Content uploaded by Jinbao Zhao
Author content
All content in this area was uploaded by Jinbao Zhao on Apr 01, 2018
Content may be subject to copyright.
Ridership and effectiveness of bikesharing: The effects of urban
features and system characteristics on daily use and turnover rate
of public bikes in China
Jinbao Zhao
a,b,
n
, Wei Deng
a
, Yan Song
b
a
School of Transportation, Southeast University, No. 2 Si-pai-lou, Nanjing 210096, China
b
Department of City and Regional Planning, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, United States
article info
Available online 28 June 2014
Keywords:
Bikesharing
Ridership analysis
Turnover rate
Personal credit cards
Universal cards
Partial least squares (PLS) regression
abstract
As a pinnacle of green transportation with transit attributes, bikesharing has become particularly popular
since the mid-2000s. Two crucial questions for the success of bikesharing adoption are how many riders
can bikesharing attract, and what influences its effectiveness. To shed light on answers to these questions,
this paper models the impacts of urban features and system characteristics on bikesharing daily use and
turnover rate, using data constructed on 69 bikesharing systems in China. Prior tomodeling, we provide an
overview of bikesharing adoption in China, describing why they have been adopted, how they have
matured, and how they have expanded. Results from data regression and comparison indicate that
bikesharing ridership and turnover rate tend to increase with urban population, government expenditure,
the number of bikesharing members and docking stations, whilst the number of public bikes shows
significant but adverse signs in impacting bikesharing ridership and turnover rate. Data comparison shows
that, to pursue an ideal bikesharing turnover rate in most Chinese cities, the bike-member (supply-
demand) ratio should be better controlled within 0.2. Moreover, this study suggests that personal credit
cards (allowing bikesharing members to pay “personal credit”rather than money if they do not return
public bikes within the free use hours) and universal cards (integrating bikesharing systems into other
urban transit systems through the use of a rechargeable smart card that can covera range of payments and
trips) can significantly raise bikesharing daily use and turnover rate. We recommend that bikesharing
operators and transit agencies take the supply-demand thresholds and the adoption of personal credit
cards and universal cards into consideration in the future bikesharing operation and development policy.
&2014 Elsevier Ltd. All rights reserved.
1. Introduction
Serving as an alternative of urban transit systems, public
bikesharing has developed and spread into a new form of mobility
across the globe since the mid-2000s (Parkes et al., 2013).
Bikesharing is viewed as an economic, efficient, and healthy means
of navigating through dense urban environments (O’Brien et al.,
2014), and it provides a variety of pickup and drop-off locations,
enabling an on-demand, very low emission form of mobility
(Parkes et al., 2013). Bikesharing users can access public bikes on
an as-needed basis without the bearing costs of bike ownership
(Shaheen et al., 2010). In addition, by integrating with public
transportation and other alternative modes, bikesharing provides
a low-carbon solution to the “last mile”challenge of urban transit
systems (Shaheen et al., 2010).
Whilst bikesharing is a relatively new form of transport in
urban areas, adoption of this evolving transit model has become
particularly popular in recent years (Shaheen et al., 2010, 2013;
O’Brien et al., 2014). There are more than 600 bikesharing systems
currently operating worldwide (DeMaio and Meddin, 2014;
Christensen and Shaheen, 2014; Hughes, 2014) and a growing
number of cities are planning to launch bikesharing to increase
bicycle use (García-Palomares et al., 2012). This growth is catching
increasing attention in planning circles in its own right.
For bikesharing's early adoption and sustainable development,
a crucial issue is the recognition of the factors affecting its
ridership and effectiveness. A bikesharing system with few riders
and low turnover rate implies a poor investment economically,
environmentally, and socially. A better understanding of factors
driving its ridership and effectiveness can help inform future
adoption policy and improve the performance of existing systems.
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/tranpol
Transport Policy
http://dx.doi.org/10.1016/j.tranpol.2014.06.008
0967-070X/&2014 Elsevier Ltd. All rights reserved.
n
Corresponding author at: School of Transportation, Southeast University,
No. 2 Si-pai-lou, Nanjing 210096, China.
E-mail address: jinbaozhao@seu.edu.cn (J. Zhao).
Transport Policy 35 (2014) 253–264
To shed light on this issue, in this study, we evaluate how urban
features and system characteristics impact bikesharing ridership
(daily use) and effectiveness (turnover rate), using data con-
structed on 69 bikesharing systems in China. The remainder of
this paper includes four additional sections. The following section
gives an overview of recent research on bikesharing, in particular,
the studies since the year of 2010. The bikesharing development in
China and the research design for this study are presented in
Section 3.Section 4 describes the modeling results and findings.
Finally, Section 5 concludes with policy implications of, and
development revelations from, the ridership and effectiveness
perspective.
2. Background
The first public-use bikesharing can be found in Amsterdam
(the Netherlands) as far back as the late 1960s, with the introduc-
tion of the famous “White Bicycles”system (Shaheen et al., 2010).
It became widely recognized in the transportation community
with the pioneering large-scale and third-generation bikesharing
system –Velo’v–launched in Lyon in 2005 (DeMaio, 2009;
Midgley, 2011). Since then, it is becoming increasingly popular in
towns and cities around the world with the growing concerns
about global motorization and the externalities associated with
driving, such as traffic congestion and greenhouse gas emissions.
To date, the existing knowledge of bikesharing is relatively thin
but is growing rapidly with bikesharing's widespread expansion.
Shaheen et al. (2010) analyzed the evolution of bikesharing around
the world. In that study, they discussed bikesharing business
models and lessons learned, highlighting the social and environ-
mental benefits associated with bikesharing. They argued that
while bikesharing is growing worldwide and can help address
many of the concerns about the global climate change, energy
security, and unstable fuel prices, its future demand and long-term
sustainability are still uncertain. More research is needed for a
better understanding of bikesharing's effects, operations, and
business models in light of its reported growth and benefits
(Shaheen et al., 2010).
Interest in bikesharing research has become particularly popular
since the important Shaheen et al. (2010) study. Two years later,
Shaheen et al. (2012) released a key report on bikesharing usage
data and user feedback from detailed interviews with governmental
agencies and bikesharing users in the United States and Canada.
Based on theuser survey (completed in Montreal, Toronto, Washing-
ton DC and the twin cities) with a decent sample size (n¼10,661),
Shaheen et al. (2012) found that the most common bikesharing trip
purpose is work- or school-related (50–56% in the two Canadian
cities and about 38% in the two American cities). Respondents in all
cities indicated that they increased bicycling, whist most of them
indicated that they drive less, as a result of bikesharing. Moreover, a
majority of respondents reported getting more exercise since
becoming a user of bikesharing. At the same time, there is evidence
from Shaheen et al. (2012) that public bikesharing is improving
urban travel connectivity, reducing driving and thus lowering
vehicle emissions.
Because of these benefits, in recent years many cities round the
world show enthusiasms in bikesharing adoption. To explore the
adoption patterns of bikesharing systems, Parkes et al. (2013)
provided an analysis on the diffusion of public bikesharing
systems in Europe and North America. They concluded that
“Europe is still in a major adoption process with new systems
emerging and growth in some existing systems”, while “in North
America, the adoption process is at an earlier stage and is gaining
momentum”. They declared that the notable and successful
systems in Paris, Lyon, Montreal, and Washington DC have sparked
great interest in bikesharing in Europe and North America, yet one
of the most potential markets for bikesharing –Asia –was missed
in their study. Considering automobile in most Asian developing
countries, such as China, is still less popular but shows a rapid
growth trend in comparison with most of European and North
American countries, it is useful to outline the adoption patterns of
bikesharing in a developing context (such as China in this paper)
to complement the research of Parkes et al. (2013).
A recent study by O'Brien et al. (2014) took a global view of
bikesharing characteristics by analyzing data from 38 systems
located in Europe, the Middle East, Asia, Australasia and the
Americas. Through the analysis of the variation of occupancy rates
over time and comparison across the system's extent, O'Brien et al.
