high rate of public Manhattan, where the public transit coverage is the
transit coverage in Manhattan, Sscenario III is highest and Uber pickups are highly concentrated,
10 Volume 0, Number 0, Month 2019

Table 8 Gini coefficient of public transit, Uber, and taxi


Weekday Weekend

Borough Public transit Uber Taxi Public transit Uber Taxi


Bronx 0.5803 0.9014 0.7384 0.6079 0.8823 0.8359
Brooklyn 0.5738 0.8024 0.8662 0.5852 0.7971 0.8859
Manhattan 0.6324 0.7200 0.7187 0.6625 0.6499 0.6964
Queens 0.7425 0.9391 0.9518 0.7359 0.8642 0.9257
Staten Island 0.4336 0.9912 0.9761 0.4951 0.9728 0.9477
Whole city 0.6653 0.9028 0.9223 0.6429 0.8634 0.9045

Figure 6 Lorenz curves of Uber and public transit.

there is a competitive relationship between the two and weekends, indicating a constant complementary
transportation modes. In the Bronx and Staten relationship.
Island, Uber also competes with public transit, but
the competitive relationship is not as significant as
in Manhattan, represented by smaller SCIs. In Transportation Equity
Queens, on the other hand, the negative SCI indi- Gini Coefficients and Lorenz Curves. The second
cates that Uber complements public transit. objective of this article is to understand the role that
To explore the diurnal variations in the relation- Uber plays in urban transportation equity. The first
ship, hourly SCIs are calculated (Figure 5). In step is to calculate the Gini coefficient of three
Manhattan, the SCI is positive most of the time transportation modes: public transit, Uber, and taxis
except between 11 p.m. and 1 a.m. and 4 a.m. to 7 (Table 8). Results show that the Gini coefficient of
a.m. on weekdays and 12 a.m. to 1 a.m. and 6 a.m. public transit is significantly lower than that of the
to 7 a.m. on the weekends. In contrast, the SCI of other two. This indicates that public transit services
Queens is negative throughout both the weekdays are more equally distributed among residents in
Uber, Public Transit, and Urban Transportation Equity 11
Table 9 Gini coefficient without and with Uber
Weekday Weekend

Borough Public transit Public transit þ Uber Public transit Public transit þ Uber
Bronx 0.5803 0.5802 0.6079 0.6073
Brooklyn 0.5738 0.5725 0.5852 0.5783
Manhattan 0.6324 0.6276 0.6625 0.6352
Queens 0.7425 0.7424 0.7359 0.7332
Staten Island 0.4336 0.4335 0.4951 0.4949
Whole city 0.6653 0.6321 0.6429 0.6349

Note: Uber data are for 17 September (weekday) and 20 September (weekend). Public transit includes
bus, subway, and rail.

Table 10 Pearson correlation coefficients


Weekday Weekend

Correlation Public transit Uber Public transit Uber


Transportation and population 0.0851 0.0406 0.0738 0.0958
Transportation and income 0.0614 0.3136 0.0502 0.3734
Transportation and minority 0.0090 0.1771 0.0139 0.2250

Note: Uber data are for 17 September (weekday) and 20 September (weekend). Public transit
includes bus, subway, and rail.
Statistically significant at the 0.001 level.

New York City, whereas the distribution of Uber transit and Uber (Table 9). The results show that
and taxi services is highly unequal. The distribution when Uber is added to the transportation system, the
of Uber is only slightly more equitable than that of Gini coefficient is only slightly smaller. Therefore,
taxis. Comparing across boroughs, several interest- Uber’s role in improving transport equity is
ing findings could be identified. First, public transit insignificant.
services are less equally distributed in Queens than
in other boroughs. This could be explained by the Correlation. Pearson correlation coefficients are
relatively insufficient public transit supply in calculated to investigate the relationship between
Queens. Second, the distribution of Uber and taxi the supply of transportation services and some
services is more equitable in Manhattan, where there socioeconomic indicators at the CBG level,
are more Uber and taxi pickups. Third, the supply including population, income, and ethnicity
of Uber and taxi services is highly unequal in Staten (Table 10).
Island, probably because there is an extremely small The coefficients between transportation supplies
number of pickups. and population are extremely small, indicating no
The Lorenz curves visualize the distribution of significant correlation between the two. This can be
Uber and public transit services among the residen- attributed to the composition of population in New
tial population (Figure 6). The black dashed line York City. As a popular tourist city, New York City
represents perfectly equitable distribution. The fur- attracted 56.5 million tourists in 2014,7 and its total
ther a curve is away from the perfect equity line, the residential population was only 8.35 million in the
more inequality it represents. Our results show that same year.8 Tourists constitute a large proportion of
the Lorenz curve of Uber is further away from the transportation users, but they are not counted
perfect equity line, indicating that the distribution toward the “total population” of the CBGs.
of Uber is less equitable than that of public transit. Commuters who work in New York City but live in
For Uber, 20 percent of residents enjoy 95 percent surrounding counties are another group of passen-
of Uber services, whereas the remaining 80 percent gers not recorded by the population statistics.
of the population in New York City share only Therefore, it is reasonable that the population cen-
about 5 percent of Uber services. In comparison, 80 sus data, which only reflect where people live, do
percent of New York City’s population share around not have a significant correlation with transporta-
34 percent of public transit. tion supplies.
As a second step, we assess whether Uber The correlation coefficients between scheduled
improves the overall transportation equity in New public transit pickups and median household income
York City through comparing the Gini coefficients of are 0.06 on the weekday and 0.05 on the weekend.
the transportation system that contains only public Such small numbers indicate that the supply of pub-
transit with the system that contains both public lic transit services in New York City is irrelevant to
12 Volume 0, Number 0, Month 2019

