You are on page 1of 12

Transportation Research Part A 77 (2015) 213–224

Contents lists available at ScienceDirect

Transportation Research Part A


journal homepage: www.elsevier.com/locate/tra

Impact evaluation of a mass transit fare change on demand


and revenue utilizing smart card data
Zi-jia Wang a,b,c, Xiao-hong Li a, Feng Chen a,b,⇑
a
School of Civil and Architectural Engineering, Beijing Jiaotong University, No.3 Shangyuancun, Haidian District, Beijing 100044, PR China
b
Beijing Engineering and Technology Research Center of Rail Transit Line Safety and Disaster Prevention, No.3 Shangyuancun, Haidian District, Beijing 100044,
PR China
c
Department of Civil, Environmental and Geomatic Engineering, University College London, Chadwick Building, Gower Street, London WC1E 6BT, England,
United Kingdom

a r t i c l e i n f o a b s t r a c t

Article history: Transit fares are an effective tool for demand management. Transit agencies can raise
Received 11 July 2014 revenue or relieve overcrowding via fare increases, but they are always confronted with
Received in revised form 25 March 2015 the possibility of heavy ridership losses. Therefore, the outcome of fare changes should be
Accepted 25 April 2015
evaluated before implementation. In this work, a methodology was formulated based on
Available online 16 May 2015
elasticity and exhaustive transit card data, and a network approach was proposed to assess
the influence of distance-based fare increases on ridership and revenue. The approach was
Keywords:
applied to a fare change plan for Beijing Metro. The price elasticities of demand for Beijing
Fare change
Metro
Metro at various fare levels and trip distances were tabulated from a stated preference
Network survey. Trip data recorded by an automatic fare collection system was used alongside the
Smart card data topology of the Beijing Metro system to calculate the shortest path lengths between all
Trip distance station pairs, the origin–destination matrix, and trip lengths. Finally, three fare increase
Demand alternatives (high, medium, and low) were evaluated in terms of their impact on ridership
and revenue. The results demonstrated that smart card data have great potential with
regard to fare change evaluation. According to smart card data for a large transit network,
the statistical frequency of trip lengths is more highly concentrated than that of the shortest
path length. Moreover, the majority of the total trips have a length of around 15 km, and
these are the most sensitive to fare increases. Specific attention should be paid to this
characteristic when developing fare change plans to manage demand or raise revenue.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction

Public transit fares not only affect the revenue of operating agencies but also travel demand due to passengers’ trip deci-
sions in response to fare changes (Webster and Bly, 1980). Therefore, urban transport agencies are able to manage travel
demand through fare policy, as has been observed in the cases of London Transport, Atlanta Rapid Transit, Los Angeles
and other metropolitan cities in developed countries (Evans, 2004; Zureiqat, 2008).

⇑ Corresponding author at: School of Civil and Architectural Engineering, Beijing Jiaotong University, No.3 Shangyuancun, Haidian District, Beijing
100044, PR China. Tel.: +86 10 51685312; fax: +86 10 51688070.
E-mail address: fengchen@bjtu.edu.cn (F. Chen).

http://dx.doi.org/10.1016/j.tra.2015.04.018
0965-8564/Ó 2015 Elsevier Ltd. All rights reserved.
214 Z.-j. Wang et al. / Transportation Research Part A 77 (2015) 213–224