(2014) proposed a classification of bikesharing systems, based on
the geographical footprint and diurnal, day-of-week and spatial
variations in occupancy rates, which laid foundations for the
analysis of larger scale bikesharing systems.
Researchers also conducted bikesharing studies at the urban
level rather than global view. For example, Jensen et al. (2010)
analyzed 11.6 million journeys of the Vélo'v in Lyon, constructing a
map showing the likely flows of the bicycles across the city. Lathia
et al. (2012) assessed the impacts of the “open policy”(that allows
casual users to use shared bikes with a debit or credit card) of the
London shared bicycle scheme, finding that open-access to the
system correlating with greater usages. García-Palomares et al.
(2012) proposed a GIS-based method to calculate the spatial
distribution of the potential demand for bikesharing trips in
Madrid, locating stations using location–allocation models, which
is of great use for managing the redistribution of bicycles among
the stations. Jäppinen et al. (2013) modeled the potential effect
of shared bicycles on public transport travel times in Greater
Helsinki. They found that the adoption of bikesharing can reduce
public transportation travel times in the study area, on average by
more than 10%.
Recent bikesharing research is also found, for example, in
Kaltenbrunner et al. (2010),Lin and Yang (2011), and Chemla
et al. (2012). These studies addressed bikesharing's issues form
different concerns, such as prediction of available public bikes
(Kaltenbrunner et al., 2010), bikesharing planning strategic (Lin
and Yang, 2011), and system rebalancing (Chemla et al., 2012;
Raviv et al., 2013; Nair and Miller-Hooks, 2014). Regarding one of
the most crucial issues as mentioned as the ridership and effec-
tiveness analysis, related studies are quite few but can also be
found. Particularly, a recent ITDP report by Gauthier et al. (2013)
gave a global analysis that looks at scale and success factors
driving bikesharing development. In their report, Gauthier et al.
(2013) argued that turnover is critical to a successful bikesharing
system, which is ideal to be four to eight daily uses per bike. Good
station locations and sufficient station coverage are critical to
ensuring that the system will have high usage and turnover.
Generally, a quality system needs 10–16 stations for every square
kilometer (approximately 300 m between stations). In addition,
there should be 10–30 bikes available for every 1000 residents
within the coverage area. A recent presentation by Hughes (2014)
further confirmed these useful findings.
The increasing trend in bikesharing research in recent years
indicates a bikesharing boom is taking place. As a continuation for
and complement of these published materials, we make an effort
to model and analyze the effects of urban population, government
expenditure, bikesharing demand (the number of bikesharing
members) and supply (the number of docking stations and public
bikes), and operation policy (using personal credit cards or not,
providing 24 h service or not, adopting universal cards or not) on
bikesharing daily use and turnover rate. The empirical analysis is
based on available data from 69 bikesharing systems in China, one
of the fastest growing markets for bikesharing across the globe.
J. Zhao et al. / Transport Policy 35 (2014) 253–264254
3. Research context and design
3.1. Research context
China, once well-known as the “Kingdom of Bicycles”in the
1970s, is now suffering the externalities associated with rapid
motorization, whereas in and Western Europe and North America,
a shift towards bicycling is occurring. From 2000 to 2012,
the private autos in China increased by more than 10 times from
6.25–88.38 million (National Bureau of Statistics of China, 2013).
The increasing trend will continue in the near foreseeable future.
To address the challenges caused by the ongoing motorization and
urbanization, such as traffic congestion and air pollution, China's
urban planners and decision-makers have gradually shifted their
emphasis from providing additional road space for driving to
examining the need for more sustainable transportation strategies.
In China, “Transit Priority”that allows public transportation to
be a more attractive and viable option for commuters has been
proposed as a national strategy to relieve people's increasing
reliance on automobiles (State Council, 2012).
As an implementation of this strategy, local governments have
adopted various policies in supporting public transport. Particu-
larly, in recent years interest in initiating, leading and funding
public bikes
1
has spread across the country, from the capital
Beijing to a small countryside Yonglian located at Zhangjiagang.
In fact, as early as May 2008, the city of Hangzhou launched the
first information technology-based public bikesharing program in
China, which was fully found by local governments and operated
by a government-owned company –the Hangzhou Public Bicycle
Operation Management Corporation. Upon its launch, users could
use their citizen cards or transit smart cards to register into
bikesharing system. After putting down a deposit (200 RBM
2
at
that time) on their cards, they could rent a public bike from a
bikesharing station. The first hour of use is free; this is followed by
incremental pricing where users pay an additional 1 RMB for the
second hour, 2 RMB for the third hour, and 3 RMB after that. If the
bike is not returned or lost, the user loses the deposit (Hangzhou
Public Bicycle, 2013).
One of the bright spots of Hangzhou bikesharing is that its
bikesharing system is integrated into other public transport
systems through the use of citizen cards or transit smart cards
(this policy will be referred as the adoption of “universal cards”in
the following discussion) that can cover a range of payments and
trips. However, most of Hangzhou's bikesharing stations do not
provide 24 h service except the 84 bikesharing stations (about
2.8% of the 2962 docking stations as the end of 2012) that locate
along its two rail transit lines (Hangzhou Public Bicycle, 2014).
After nearly 6 years' development, the Hangzhou Public Bicycle
System has surpassed Vélib' as one of the most famous bikeshar-
ing programs in the world (Shaheen et al., 2011). As of December
2012, the service operated 69,750 bicycles and 2962 fixed stations,
with an average daily use of 25.76 ten thousand and a free charge
rate of 96% (Hangzhou Public Bicycle, 2013).
Another pioneer in developing bikesharing in China is the city
of Wuhan. Only 6 months after Hangzhou, Wuhan launched its
first bikesharing program in November 2008, with 1000 public
bikes at that time. But different from Hangzhou's “government-
arranging”model, Wuhan adopted a “government-leading and
private-participating”model. Wuhan governments also subsidize
bikesharing but partially. They encourage private companies to
operate bikesharing and grant their right of advertising. In addi-
tion, unlike the application approach for a bikesharing card in
Hangzhou, citizens in Wuhan could use their local ID (Resident
Identity Cards, student cards, etc.) to apply for a “personal credit
card”
3
to register into the system. Bikesharing members did not
need to put down a deposit on their personal credit cards at the
early adoption. This “zero-cost”policy evoked a member boom in
Wuhan, but also caused some challenges. For example, as of July
2011, the number of personal credit card holders was about
1 million, whilst the number of public bikes was less than 0.05
million (Cui, 2011). Many members could not rent a bike, in
particular, during peak hours. To address this, Wuhan's bikeshar-
ing operators adopted the Hangzhou model since July 2011. Whilst
this policy transformation relieved the tight supply of public bikes,
yet lead to a long depression of bikesharing use in Wuhan.
However, some cities, including Shanghai, Chengdu, and
Suzhou, still use personal credit cards yet improve this policy
from several aspects: (1) members need to put down a deposit on
their cards, (2) members should return public bikes within 1 h (in
comparison with 4 h upon the launch of the bikesharing program
in Wuhan) and if not, (3) users are penalized 5–10 personal credit
points (in comparison with only 1 RMB for universal card holders).
Personal credit cards play a role in encouraging residents to apply
a bikesharing card and use public bikes, this policy, however,
needs stronger financial support from local governments.
Compared to personal credit cards policy, most cities in China
follow the Hangzhou model: (1) users are required to provide
resident identity cards (usually Resident Identity Cards or Hukou)
to register into the system, (2) users are required to put down a
deposit (usually 200–300 RMB) on their bikesharing cards or
universal cards and, (3) the bicycle usage is limited by time
(usually the use for the first 1 h is free). Following this model,
the diffusion of public bikesharing systems in China began to
accelerate, in particular, since the year of 2010 (Fig. 1). Between
2010 and 2012, around 20 new systems were introduced each year,
and the curve reached its steepest gradient in 2013 with 29 new
systems being introduced during the first 10 months. Based on
available data, as of November 2013, around 98 cities in China
launched bikesharing programs. The service operated approxi-
mately 419,800 public bikes, 15,603 bikesharing stations, 4,357,382
bikesharing members, with a daily use of about 1,586,228.