neighborhood income level. In contrast, the correl- public transit, the relationship is still competitive.
ation coefficients between Uber pickups and income On the contrary, in Queens, where public transit
are 0.31 on weekdays and 0.37 on the weekend. coverage is lower, Uber has a much higher comple-
These numbers demonstrate a weak positive rela- mentary effect on public transit. Our results are con-
tionship between the supply of Uber services and sistent with Feigon and Murphy’s (2016)
the income level of CBGs, indicating that more observations that ridesourcing is frequently used
Uber pickups tend to locate in areas with when public transit is sporadic. We do not sup-
higher income. port Feigon and Murphy’s (2016) general conclu-
Regarding the relationship between transporta- sion that ridesourcing complements public transit,
tion supplies and the percentage of minorities, however, because our findings show that Uber
results are quite divergent for the two transporta- also competes with public transit in other time
tion modes. The coefficients between scheduled periods of the day. In light of the limitation of
public transit pickups and percentage of minor- Williams’s (2017) attempt to validate Uber as a
ities are very small (0.009 on weekdays and 0.014 feeder of public transit, this research investigates
on weekends), indicating that the supply of public the situation in which Uber pickups take place at
transit services is irrelevant to the distribution of a close distance to public transit stations and
minority groups in the city. In contrast, there is a argues that there are three different scenarios
weak negative correlation (0.18 on weekdays and depending on the riders’ origins and destinations.
0.23 on weekends) between Uber pickups and Even when Uber does serves as a public transit
minorities, indicating that fewer Uber pickups are feeder, it does not necessarily complement pub-
located in the CBGs with a higher minor- lic transit.
ity population. The second objective of this article is to explore
Considering the potential impacts of spatial auto- how Uber affects urban transportation equity.
correlation on the correlation analysis results, the Results of Gini coefficients and Lorenz curves dem-
Moran’s I is calculated for the number of public onstrate that the distribution of Uber services is
transit pickups, the number of Uber pickups, popu- highly unequal and a very small proportion of the
lation, income, and percentage of minorities, which population in New York City enjoys most of the
shows weak spatial autocorrelation (Appendix B), Uber services. Also, Uber’s role in improving trans-
not strong enough to impede our main conclusions. port equity is insignificant. The supply of Uber and
By calculating the SCI and partial correlation coeffi- public transit services has no significant correlation
cient Rp, we identify both a significant total correl- with the number of people living in the CBGs. The
ation and a strong cross-correlation free of spatial supply of public transit services is irrelevant to the
decay effects (Appendix C). median household income of the CBGs. In contrast,
we observe a weak positive relationship between the
number of Uber pickups and the income level. This
Summary and Conclusion result enriches our understanding of how Uber
serves low-income neighborhoods. The rider experi-
The first objective of this article is to examine the ment by Smart et al. (2015) shows that UberX offers
spatiotemporal relationship between Uber and pub- cheaper and faster services than taxis in low-income
lic transit. In general, our analysis confirms previous neighborhoods in Los Angeles. The study by
arguments that Uber both complements and com- Hughes and MacKenzie (2016) concludes that wait-
petes with public transit, beyond which we identify ing time for UberX is shorter in low-income areas.
how the relationship varies over time and space. Yet our study finds that there tend to be fewer Uber
Uber competes with public transit during most pickups in areas with lower income. With regard to
hours of the day and in areas with good public tran- ethnicity, a weak negative correlation is identified
sit coverage, whereas it complements public transit between the number of Uber pickups and the per-
at midnight and in places where public transit serv- centage of minorities in CBGs. This result resonates
ices are inadequate. Temporally, the relationship with Ge et al.’s (2016) rider experiment that
between Uber and public transit exhibits diurnal and observes discrimination of Uber and Lyft against
weekly rhythms. Whereas the relationship is more minority riders.
competitive during most hours of the day, it tilts This article is among the few efforts to provide
toward complementary at midnight, with evidence empirical evidence on the impact of the sharing
of more Uber pickups located in areas not covered economy on transportation systems and urban
by public transit buffers and negative SCIs in most equity using big data analytics. Our results have
boroughs. Spatially, the relationship varies across interesting implications for policymaking. Because
boroughs in New York City. In Manhattan, where the assumption that TNCs only serve as feeders of