3.2. Research design
It is foreseeable that the adoption of bikesharing will be
continually ongoing in China, whilst bikesharing ridership and
turnover performance and the potential influence factors have
rarely been examined. To address this issue, this paper is designed
to explore and examine the factors that might affect bikesharing
daily use and turnover, particularly focusing on the variables
measuring urban features and system characteristics. We endea-
vored to collect data for all the 98 cities that have launched
bikesharing systems as we knew, whilst 29 of them were missed in
the data modeling. This is mainly because of the data unavail-
ability, since some key indicators for the 29 systems, such as
ridership, the number of bikesharing members, etc., could not be
1
Bikesharing systems go by a variety of names around the world: “bicycle
sharing”or simply “bike-share”in North America, “cycle hire”in the United
Kingdom and “public bike”in China (Gauthier et al., 2013). Bikesharing in China
is more well-known as “public bike”because most bikesharing systems are fully
found and operated by public agencies (such as the Department of Transportation
and Urban Management Bureau) rather than by private operators.
2
RMB, or Renminbi, is the currency of China. At the time of writing, the
exchange rate is RMB 1 to USD 0.164 approximately.
3
A“personal credit card”is different from a bank credit card, since there is no
money but personal credit points in the former. Upon its launch, members holding
Wuhan's personal credit cards can rent public bikes for free at the first four hours. If
the bike is not returned with four hours, users are penalized by one bad credit
point. If their bad credit points are three or more, they cannot rent a public bike in
the future.
J. Zhao et al. / Transport Policy 35 (2014) 253–264 255
obtained. Besides estimating models of factors that contribute to
bikesharing daily use and turnover rate by adopting ordinary least
squares (OLS) and partial least squares (PLS) regression models
(Tobias, 1999), we also make comparison of system daily use and
turnover rate versus various significant variables, such as popula-
tion, bikesharing members, docking stations, public bikes, etc.
Data related to bikesharing system and urban characteristics
were collected for the 69 cities –including 3 central municipa-
lities, 43 prefecture-level cities, and 23 county-level cities –from
Urban Statistical Yearbook of China (National Bureau of Statistics of
China, 2012), bikesharing operators, online documents, etc. Infor-
mation on the key characteristics of bikesharing systems in China,
including the opening month, operation days, turnover rate, daily
use, the number of docking stations and public bikes, are listed in
Appendix A. Table 1 summarizes the dependent and independent
variables adopted in this study. As Table 1 shows, bikesharing daily
use
4
averaged 27,551 passengers, with a range from 300 in
Jiayuguan city to 257,600 in Hangzhou city. Turnover rate (TR),
which is adopted to measure the effectiveness degree of public
bikes and is calculated as
TR
i
¼U
i
B
i
ð1Þ
where TR
i
is the turnover rate of public bikes (per bicycle per day)
of city i,U
i
is the daily use of public bicycles of city i, and B
i
is the
number of public bikes of city i. The higher the rate, the more
effective a system is. Of the 69 systems, turnover rate averaged
4.2 times per bike per day, from a low of 0.7 in Jiaxing city to a
high of 9.5 in Taiyuan city.
Regarding the independent variables, potential factors that
might affect bikesharing daily use and turnover rate fall into three
categories in this study: (1) urban features; (2) system character-
istics and; (3) composite indictors.
3.2.1. Urban features
Urban features consist of combinations of urban socioeconomic
and demographic variables. Four independent variables –population,
job, income per capita, and government expenditure –are adopted.
Population and job are often first chosen as variables in transit
ridership analysis since they may influence ridership in three
primary ways: (a) by increasing the number of potential origins
and destinations served by the system; (b) by making it more likely
that someone living near transit has attractive transit destinations;
and (c) by increasing the congestion and parking costs of transit's
primary competitor, the car (Guerra and Cervero, 2011). Government
expenditure is expected to have a positive impact on bikesharing
daily use and turnover rate since most of public bikesharing
programs heavily rely on government subsidies in China. The
government expenditure adopted in this study is the overall govern-
ment spending on urban social affairs, since most empirical cities in
this study did not release valid data on the cost of bikesharing
programs. Ignoring the difference between overall urban public
finance and bikesharing program cost might erode the significance
of the variable.
3.2.2. System characteristics
Considering the availability of good quality and readily acces-
sible data, we account for bikesharing characteristics from con-
sideration of bikesharing demand, supply, and operation policy.
In discussing system ridership and effectiveness, it is important
to consider the impacts of bikesharing demand and bikesharing
supply. We take advantage of the registration policy adopted by
most bikesharing systems in China to obtain the number of
bikesharing members to examine the impact of bikesharing
demand on system ridership and effectiveness. For bikesharing
supply, the number of docking stations and public bikes are likely
to determine supply levels. From demand and supply side, it is
hypothesized that systems with more members, docking stations,
and public bikes have higher daily use, yet their impacts on
turnover rate are still to be seen. Moreover, since the time that
cities launched bikesharing programs and released bikesharing
related data are different from city to city, a variable reflecting how
long the system has operated –the days between the system's
opening day and the data releasing day –is selected to accom-
modate the differences.
For bikesharing operation policy, as mentioned, Chinese cities
vary in policy adoption. Some cities penalize personal credit whilst
some penalize money if bikesharing users do not return public
bikes within the free use time. Twenty-four hour service has been
realized in some cities while some do not open their bikesharing
systems for 24 h. In addition, some cities have adopted universal
cards that can cover a range of transit payments and trips while
some have not. To accommodate these differences, three dummy
variables: (1) using personal credit cards or not (1–0), (2) providing
24 h service or not (1–0) and, (3) using universal cards or not (1–0)
are included in this study.
3.2.3. Composite indictors
Two composite indictors that reflect the ratios between deposit
and income level, and between penalty and income level are
assumed to have negative effects on bikesharing daily use and
turnover rate. The first composite indictor is introduced as
DIR
i
¼D
i
I
i
ð2Þ
where DIR
i
is the ratio between deposit and income level of city i,
D
i
is the deposit of bikesharing of city i, and I
i
is the income per
capita of city i. Moreover, in some Chinese cities, bikesharing users
are charged penalty if they do not return public bikes within the
free use hours, and the penalty vary from city to city. By
compositely considering a city's income level, the ratio between
0
10
20
30
40
50
0
20
40
60
80
100
May Sep Jan May Sep Jan May Sep Jan May Sep Jan May Sep Jan May Sep
Increase number of cities Cumulative number of cities
2009 2011 20132010 2012
Year
Month
Cumulative number of cities
Increase number of cities by month
Fig. 1. Diffusion of bikesharing systems in China.
4
Daily use includes two types of users: membership users and occasional
users. Most of bikesharing users in China are membership users from local
residents because of the registration policy. There are some occasional users but
the proportion is often quite low. For example, in Xi’an, this proportion was only 3%
(see Zhao, 2012). This is because the deposit for occasional users is usually higher
than that for local residents (700 RMB vs. 200 RMB in Xi’an, and 1200 RMB vs. 200
RMB in Zhuzhou). The proportion of occasional users varies from city to city but
most cities do not count it separately. Ignoring the proportion of occasional users
may erode the significance of the variables, particularly the number of bikesharing
members.
J. Zhao et al. / Transport Policy 35 (2014) 253–264256
penalty fare and income level is introduced as
PIR
i
¼P
i
I
i
ð3Þ
where PIR
i
is the penalty fare-income level ratio of city i,P
i
is the
penalty of bikesharing of city i,andI
i
is the income per capita of city i.
3.2.4. Possible missing variables
The present study missed several variables that may affect
bikesharing ridership and effectiveness. Neighborhood environ-
ments, personal characteristics, and service attributes (rebalan-
cing, maintenance, etc.) might influence bikesharing daily use and
turnover rate significantly. However, due to data limitations, we
are unable to separate out the various ways in which the selected
variables affect bikesharing ridership and effectiveness. Even
though, some useful findings can be drawn from data regression
and comparison.
4. Result analysis
Bikesharing daily use may vary as a function of urban socio-
economic and demographic, system scale, and operation policy.