public transit coverage is extremely high, Uber public transit and thus merely complement public
mostly competes with public transit (except for mid- transit has been proved erroneous by our study,
night). Even when Uber does serve as a feeder for public agencies should take precautions prior to
Uber, Public Transit, and Urban Transportation Equity 13
partnering with TNCs. On the other hand, the transportation infrastructures and different urban
contrasting results between Manhattan and forms. 䊏
Queens imply that how ridesourcing works with
public transit is determined by the public trans-
portation infrastructure. Ridesourcing might Acknowledgments
have a stronger complementary effect in cities
and areas with a relatively poor public transit The first two authors contributed equally to this
network. The challenging question for policy- article. They are joint first authors. Should you
makers is, then, should we rely on TNCs to have any question, please address to the
improve urban mobility or should we continue to corresponding author Hui Kong via
invest in and improve public transit networks? kong.hui@smart.mit.edu.
How should public agencies effectively partner
with TNCs to improve urban mobility while tak-
ing account of the equity issue? One possible
Notes
1
solution would be to limit ridesourcing on major GitHub is a software development platform. The data
urban corridors with adequate public transit we used are available at https://github.com/
coverage and encourage it in areas with insuffi- fivethirtyeight/uber-tlc-foil-response.
2
cient public transportation infrastructure. MTA Data Feeds: http://web.mta.info/developers/
Despite the significance of our results, this study developer-data-terms.html#data. The data we used were
updated on 23 August 2017.
has two limitations that necessitate further research. 3
NYC Taxi and Limousine Commission (TLC): https://
The first limitation is data restriction. Because the
docs.google.com/document/d/1iomkE6nF9kwsJSv3BQ0
data are from September 2014, they are a bit out- p65SNe1TDpR0k8mOTVV7FAvY/edit#
dated, considering the accelerated pace of mobility 4
U.S. Census Bureau, TIGER/Line Shapefiles: https://
service growth during the past four years. The data www.census.gov/geo/maps-data/data/tiger-line.html.
used in this study do not capture the effect of 5
Baruch College, The William and Anita Newman
UberPOOL (a ridesplitting service that allows pas- Library: www.baruch.cuny.edu/confluence/display/
sengers with similar routes to share a ride and split geopotal/NYWþMassþTransitþSpatialþLayers.
6
the fare), because UberPOOL was launched in New U.S. Census Bureau: https://factfinder.census.gov/faces/
York City in December 2014. Furthermore, dropoff nav/jsf/pages/index.xhtml.
7
records are not available in our data set. Without NYC & Company: http://www.nycandcompany.org/
dropoff information, we cannot figure out the exact research/nyc-statistics-page.
8
U.S. Census Bureau: https://factfinder.census.gov/faces/
relationship between Uber and public transit when
nav/jsf/pages/index.xhtml.
Uber pickups are located in the 100-m buffer area.
Either a comprehensive data set with both pickup
and dropoff records or rider surveys is necessary to
ORCID
further advance our understanding of ridesourcing’s
impact on urban mobility. More crucially, given the Scarlett T. Jin http://orcid.org/0000-0002-
increasing importance of ridesourcing in the urban 2714-6873
transportation system, city governments should Hui Kong http://orcid.org/0000-0001-
require TNCs to provide detailed origin–destination 6048-9440
data for the government and researchers to accur-
ately assess their impacts on urban efficiency, equity, Literature Cited
and sustainability. For those cities that initiated
cooperation with TNCs, the collection of such data Anderson, D. N. 2014. Not just a taxi? For-profit rideshar-
is pivotal to justify and evaluate the partner- ing, driver strategies, and VMT. Transportation 41 (5):
ship programs. 1099–1117.
The second limitation comes from the special Bialik, C., A. Flowers, R. Fischer-Baum, and D. Mehta.
characteristics of New York City. New York City 2015. Uber is serving New York’s outer boroughs more
was chosen as the study area due to data availability, than taxis are. Accessed July 1, 2017. https://fivethir-
yet it is a unique city with an exceptionally sophisti- tyeight.com/features/Uber-is-serving-new-yorks-outer-
boroughs-more-than-taxis-are/.
cated public transit network, which is different from
Blumenberg, E. 2017. Social equity and urban trans-
many cities in the United States and worldwide.
portation. In The geography of urban transportation, 4th
Cautions need to be exercised when applying the ed., ed. G. Giuliano and S. Hanson, 332–58. New York:
results of our study to other cities like Los Angeles, Guilford.
where public transit is less well developed and Brustein, J. 2016. Uber and Lyft want to replace public
dependence on the automobile is high. In light of buses. Accessed July 1, 2017. https://www.bloomberg.
these considerations, future studies should explore com/news/articles/2016-08-15/uber-and-lyft-want-to-
how ridesourcing affects cities with different public replace-public-buses.
14 Volume 0, Number 0, Month 2019