Cities with more population and sufficient expenditure tend to
drive bikesharing development and thus the daily use. Systems
with more members, public bikes, and stations may have higher
ridership. We first probe this hypothesis by adopting ordinary least
squares (OLS) regression models. Prior to modeling, what can be
predictable is that some independent variables adopted in this
study are likely to be highly collinear. For example, the more the
docking stations, often the more the public bikes, and usually
the more bikesharing members are. OLS can distinguish multi-
collinearity by computing variables' variance inflation factors (VIF)
and a general rule is that the VIF should not exceed 10 (Kutner et al.,
2004). Multi-collinearity can also be tested by generating Pearson
correlation coefficients for all pairwise combinations and the
danger level for independent variables' multi-collinearity is 0.7
(Clark and Hosking, 1986). Anyway, some independent variables
have to be omitted even they may have significant influence on
bikesharing daily use and turnover rate. In this case, partial least
squares (PLS) regression models, which is a useful tool for con-
structing predictive models when the factors are many and highly
collinear (Tobias, 1999), are adopted by the authors to contrast the
modeling results of OLS. A statistic summarizing the contribution a
variable makes to the PLS model is the variable importance for
projection (VIP) and a value less than 0.8 is considered to be “small”
for the VIP (Wold, 1994).
4.1. Results from daily use analysis
We first provide the daily use modeling results by adopting OLS
and PLS models in Table 2. The first OLS model is initially applied
to the entire set of independent variables. Among the variables in
the first OLS model, variables with a VIF smaller than 10 are then
applied to the second OLS model. The third model is established by
using PLS and is applied to the entire set of independent variables.
Table 2 confirms many previous expectations. It is firstly worth
noting that four independent variables, including jobs, government
expenditure, and the number of bikesharing members and public
bikes are highly correlated with one or more of the others. After
excluding these four variables, the VIFs of all the other independent
variable are smaller than4 in the second OLS model, and four variables
are turned out to be significantly associated with bikesharing daily use
in this model. It is found that bikesharing daily use tends to increase
significantly with urban population and the number of docking
stations. All else being equal, the results imply an increase of 31
bikesharing passengers for every 10 thousand population, and an
increase of 121 passengers for each docking station. Moreover, both
the adoptions of personal credit cards and universal cards tend to
increase bikesharing daily use significantly.
Modeling results from the second OLS model generally make
sense, yet it dismisses some variables that might impact ridership
significantly. Indeed, we feel that government expenditure is exogen-
ous to bikesharing daily use in China since bikesharing adoption and
development heavily rely on government investments. In addition,
from the demand and supply considerations, the number of
Table 1
Summary of variables.
Variables Mean SD Min Median Max Expected sign
Dependent variables
Daily use
a
27,551 52,185 300 600 0 257,600
Turnover rate 4.2 2.1 0.7 4.0 9.5
Independent variables
Urban features
b
Population (10 thousand) 164.8 240.4 9.8 101.8 1350.6 þ
Jobs (10 thousand) 54.8 101.9 2.3 21.1 614.9 þ
Income per capita (RMB) 43,882 9846 25,946 42,60 0 77,146 ?
Government expenditure (million RMB)
c
28,297 62,110 1218 10,106 381,695 þ
System characteristics
a
Number of members 70,368 163,899 500 12,064 986,00 0 þ
Number of docking stations 196 351 7 82 2435 þ
Number of public bikes 5883 12,019 150 2000 75,00 0 ?
Operation days since the system opened 427 348 93 367 1560 ?
Using personal credit cards or not (1–0) 0.16 0.37 0 0 1 þ
Providing 24 h service or not (1–0) 0.57 0.50 0 1 1 þ
Using universal cards or not (1–0) 0.20 0.41 0 0 1 þ
Composite indictors
Deposit-income per capita ratio (10
%3
) 4.79 3.30 0 4.92 18.26 –
Penalty (after the first free use hours)-income per capita ratio (10
%4
) 0.19 0.10 0 0.22 0.38 –
a
Data from various sources, including local bikesharing agencies, online documents, and websites (such as, http://www.hzsggzxc.com/ (Hangzhou Public Bicycle), http://
www.bjysj.gov.cn/bjggzxc/ (Beijing Public Bicycle), http://zxc.gzzc.com.cn/ (Guangzhou Public Bicycle), etc.) to name very a few here.
b
Data from China City Statistical Yearbook 2012 (National Bureau of Statistics of China, 2012).
c
Government expenditure on urban public affairs.
J. Zhao et al. / Transport Policy 35 (2014) 253–264 257
bikesharing members and public bikes can be significant factors
determining bikesharing use. All the three factors, however, are
excluded in the second OLS model because of multi-collinearity. PLS
regression, therefore, is developed to extract these latent factors,
which can account for as much of the manifest factor variation as
possible while modelling the responses well (Tobi as , 1999).
As PLS modeling results show, in addition to the four significant
variables in the second OLS model, another three independent
variables have a VIP larger than 1.0: government expenditure, the
number of bikesharing members, and the number of public bikes.
These variables erode the coefficients of the four significant
variables in the second OLS model yet turn out to be important
variables in generating bikesharing ridership.
Fig. 2 shows the comparisons of bikesharing daily use versus
five variables that affect bikesharing daily use significantly in
Table 2. Whilst there is clearly variation among cities of bikeshar-
ing daily use, the figure confirms most of the results in Table 2.
Generally, bikesharing daily use tends to increase with a city's
population, government expenditure, bikesharing members, dock-
ing stations, and public bikes. It is further noted that bikesharing
daily use tends to increase systematically with the number of
public bikes in the comparison analysis, with a linear R
2
value of
0.849. The regression results from Table 2 and the comparisons of
Fig. 2 provide bikesharing operators and transit agencies direct
formulas to estimate bikesharing daily use roughly.
4.2. Results from turnover rate analysis
To find suitable formulas to model the influence of indepen-
dent variables on bikesharing turnover rate, we first plot a range of
cities based on cluster analysis and comparisons between turnover
rate and the selected independent variables. By taking various
regressions, including linear regression, logarithm regression, and
index regression, we find the logarithm regression fits the com-
parisons superiorly in comparison with other regression methods
(Fig. 3), reflected by its relatively higher R
2
values.
Fig. 3 shows the linear regression lines (cluster analysis) and
logarithm regression lines (scatter analysis) of the turnover rates
on the variables measuring bikesharing demand (bikesharing
members) and supply (docking stations and public bikes). Regard-
ing single factor analysis, it is found that bikesharing turnover rate
tends to increase with bikesharing members and docking stations,
as well as public bikes. Whilst the results make sense, single factor
comparison does not consider other factors' impacts. By taking
logarithm of variables measuring urban features and bikesharing
demand and supply, we estimate the impacts of selected inde-
pendent variables on bikesharing turnover rate, using a OLS model
by deleting invariables with a Pearson correlation coefficient
larger than 0.7 and again, using a PLS model that is applied to
the entire set of independent variables. The modeling results are
shown in Table 3.
It is first noted that the four independent variables deleted in
Table 3 by adopting Pearson correlation coefficient test are the
same as by using VIF of 10 as the cut off value in Table 2. In
addition, we find that the dummy variable indicating a system
providing 24 h service or not is significant to be associated with
bikesharing turnover rate, whilst this variable does not turn out to
be related to bikesharing daily use significantly in Table 2.
By adopting the PLS model, more independent variables
are found to be important in explaining bikesharing turnover rate.
The results indicate that population, government expenditure, the
number of bikesharing members and docking stations, and the
Table 2
OLS and PLS regression models of bikesharing daily use.
Variable OLS model with all variables OLS model with variables whose
VIF is smaller than 10.
PLS model
Coeff. p-value VIF
a
Coeff. p-value VIF
a
Coeff. B'
b
VIP
c
(Constant) %13613.295 0.9018 0 %23365 0.4638 0 %15573.287 0
Urban features
Population (10 thousand) 33.697
nn
0.0239 9.713 30.739
nn
0.0160 2.611 23.644
††
0.101 1.109
Jobs (10 thousand) %185.816 0.2272 14.656 41.667 0.081 0.639
Income per capita (RMB) %0.129 0.7725 4.233 %0.110 0.7227 3.668 %0.066
†
%0.073 0.806
Government expenditure (million RBM) 0. 291
nn
0.0346 15.192 0.392
††
0.078 1.032
System characteristics
Number of members 0.056 0.2489 13.951 0.064
††
0.202 1.845
Number of docking stations 55.193
nnn
0.0034 8.864 121.497
nnn
o0.0001 1.688 29.9614
††
0.201 1.852
Number of public bikes 1.633
nn
0.0209 15.025 0.889
††
0.204 1.868
Operation days since the system opened 3.3164 0.6889 1.816 6.860 0.4667 1.657 8.456
†
0.123 0.882
Using personal credit cards or not (1–0) 10120.978
nn
0.0238 6.255 16394.267
nn
0.0142 3.370 12474.332
††
0.041 1.022
Providing 24 h service or not (1–0) 2025.963 0.6678 1.212 3697.595 0.4939 1.121 3336.613 0.031 0.587
Using universal cards or not (1–0) 12128.592
nn
0.0532 1.301 19598.569
nn
0.0295 1.227 14838.563
††
0.065 1.182
Composite indictors
Deposit-income per capita ratio (10
%3
)%287.473 0.9816 4.373 251.603 0.9780 1.716 %236.896 %0.015 0.281
Penalty (after the first free use hours)-income per capita ratio (10
%4
)%1387.231 0.5834 8.625 %2109 0.5698 3.551 %5097.215 %0.050 0.298
Summary statistics
N69 69 N69
R
2
0.9084 0.8610 Prob PRESS 40.1
Adj. R
2
0.8868 0.8398
a
VIF: variance inflation factor.
b
B': standardized coefficient.
c
VIP: variable importance for projection.
nn
significant at the 0.05 level,
nnn
significant at the 0.01 level.
†
Variable with a VIP larger than 0.8 is not considered to be “small”for projection (Wold, 1994),
††
variable with a VIP larger than 1.0.
J. Zhao et al. / Transport Policy 35 (2014) 253–264258
three dummy variables are very important (VIP41.0) variables in
promoting bikesharing turnover rate. Whilst Fig. 3 shows that
bikesharing turnover rate tends to increase with the number of
public bikes, the PLS modeling results indicate that, all else being
equal, increasing public bikes tends to significantly lower bike-
sharing turnover rate. This seems a paradox yet can be explained
by at least two reasons: First, according to the definition for
turnover rate (Formula 1), the magnitude of turnover rate varies
inversely with the number of public bikes. Second, the relationship
of turnover rate with the number of public bikes in Fig. 3 does not
account for other independent variables, in particular, the number
of bikesharing members that reflects bikesharing demand. Under
the condition of all else being equal, particularly the potential
demand equal, bikesharing turnover could decline with the
number of public bikes. Moreover, the number of docking stations
carries important and positive impact on bikesharing turnover
rate. Its positive impact might be due to the fact that bikesharing
members can pick-up and drop-off public bikes more conveniently
if more docking stations are installed, thus improve bikesharing
turnover rate.
The significant but adverse signs of the number of public bikes
in explaining bikesharing ridership and turnover rate carry an
important message to bikesharing operators and planners. Model-
ing results from Table 2 indicate that bikesharing ridership tends
to increase with the number of public bikes, but the same
independent variable turns out to be negatively associated with
turnover rate. To pursue an ideal turnover rate (higher than four
daily uses per bike as Gauthier et al. (2013) expected), the number
of public bikes should be neither too many nor too few, reflecting
the relationship between bikesharing demand and bikesharing
supply. Fig. 4 plots a range of cities based on a comparison
between their bike-member (supply-demand) ratio and bikeshar-
ing turnover rate. Generally, turnover rate declines obviously
when the bike-member ratio is larger than 0.2. In other words, a
system with an ideal turnover rate needs at least 5 potential users
(bikesharing members in China) for each public bike.
y = 102.23x + 10705
R² = 0.2219
0
50000
100000
150000
200000
250000
300000
0 500 1000 1500
Population (10 thousand)
y = 0.3029x + 18980
R² = 0.13
0
50000
100000
150000
200000
250000
300000
0100000200000300000400000500000
Government expenditure (million RMB)
y = 0.2958x + 6737.5
R² = 0.863
0
50000
100000
150000
200000
250000
300000
0 500000 1000000 1500000
Bikesharing members
Bikesharing daily use
y = 134.82x + 1155.7
R² = 0.8244
0
50000
100000
150000
200000
250000
300000
0 500 1000 1500 2000 2500 3000
Docking stations
y = 3.9997x + 4021
R² = 0.8486
0
50000
100000
150000
200000
250000
300000
0 20000 40000 60000 80000
Public bikes
Fig. 2. Comparisons of bikesharing daily use versus urban population, government expenditure, bikesharing members, docking stations, and public bikes.
J. Zhao et al. / Transport Policy 35 (2014) 253–264 259
4.3. Bikesharing ridership and effectiveness by type
Table 4 presents the average daily use and turnover rate of
bikesharing systems, classed by using personal credit cards or not,
providing 24 h service or not, and using universal cards or not. The
average daily use and turnover rate of the 11 bikesharing systems
by adopting personal credit cards are respectively 2.44 and 1.46
times the daily use and turnover rate of the 58 systems by
charging penalty on bikesharing users if they do not return bikes
within the free use hours. Compared to the systems without
providing 24 h service, the systems providing 24 h service have a
fewer daily use but a higher turnover rate. The effectiveness of the
later is 1.23 times higher that of the former. Averagely, systems
using universal cards have more ridership and higher turnover
rate, which are respectively 2.59 and 1.27 times the systems
without adopting universal cards. Whilst encouraging as these
results are, the percentages of adopting these three innovations
are quite low in China. For example, only 15.9% (11 out of 69) cities
using personal credit cards and only 20.3% (14 out of 69) cities
adopting universal cards.
We also develop models of bikesharing ridership and turnover
rate by type, particularly for the systems using personal credit
cards, providing 24 h service, and adopting universal cards. To
preserve the observations to be sufficient for modeling, we did not
classify these three different systems into sub-types. For example,
the systems using personal credit cards can further be classified
into providing 24 h service or not and adopting universal cards or
not. By doing this, however, the number of factors may be greater
than the number of observations, which is likely to get a model
that fits the sampled data perfectly but that will fail to predict new
data well (Tobias, 1999). Moreover, the modeling results by type
are compared to the modeling results of overall systems.
Table 5 presents the modeling results of the factor analysis.
Whilst the coefficients vary obviously among different types of
y = 0.8742x + 1.5309
R² = 0.3007
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
8.00
9.00
10.00
0123456
Number of memberships, grouped by
1 (<5000), 2(5000-10000), 3(10000-15),
4(20000>100000), 5 (>=100000)
y = 0.7114ln(x) - 2.7957
R² = 0.2943
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
8.00
9.00
10.00
0 500000 1000000 1500000
Bikesharing members
y = 0.5242x + 2.847
R² = 0.1157
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
8.00
9.00
10.00
0123456
Number of docking stations, grouped by
1 (<50), 2(50-100), 3(100-200), 4(200-400), 5(>=400)
Turnover rate
y = 0.6091ln(x) + 1.4595
R² = 0.1267
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
8.00
9.00
10.00
0 500 1000 1500 2000 2500 3000
Docking stations
y = 0.4688x + 2.7716
R² = 0.1114
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
8.00
9.00
10.00
0123456
Public bikes, grouped by
1 (<750), 2(750-1500), 3(1500-3000),
4(3000-5000), 5 (>=5000)
y = 0.4852ln(x) + 0.4483
R² = 0.0931
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
8.00
9.00
10.00
0 20000 40000 60000 80000
Public bikes
Fig. 3. Comparisons of bikesharing turnover rate versus bikesharing members, docking stations, and public bikes.