Chen, Y. 2013. New approaches for calculating Moran’s provision. Journal of Transport Geography 37:82–92. doi:
index of spatial autocorrelation. PLoS ONE 8 (7):e68336. 10.1016/j.jtrangeo.2014.04.016.
———. 2015. A new methodology of spatial cross-correl- Litman, T. 2002. Evaluating transportation equity. World
ation analysis. PLoS ONE 10 (5):e0126158. Transport Policy & Practice 8 (2):50–65.
Church, A., M. Frost, and K. Sullivan. 2000. Transport Lucas, K. 2012. Transport and social exclusion: Where are
and social exclusion in London. Transport Policy 7 (3): we now? Transport Policy 20:105–13. doi:10.1016/j.tran-
195–205. doi: 10.1016/j.tranpol.2012.01.013. pol.2012.01.013.
Clewlow, R. R., and G. S. Mishra. 2017. Disruptive trans- McGrath, F. 2015. The demographics of Uber’s U.S.
portation: The adoption, utilization, and impacts of users. GlobalWebIndex. Accessed July 1, 2017. https://
ride-hailing in the United States. Report UCD-ITS- www.globalwebindex.net/blog/the-demographics-of-
RR-17-07, Institute of Transportation Studies, Ubers-us-users.
University of California, Davis. Murray, A. T., R. Davis, R. J. Stimson, and L. Ferreira.
Cramer, J., and A. B. Krueger. 2016. Disruptive change in 1998. Public transportation access. Transportation Research
the taxi business: The case of Uber. The American Part D: Transport and Environment 3 (5):319–28.
Economic Review 106 (5):177–82. doi:10.1257/ Nie, Y. 2017. How can the taxi industry survive the tide of
aer.p20161002. ridesourcing? Evidence from Shenzhen, China.
Delbosc, A., and G. Currie. 2011. Using Lorenz curves to Transportation Research Part C 79:242–56. doi:10.1016/
assess public transport equity. Journal of Transport j.trc.2017.03.017.
Geography 19 (6):1252–59. Odland, J. 1988. Spatial autocorrelation. London: Sage.
Demetsky, M., and B. Lin. 1982. Bus stop location and Rayle, L., D. Dai, N. Chan, R. Cervero, and S. Shaheen.
design. Transportation Engineering Journal 108 (4): 2016. Just a better taxi? A survey-based comparison
313–27. of taxis, transit, and ridesourcing services in San
Feigon, S., and C. Murphy. 2016. Shared mobility and the Francisco. Transport Policy 45:168–78. doi:10.1016/
transformation of public transit. Transportation Research j.tranpol.2015.10.004.
Board Research Report. Washington, DC: The National Schor, J. 2016. Debating the sharing economy. Journal
Academies Press. doi:10.17226/23578. of Self-Governance and Management Economics 4 (3):7–22.
Frenken, K., and J. Schor. 2017. Putting the sharing econ- Shaheen, S., and N. Chan. 2016. Mobility and the sharing
omy into perspective. Environmental Innovation and economy: Potential to facilitate the first- and last-mile
Societal Transitions 23:3–10. Accessed July 1, 2017. doi: public transit connections. Built Environment 42 (4):
10.1016/j.eist.2017.01.003. 573–88. doi:10.2148/benv.42.4.573.
Ge, Y., C. R. Knittel, D. MacKenzie, and S. Zoepf. 2016. Shaheen, S., A. Cohen, and I. Zohdy. 2016. Shared
Racial and gender discrimination in transportation net- mobility: Current practices and guiding principles.
work companies. Working Paper No. 22776, National Report, U.S. Department of Transportation, Federal
Bureau of Economic Research, Cambridge, MA. Highway Administration, Washington, DC. Accessed
Gehrke, S. R., A. Felix, and T. Reardon. 2018. Fare April 27, 2018. https://ops.fhwa.dot.gov/publications/
choices: A survey of ride-hailing passengers in Metro fhwahop16022/fhwahop16022.pdf.
Boston. MAPC Research Brief. Accessed June 21, 2018. Smart, R., B. Rowe, A. Hawken, M. Kleiman, N.
http://www.mapc.org/wp-content/uploads/2018/02/ Mladenovic, P. Gehred, and C. Manning. 2015. Faster
Fare-Choices-MAPC.pdf. and cheaper: How ride-sourcing fills a gap in low-
Gl€oss, M., M. McGregor, and B. Brown. 2016. Designing income Los Angeles neighborhoods. Los Angeles, CA:
for labour: Uber and the on-demand mobile workforce. BOTEC Analysis Corporation. Accessed July 1, 2017.
In 2016 CHI conference on human factors in computing sys- http://botecanalysis.com/wpcontent/uploads/2015/07/
tems, Santa Clara, CA, USA, 7–12 May 2016, 1632–43. LATS-Final-Report.pdf
San Jose, CA. doi:10.1145/2858036.2858476 Smith, A. 2016. Shared, collaborative and on demand: The new
Hawas, Y. E., M. N. Hassan, and A. Abulibdeh. 2016. A digital economy. Washington, DC: Pew Internet &
multi-criteria approach of assessing public transport American Life Project.
accessibility at a strategic level. Journal of Transport Smith, P. N., and C. J. Taylor. 1994. A method for the
Geography 57 (2):19–34. rationalisation of a suburban railway network.
Henao, A. 2017. Impacts of ridesourcing—Lyft and Transportation Research Part A: Policy and Practice 28 (2):
Uber—on transportation including VMT, mode 93–107.
replacement, parking, and travel behavior. PhD thesis, Wang, M., and L. Mu. 2018. Spatial disparities of Uber
University of Colorado, Boulder. accessibility: An exploratory analysis in Atlanta, USA.
Hughes, R., and D. MacKenzie. 2016. Transportation net- Computers, Environment and Urban Systems 67:169–75.
work company wait times in greater Seattle, and rela- Williams, R. 2017. How Uber works together with
tionship to socioeconomic indicators. Journal of Southern California’s growing rail network. Accessed
Transport Geography 56:36–44. doi:10.1016/ September 21, 2017. https://medium.com/uber-under-
j.jtrangeo.2016.08.014. the-hood/how-uber-works-together-with-southern-cali-
Jin, S. T., H. Kong, R. Wu, and D. Z. Sui. 2018. fornias-growing-rail-network-fa8b1aba03b4.
Ridesourcing, the sharing economy, and the future of Wu, C., and A. T. Murray. 2005. Optimizing public transit
cities. Cities 76:96–104. doi: 10.1016/j.cities.2018.01.012. quality and system access: The multiple-route, maximal
Kaplan, S., D. Popoks, C. G. Prato, and A. Ceder. 2014. covering/shortest-path problem. Environment and
Using connectivity for measuring equity in transit Planning B: Planning and Design 32 (2):163–78.
Uber, Public Transit, and Urban Transportation Equity 15
SCARLETT T. JIN is a PhD Candidate in the where wij could be defined as
Department of Geography at The Ohio State
dij1
University, Columbus, OH 43210. E-mail: wij ¼ Pn Pn ; (A.4)
1
jin.760@osu.edu. Her research interests include j¼1 i¼1 dij
urban development in China, shrinking cities, the
where dij is the Euclidean distance between block i
sharing economy, and shared mobility.
and j, and we define dij ¼ 0 when i ¼ j.
HUI KONG is a Postdoctoral Associate at JTL
Urban Mobility Lab of Singapore–MIT Alliance for SCI denotes the global spatial cross-correlation
Research and Technology (SMART), Singapore index. The SCI is the correlation coefficient
138602. E-mail: kong.hui@smart.mit.edu. She between x and y through the spatial distances and
obtained her PhD from the Department of geographical elements. It represents a correlation
Geography at the Ohio State University. Her with distance effect. It falls between 1 and 1. By
research interests include shared mobility, smart analogy with the improved formula of Moran’s I for
transportation, GIScience, and urban design. spatial autocorrelation (Chen 2013), the measure-
ment of SCI is
DANIEL Z. SUI is an Arts & Sciences
SCI ¼ xT Wy ¼ yT Wx: (A.5)
Distinguished Professor in the Department of
Geography at The Ohio State University,
Columbus, OH 43210. E-mail: sui.10@osu.edu. His
research interests include GIScience, location-based
social media, convergence research, and the emerg-
ing platform economy. Appendix B: Spatial Autocorrelation
Index
Appendix A: Spatial Cross-
Correlation Index Moran’s I is calculated to measure the spatial auto-
correlation. The formula is as follows (Chen 2013):