J. Zhao et al. / Transport Policy 35 (2014) 253–264260
system, most of the independent variables have the same expected
sign as that in Tables 2 and 3. Generally, regarding different
systems types, bikesharing ridership and turnover tend to increase
with population, government expenditure, the number of bike-
sharing members and docking stations. The signs of the number of
public bikes, again, are found to be adverse in explaining bike-
sharing ridership and turnover rate. In addition, we find that the
coefficients of the number of bikesharing members by type are
larger than that of overall systems, indicating that using personal
credit cards, providing 24 h service, and/or adopting universal
cards may promote members to use bikesharing more frequently.
5. Conclusions
At a time when vehicle travel is increasing and the associated
problems are aggravating, urban planners and decision-makers
have gradually shifted their emphasis from providing additional
road space for driving to examining the need for more sustainable
transportation strategies. Transit priority that allows public trans-
portation to be a more attractive and viable option for commuters
has been proposed as a promising strategy to help relieve people's
heavy reliance on automobiles. As a pinnacle of green transporta-
tion with transit attributes, public bikesharing has become parti-
cularly popular since the mid-2000s.
In recent years adoption of and studies on this widespread urban
public transportation “revolution”have spread across the globe
(Shaheen et al., 2010; Midgley, 2011; Lin and Yang, 2011;
García-Palomares et al., 2012; Parkes et al., 2013; O’Brien et al.,
Table 3
OLS and PLS regression models of bikesharing turnover rate.
Variable Log OLS model with variables whose Pearson correlation coefficients
are smaller than 0.7.
Log PLS model
Coeff. p-value VIF
a
Coeff. B'
b
VIP
c
(Constant) %1.847 0.9261 0 %2.292 0
Urban features
Log(population (10 thousand)) 0.531
nn
0.0214 1.996 2.385
††
0.571 1.243
Log(jobs (10 thousand)) %1.625
†
%0.432 0.993
Log(income per capita (RMB)) 0.2028 0.9616 2.948 %0.027
†
%0.001 0.958
Log(government expenditure (million RBM)) 0.543
††
0.138 1.150
System characteristics
Log(number of members) 3.446
††
1.141 2.019
Log(number of docking stations) 0.966
nn
0.0125 1.451 0.117
††
%0.029 1.237
Log(number of public bikes) %2.062
††
%0.563 1.164
Log(operation days since the system opened) 0.240 0.7623 2.345 %0.589
†
%0.099 0.879
Using personal credit cards or not (1–0) 1.116
n
0.0898 3.339 1.045
††
0.186 1.199
Providing 24 h service or not (1–0) 0.888
n
0.0749 1.126 0.682
††
0.164 1.080
Using universal cards or not (1–0) 0.717
nn
0.0315 1.086 0.609
††
0.121 1.321
Composite indictors
Deposit-income per capita ratio (10-3) %0.009 0.9924 1.769 %0.0 06 %0.009 0.412
Penalty (after the first free use hours)-income per capita ratio (10-4) %3.240 0.6023 3.840 %4.241
†
%0.213 0.938
Summary statistics
N69 N69
R
2
0.4694 Prob PRESS 40.1
Adj. R
2
0.3579
a
VIF: variance inflation factor.
b
B': standardized coefficient.
c
VIP: variable importance for projection.
n
Significant at the 0.1 level,
nn
significant at the 0.05 level.
†
Variable with a VIP larger than 0.8 is not considered to be “small”for projection (Wold, 1994),
††
variable with a VIP larger than 1.0.
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
8.00
9.00
10.00
0
Turnover rate
y = -2.566ln(x) - 1.2121
R² = 0.5688
0.2 0.4 0.6
Bike-membership ratio
Fig. 4. Comparison of turnover rates versus bike–member ratio.
Table 4
Average daily use and turnover rate of bikesharing systems, classified by using
personal credit cards or not, providing 24 h service or not, and using universal
cards or not.
Classes Daily use Turnover rate
Average SD Average SD
Using personal credit cards or not
Yes (n¼11) 54,694 73,731 5.71 1.48
No (n¼58) 22,404 46,108 3.9 2.05
Providing 24 h service or not
Yes (n¼39) 21,023 35,653 4.56 1.94
No (n¼30) 36,038 67,736 3.71 2.15
Using universal cards or not or not
Yes (n¼14) 53,951 83,634 5.04 2.53
No (n¼55) 20,831 38,971 3.98 1.90
J. Zhao et al. / Transport Policy 35 (2014) 253–264 261
2014). Two key questions for the success of bikesharing programs are
how many ridership bikesharing systems can attract, and what
influences their effectiveness. To address these, this paper evaluates
factors affecting bikesharing daily use and turnover rate by analyzing
data from 69 bikesharing systems in China. In addition to comparing
daily use and turnover versus various variables, we model bikesharing
daily use and turnover rate respectively as a function of independent
variables measuring urban features and system characteristics, by
adopting OLS and PLS regression models.
Regression results indicate that both bikesharing ridership and
turnover rate tend to increase with urban population, government
expenditure, the number of bikesharing members and docking
stations. While the number of public bikes has a significantly
positive impact on bikesharing ridership, it might lower bikeshar-
ing turnover rate at the same time, with all else being equal.
Moreover, personal credit cards and universal cards can signifi-
cantly raise bikesharing daily use and turnover rate. The models
and comparisons displayed in this study provide urban planners
and transit agencies with direct demand models for bikesharing
ridership and turnover rate estimation.
The comparison between bike–member ratio and bikesharing
turnover rate indicates that, to pursue an ideal turnover rate, the
number of public bikes should be neither too many nor too few.
Regarding the bikesharing systems in China, the ideal cut off value
for the bike–member ratio is 0.2. In other words, for each public bike
there should be at least 5 potential users (bikesharing members in
this study). The results should catch bikesharing operators' and
transit agencies' attention especially. Whilst the more public bikes,
usually the higher the bikesharing ridership level is, yet all else being
equal, increasing public bikes may significantly decrease system's
effectiveness.
Our analysis also recommends bikesharing operators and transit
agencies to consider the policies of adopting personal credit cards
and universal cards, in particular, the adoption of the latter. Indeed,
the free charge rate of bikesharing use is quite high in China, for
example, 96% in Hangzhou. Most bikesharing users in China can
return public bikes within the first free 1 h (the time threshold
adopted by most bikesharing systems in China). Bikesharing opera-
tors can only earn quite a small profit from the use fee penalty. Why
not considering using personal credit cards to increase both the
bikesharing daily use and turnover rate with quite a few more
subsidies?
5
Unfortunately, to date, only 15.9% (11 out of 69) cities in
China adopt this strategy. In addition, only 20.3% (14 out of 69) cities
integrate bikesharing systems into other urban transit systems
through the use of universal cards. During the early adoption of
Table 5
PLS models of bikesharing ridership and turnover by system type.
Variable
a
PLS model of ridership Log PLS model of turnover rate
Coeff. B'
b
VIP
c
Coeff. B'
b
VIP
c
All systems (N¼69)
(Constant) %13376.086 0 %0.062 0
Population (10 thousand) 30.582
††
0.016 1.035 1.108
††
0.265 1.463
Government expenditure (million RBM) 0.080
†
0.095 0.906 0.417
†
0.142 0.821
Number of members 0.066
††
0.208 1.320 3.927
††
1.366 2.085
Number of docking stations 24.073
††
0.364 1.316 0.596
†
0.151 0.954
Number of public bikes 1.539
††
0.354 1.334 %2.565
††
%0.701 1.346
Systems use personal credit cards (N¼11)
(Constant) 5432.175 0 1.747 0
Population (10 thousand) 20.813
†
0.112 0.899 0.873 0.284 0.787
Government expenditure (million RBM) 0.26 8
†
0.408 0.979 1.938
†
0.513 0.852
Number of members 0.112
††
0.369 1.221 3.307
††
1.428 1.848
Number of docking stations 5.468
††
0.026 1.243 4.01
††
1.347 2.660
Number of public bikes 1.794
††
0.521 1.233 %6.29
††
%2.703 1.174
Systems provide 24 h service (N¼39)
(Constant) %2173.70 0 0.341 0
Population (10 thousand) 16.213
†
0.841 0.878 1.895
††
0.484 1.573
Government expenditure (million RBM) 0.078
†
0.143 0.923 1.438 0.434 0.792
Number of members 0.131
††
0.291 1.200 5.629
††
1.765 1.946
Number of docking stations 39.185
††
0.151 1.180 %0.767
†
%0.176 0.919
Number of public bikes 2.040
††
0.639 1.276 %3.168
††
%1.371 1.145
Systems adopt universal cards (N¼14) 1.696
(Constant) 20796.35 0 0.230 0
Population (10 thousand) 49.902
†
0.067 0.961 1.420
††
0.217 1.014
Government expenditure (million RBM) %1.195
†
%0.233 0.964 %1.475
†
%0.443 0.982
Number of members 0.102
††
0.107 1.134 4.050
††
1.164 1.696
Number of docking stations 83.843
††
0.649 1.212 1.883
††
0.662 1.338
Number of public bikes 1.904
††
0.391 1.210 %3.917
††
%1.272 1.032
a
Log independent variables' values for turnover rate modeling.
b
B
0
: standardized coefficient.
c
VIP: variable importance for projection.
†
Variable with a VIP larger than 0.8 is not considered to be “small”for projection (Wold, 1994),
††
Variable with a VIP larger than 1.0.
5
Practice of the adoption of personal credit cards indicates that this policy can
evoke the enthusiasm for applying a bikesharing member, thus increase the
potential bikesharing use and improve the turnover rate. Personal credit cards,
based on our insights, however, also have two main shortcomings: 1) this policy
heavily relies on government subsidy, which may corrode system's sustainability
and, 2) this zero-cost model can play a role in increasing bikesharing daily use but
also may prolong bikes’occupation time. Some cities in China, such as Wuhan, have
overridden this policy after several years’practice, yet suffered lots during the
transformation, such as a sharp decrease in bikesharing members and a long
depression in bikesharing use. Personal credit cards are still adopted by some cities
in China, including, Shanghai, Chengdu, and Suzhou, transit authorities are
encouraged to adopt this model but also need to do enough feasibility study on
the cost-effectiveness analysis of this policy.
J. Zhao et al. / Transport Policy 35 (2014) 253–264262
bikesharing, integrating bikesharing into other transit systems may
meet with various challenges from technology and investment
issues. These challenges, however, could be resolved gradually as
bikesharing continues to catch people's attention, survive to policy
developments, and integrate technological innovation.
At the conclusion of this study, a host of other questions
regarding bikesharing remain unanswered. What other factors
influence bikesharing ridership and effectiveness? In addition to
the variables measuring urban features and system characteristics,
one would expect that good bikesharing neighborhoods (built
environment design) and high level of service (rebalancing, main-
tenance, etc.) can also raise bikesharing ridership and effective-
ness. Second, would personal credit cards be sustainable in
bikesharing adoption? Bikesharing users would expect that there
is less cost in using public bikes, whereas local governments and
bikesharing operators would hope the systems to be more cost-
effective. How to look for a balance between cost and effective-
ness? Third, universal cards turn out to be an effective policy in
raising both bikesharing daily use and turnover rate, what are the
other policies having the same positive impacts? Opening the
system (for example, anyone in possession of a debit or credit card
can access to bikesharing in London) is found to correlate with
greater daily use. However, registration policy required by most
cities in China is a major barrier to the adoption of “open policy”.
Would it be possible to override the registration policy to create an
opening bikesharing system in China? Whilst we did not shed
light on answers to these questions, it is believed that these
questions can be addressed in the near future, as bikesharing
continues to be more diversified in system design and policy
adoption.
Acknowledgments
This research is supported by the Chinese Scholarship Council
(No. 201306090063) and the Scientific Research Foundation of
Graduate School of Southeast University (No. YBPY1405). We want
to express our gratitude to Professor Greg Marsden and two
anonymous reviewers for their helpful comments and suggestions.
Appendix A
See Table A1.
References
Chemla, D., Meunier, F., Calvo, R.W., 2012. Bike Sharing System: Solving the Static
Rebalancing Problem. Available from: 〈http://hal-enpc.archives-ouvertes.fr/
docs/00/72/66/17/PDF/SVOCPDP_HAL.pdf〉(accessed 20.02.14.).
Christensen, M., Shaheen, S.A., 2014. Can Bikeshare Systems Ever Stand on their
Own Two Wheels? Available from: 〈http://theconversation.com/can-bikeshar
e-systems-ever-stand-on-their-own-two-wheels-22981〉(accessed 18.02.14.).
Table A.1
Information on the key characteristics of bikesharing systems in China (ranked by
turnover rate).
City Opening
month
a
Operation
days
b
Turnover
rate
a
Daily
use
a
Docking
stations
a
Public
bikes
a
Jiaxing 12–2011 315 0.7 1000 50 1500
Binzhou 8–2013 112 1.1 450 25 400
Kaixian 1–2011 220 1.2 600 20 500
Zhoushan 10–2009 190 1.4 700 15 500
Qingzhou 9–2010 350 1.4 14,000 506 10,000
Guangyuan 8–2010 120 1.5 1500 29 1000
Hohhot 10–2013 100 1.6 4000 50 250 0
Ordos 8–2013 102 1.8 2189 44 1200
Dujiangyan 4–2010 197 1.9 1500 80 800
Yongchuan 6–2011 120 2.0 1000 25 50 0
Jiayuguan 5–2013 100 2.0 300 7 150
Ningbo 9–2013 93 2.0 40 00 100 20 00
Ninghai 6–2011 358 2.1 4500 100 2200
Qidong 3–2013 180 2.1 2500 60 1200
Ninghai 6–2011 670 2.3 6876 151 3000
Table A.1 (continued)
City Opening
month
a
Operation
days
b
Turnover
rate
a
Daily
use
a
Docking
stations
a
Public
bikes
a
Wujiang 12–2011 102 2.3 3568 100 1532
Zhuhai 12–2012 227 2.4 12,000 195 5000
Yiwu 10–2013 93 2.5 2500 50 1000
Baoji 9–2013 307 2.6 5149 10 0 20 00
Shaoxing 6–2011 160 2.6 9000 82 3500
Heihe 5–2012 98 2.8 6333 62 2300
Yantai 8–2010 1110 2.8 360 0 110 1300
Huizhou 4–2012 191 2.8 5569 100 2000
Wuxi 1–2010 960 3.0 4500 27 1500
Nantong 6–2011 715 3.0 12,00 0 60 40 00
Wuhan 11–2008 1460 3.1 220,000 1218 70,000
Nanjing 1–2011 101 3.2 18,000 307 560 0
Pinghu 7–2013 950 3.2 3210 30 1000
Wuhu 9–2012 365 3.3 39,000 478 12,000
Yinchuan 7–2010 1050 3.3 1060 7 320
Zaozhuang 7–2012 210 3.3 1200 17 360
Linhai 6–2011 720 3.6 4493 52 1250
Foshan 8–2010 600 3.6 67,200 441 18,520
Hangzhou 5–2008 1560 4.0 257,600 2435 65,0 00
Jiangyin 11–2008 305 4.0 2800 23 70 0
Liuyang 1–2012 485 4.0 4000 50 1000
Taizhou 2–2010 820 4.0 40,0 00 205 10,000
Changde 10–2012 200 4.0 4000 63 1000
Aksu 5–2013 106 4.2 2118 20 500
Wenling 1–2012 415 4.5 22,500 120 5000
Guangzhou 8–2009 730 4.6 23,000 116 5000
Nanchang 9–2008 1125 4.8 24,000 120 50 00
Tongliang 10–2010 300 5.0 2500 16 50 0
Fuzhou 6–2011 670 5.0 7000 59 1400
Zhangjiagang 6–2010 990 5.0 16,000 152 3200
Shenzhen 9–2010 630 5.1 45,932 30 0 9054
Mengzi 10–2013 93 5.2 2216 43 426
Shenmu 5–2013 221 5.3 4000 30 750
Huangyan 1–2012 630 5.4 21,430 114 4000
Mianyang 10–2012 395 5.8 5479 51 950
Shanghai 3–2009 1140 5.8 162,400 574 28,00 0
Zhenjiang 4–2013 95 6.0 60 00 40 1000
Jiyuan 8–2013 120 6.0 3000 32 500
Shaoyang 10–2013 180 6.1 3200 30 528
Yongcheng 1–2013 297 6.2 6851 55 1100
Beijing
c
6–2011 399 6.3 35,000 280 5600
Lishui 2–2013 210 6.3 5000 26 800
Sunning 9–2011 370 6.3 7500 109 1190
Chizhou 9–2009 303 6.7 20,0 00 67 300 0
Suzhou 8–2010 365 6.8 101,450 640 15,000
Zhongshan 10–2011 390 6.8 47,300 267 7000
Chengdu 6–2010 450 7.0 10,500 156 1500
Changshu 5–2011 370 7.2 21,600 150 30 00
Yixing 12–2012 152 7.3 19,000 115 2600
Zhuzhou 5–2011 730 7.5 150,000 502 20,0 00
Wenzhou 12–2012 425 7.6 53,50 0 254 7000
Kunshan 9–2010 510 7.8 46,666 30 0 60 00
Xuzhou 9–2012 377 8.0 64,000 297 8000
Taiyuan 9–2012 265 9.5 190,000 1000 20,000
a
Data from various sources, including local bikesharing agencies, online
documents, unpublished records, and websites (such as, http://www.hzsggzxc.