The spatial cross-correlation index (SCI) is calcu- I ¼ xT Wx (B.1)


lated through the following steps. First, the standar-
dized values of the two measurements are obtained. The Moran’s I of all of the indicators used earlier is
Suppose there are n elements (e.g., CBG) in the
calculated (Table B.1). The threshold value of
study area (New York City) and there are two
Moran’s I has been regarded as either zero (Chen
measurements whose correlation we need to quan-
2013) or 1/(n  1) (Odland 1988). Therefore, all
tify: the number of Uber pickups and scheduled
of the indicators have weak spatial autocorrelations.
public transit pickups. A pair of vectors of the two
measurements can be defined as
Because Uber pickups, income, and minority
X ¼ ½x1 ; x2 ; . . . xn T ; Y ¼ ½y1 ; y2 ; . . . yn T ; (A.1) have relatively stronger autocorrelations (but are
still too weak to influence our conclusions), we cal-
where xi and yi are the measurements of the ith
element (i ¼ 1, 2, … , n), and T denotes the trans- culate two additional indexes to distinguish the spa-
pose. The pair of standardized scores of X and Y is tial and nonspatial effects in the correlations.

Xlx Y ly
x¼ ;y ¼ ; (A.2)
rx ry
Table B.1 Spatial autocorrelation index
where lx and ly represent the means of X and Y,
Weekday Weekend
and rx and ry denote the population standard devia-
Public transit 0.0212 0.014
tions of X and Y. Uber 0.0767 0.1261
Population 0.0134 0.0134
Income 0.1311 0.1311
The indexes of spatial cross-correlation also Minority 0.1613 0.1613
depend on a distance matrix based on the spatial
distance between each pair of elements. The n-by-n Note: Uber and taxi data are for 17 September (weekday) and
20 September (weekend). Public transit includes bus, subway,
spatial weights matrix could be defined as and rail.
W ¼ ½wij nn ; (A.3)
16 Volume 0, Number 0, Month 2019

Table C.1 The direct and indirect correlations


Weekday Weekend

Correlation Direct correlation Indirect correlation Direct correlation Indirect correlation


Uber and income 0.2435 0.0701 0.2841 0.0893
Uber and minority 0.1173 0.0598 0.1457 0.0793

Note: Uber data are for 17 September (weekday) and 20 September (weekend).
Statistically significant at the 0.001 level.

Appendix C: Distinguishing Direct and geographical elements. It is the direct correlation


Indirect Correlation in Pearson’s without distance effects. Rp can simply be calculated
Correlation Index using R0 and SCI:
Rp ¼ R0  SCI (C.1)
The Pearson correlation coefficient (R0 ) includes
two parts: a correlation based on the distance decay Due to the weak spatial autocorrelation in the num-
effect and a correlation without distance effects. ber of Uber pickups, income, and the percentage of
minority populataion, we calculate the SCI and Rp
The global spatial cross-correlation index, SCI, between Uber and income and Uber and minority to
denotes the correlation between two variables distinguish the direct (without spatial decay) and
through the spatial distances and geographical ele- indirect (with spatial decay) correlation between them
ments. It represents an indirect correlation with dis- (Table C.1). Results indicate a weak spatial cross-cor-
tance effects. The partial correlation coefficient Rp relation between either Uber and income or Uber
represents the correlation between the two varia- and minority, whereas the direct correlations with
bles, which is free of the spatial distance and other spatial effects are relatively strong and significant.

You might also like