com/ (Hangzhou Public Bicycle), http://www.bjysj.gov.cn/bjggzxc/ (Beijing Public
Bicycle), http://zxc.gzzc.com.cn/ (Guangzhou Public Bicycle) to name very a
few here).
b
The days between the system opening day and the data releasing day.
c
Whilst the history of Beijing's public can be dated back to the 1990s, the first
bikesharing system supported by the local governments and public agencies was
launched in June 2011 in the districts of Dongcheng and Chaoyang.
J. Zhao et al. / Transport Policy 35 (2014) 253–264 263
Clark, W.A.V., Hosking, P.L., 1986. Statistical Methods for Geographers. Wiley, New
York.
Cui, Y., 2011. Wuhan's Public Bikes. Available from: 〈http://www.hb.xinhuanet.com/
newscenter/2011-08/25/content_23546188.htm〉(accessed 21.02.14.).
DeMaio, P., 2009. Bike sharing: history, impacts, models of provision, and future. J.
Public Transp. 12, 41–56.
DeMaio, P., Meddin, R., 2014. The Bike-sharing World Map. Available from: 〈http://
bit.ly/K9pKmO〉(accessed 23.02.14.).
García-Palomares, J.C., Gutiérrez, J., Latorre, M., 2012. Optimizing the location of
stations in bike-sharing programs: a GIS approach. Appl. Geogr. 35, 235–246.
Gauthier, A., Hughes, C., Kost, C., Li, S., et al., 2013. The Bike-share Planning Guide.
ITDP Report. Available from: 〈http://www.itdp.org/documents/ITDP_Bike_Shar
e_Planning_Guide.pdf〉(accessed 19.02.14.).
Guerra, E., Cervero, R., 2011. Cost of a ride: the effects of densities on fixed-guide
way transit ridership and costs. J. Am. Plan. Assoc. 77, 267–290.
Hangzhou Public Bicycle, 2013. An Introduction of Hangzhou Public Bicycle.
Available from: 〈http://www.hzsggzxc.com/about.aspx?c_kind=521&c_k
ind2=522&c_kind3=531〉(accessed 10.11.13.).
Hangzhou Public Bicycle, 2014. Available from: 〈http://www.hzsggzxc.com/about.
aspx?c_kind=521&c_kind2=525&c_kind3=906〉(accessed 19.02.14.).
Hughes, C., 2014. Planning Bike Share for Better Performance. Presented at
Transforming Transportation 2014. Available from: 〈http://www.slideshare.
net/EMBARQNetwork/planning-bike-share-for-better-performance-colin-hugh
es-itdp〉(accessed 19.02.14.).
Jäppinen, S., Toivonen, T., Salonen, M., 2013. Modelling the potential effect of shared
bicycles on public transport travel times in Greater Helsinki: an open data
approach. Appl. Geogr. 43, 13–24.
Jensen, P., Rouquier, J.B., Ovtracht, N., Robardet, C., 2010. Characterizing the speed
and paths of shared bicycles in Lyon. Transp. Res. Part D 15, 522–524.
Kaltenbrunner, A., Meza, R., Grivolla, J., Codina, J., Banchs, R., 2010. Urban cycles and
mobility patterns: exploring and predicting trends in a bicycle-based public
transport system. Pervasive Mobile Comput. 6, 455–466.
Kutner, M.H., Nachtsheim, C.J., Neter, J., 2004. Applied Linear Regression Models,
4th ed. McGraw-Hill, Irwin.
Lathia, N., Ahmed, S., Capra, L., 2012. Measuring the impact of opening the London
shared bicycle scheme to casual users. Transp. Res. Part C 22, 88–102.
Lin, J.R., Yang, T.H., 2011. Strategic design of public bicycle sharing systems with
service level constraints. Transp. Res. Part E 47, 284–294.
Midgley, P., 2011. Bicycle-Sharing Schemes: Enhancing Sustainable Mobility in
Urban Areas. United Nations Department of Economic and Social Affairs.
Background Paper No. 8.
Nair, R., Miller-Hooks, E., 2014. Equilibrium network design of shared-vehicle
systems. Eur. J. Oper. Res. 235 (1), 47–61.
National Bureau of Statistics of China, 2012. Urban Statistical Yearbook of China.
China Statistics Press, Beijing p. 2012.
National Bureau of Statistics of China, 2013. Statistic Yearbook of China. Available
from: 〈http://www.stats.gov.cn〉(accessed 20.02.14.).
O'Brien, O., Cheshire, J., Batty, M., 2014. Mining bicycle sharing data for generating
insights into sustainable transport systems. J. Transp. Geogr. 34, 262–273.
Parkes, S.D., Marsden, G., Shaheen, S.A., Cohen, A.P., 2013. Understanding the
diffusion of public bikesharing systems: evidence from Europe and North
America. J. Transp. Geogr. 31, 94–103.
Raviv, T., Tzur, M., Forma, I.A., 2013. Static repositioning in a bike-sharing system:
models and solution approaches. EURO J. Transp. Logist. 2 (3), 187–229.
Shaheen, S.A., Cohen, A.P., Martin, E.W., 2013. Public bikesharing in North America:
early operator understanding and emerging trends. Transp. Res. Rec. 2387,
83–92.
Shaheen, S.A., Guzman, S., Zhang, H., 2010. Bikesharing in Europe, the Americas,
and Asia: past, present, and future. Transp. Res. Rec. 2143, 159–167.
Shaheen, S.A., Martin, E.W., Cohen, A.P., Finson, R.S., 2012. Public Bikesharing in
North America: Early Operator and User Understanding. Available from: 〈http://
transweb.sjsu.edu/project/1029.html〉(accessed 02.06.14.).
Shaheen, S.A., Zhang, H., Martin, E., Guzman, S., 2011. China's Hangzhou public
bicycle: understanding early adoption and behavioral response to bikesharing.
Transp. Res. Rec. 2247, 33–41.
State Council, 2012. State Council's Guidance on the Priority Development of Urban
Transit. Available from: 〈http://www.gov.cn/zwgk/2013-01/05/content_2304962.
htm〉(accessed 20.02.14.).
Tobias, R.D., 1999. An Introduction to Partial Least Squares Regression. SAS Institute,
Cary, NC, pp. 1–8.
Wold, S., 1994. Exponentially weighted moving principal component analysis and
projections to latent structures. Chemometr. Intell. Lab. Syst. 23, 149–161.
Zhao, B., 2012. Deposit of Xi’an's Bikesharing: 200 RMB for Local Residents vs. 700
RMB for Occasional Users. Available from: 〈http://news.hsw.cn/system/2012/
08/01/051406079.shtml〉(accessed 21.02.14.).
J. Zhao et al. / Transport Policy 35 (2014) 253–264264