As the capital of the largest developing country, Beijing’s public transit system provides significant numbers of daily res-
idential trips and is approaching its capacity. However, its financial performance is deteriorating concurrent with the rapid
growth of newly built lines. It is ascribed to the extraordinarily low fare charges and unsustainable government subsidies.
This is particularly true for the metro. By the end of May 2015, the metro system consists of 17 lines and 277 stations (trans-
fer stations being counted more than once) and the total network length is 465 km. The ridership of Beijing Metro reached
3.209 billion trips in 2013 (based on the count of unlinked trips as every transfer in the system is counted as an additional
trip). This represents a daily Demand of more than 10 million trips on a busy weekday. The stations and trains are heavily
crowded during rush hours, reaching 8 persons per m2, as shown in Fig. 1. Congestion management measures such as hold-
ing incoming passengers at the entrance are applied to 40 stations during peak hours.
Given such a large system, the fare structure is simple: a flat fare of 2 RMB for the entire network except the Capital
Airport Line, which charges 25 RMB. The fare is roughly a third of that in Shanghai or Guangzhou, China. The fare revenue
of Beijing Metro cannot cover its operating costs, not to mention construction costs, and a 20 billion RMB subsidy was pro-
vided by the government for the bus and rail operation company in 2013 (FBB, 2014). The subsidy is expected to increase in
order to pay for and maintain more newly built lines every year. Based on the long-term planning of Beijing Metro, estimated
investment will rise to 297.8 billion RMB by 2020 (Zhang et al., 2012). Currently, Beijing government can provide only 10 bil-
lion RMB per year as a special fiscal budget.
Although various financing modes have been investigated and implemented, including a public–private partnership with
Hong Kong MTR for Line 4 and a build-transfer mode for the Olympic Line, the financing gap for Beijing Metro’s construction
and operation is still extremely large. Mass public transit is regarded as a quasi-public good, and welfare policies are irre-
versible, so fare increase proposals have been opposed intensely by the public and several fare change plans have
deadlocked.
However, the current fare policy has threatened the level of service and the sustainable development of Beijing Metro
significantly, and fare change is imperative. The initial proposal is to increase the base fare and charge for a linked trip
according to its length. Before the new fare policy is implemented, passengers’ responses should be investigated; that is,
the impact on travel demand, and the revenue after fare increases should be assessed (Chen and Lin, 2005; Kremers
et al., 2002).
The objective of this study is to develop a method to evaluate the impact of fare increases on the demand and revenue for
a transit network using a large amount of smart card data. To this end, this contribution first establishes an estimation model
of revenue pertaining to fare increases. The model is formulated with price elasticity, current metro demand, and shortest
paths between any stations in metro network. For the data needed in the model, the price elasticities are collected by a stated
preference (SP) survey. Current demand is extracted from an automatic fee collection (AFC) system by assigning each record
to an origin–destination (OD) pair according to its entry and exit station. And shortest paths between any stations are
obtained by a network approach. Finally, using this model, the impacts on demand and revenue are estimated for each fare
increase alternative, and corresponding discussion and conclusions are provided.

Fig. 1. Overcrowding problem in stations and trains of Beijing Metro. Source: Taken in the field study.
Z.-j. Wang et al. / Transportation Research Part A 77 (2015) 213–224 215

2. Literature review

A large number of studies have analyzed passengers’ responses to fare change using SP surveys, revealed preference (RP)
surveys, or a combination of the two. Passengers’ responses to fare change are generally referred to as price elasticity of
demand, which is the percentage change of travel demand due to a 1% change in fare. Price elasticity is a vital factor in pre-
dicting transit ridership and revenue change due to fare change. A literature review indicates that the price elasticity of travel
demand varies greatly in terms of such variables as time span, transit mode, time of day, original fare level, income level, trip
distance, data paradigm, type and direction of price change, demographic and geographic conditions (ATOC, 2002; Dargay
and Hanly, 2002; Hensher, 2008; Litman, 2004). In general, without considering such contexts as mode or peak hour, a
rule-of-thumb value of 0.3 is often taken as the default elasticity based on earlier studies (Bresson et al., 2003; Collins,
1982; Goodwin, 1992). As more and more data pertaining to demand change before and after fare changes became available,
price elasticities of travel demand and their contexts have been collected and analyzed through meta-analysis. Holmgren
(2007) collected 81 cases and found the mean value is 0.38, which is close to the aforementioned rule-of-thumb despite
a large variation. Hensher (2008) drew a mean elasticity of 0.395 from 319 observations, which is consistent with those
of Holmgren’s and the rule-of-thumb value. Paulley et al. (2006) proposed that a price elasticity of 0.4 is closer to what
was revealed by recent data than the generally used value of 0.3.
In fact, the general price elasticities referred to here are seldom applied in a particular case because elasticities vary sub-
stantially with contexts, such as time span, transit mode, and time of day. Therefore, price elasticities should be analyzed in a
specific context. For example, time span matters because residents’ trip decisions differ in response to transit fare change in
terms of time; they may choose to purchase a car to cope with fare increases in the long run, and thus, short-term and
long-term elasticities differ. Paulley et al. (2006) suggested 1–2 years can be regarded as the short term. Long-term price
elasticities can be twice as much as short-term ones, up to 1 (Litman, 2004; Paulley et al., 2006; Sharaby and Shiftan,
2012), which indicates the price elasticity of transit demand is 0.2 to 0.5 for the first year, and 0.6 to 0.9 five years
later. Without particular specification, price elasticities of transit demand refer to short-term elasticities. For transit modes,
the price elasticity for rail transit is generally much lower than that of buses because rail riders’ income is higher (Litman,
2004; Paulley et al., 2006). Paulley et al. (2006) proposed that the short-term price elasticities for bus and metro services are
0.4 and 0.3, respectively. Riders of buses respond to fare change more quickly than riders of the metro do (Dygert et al.,
1977). With regard to time of day, off-peak price elasticity is generally 1.5–2 times as that of peak hours because trips during
off-peak periods are more flexible (Litman, 2004; Paulley et al., 2006). According to data collected in the UK, price elasticities
for bus, metro, and suburban rail services are 0.26, 0.26, and 0.34, respectively, for peak-hour demand, and 0.48,
0.42, and 0.79, respectively, for off-peak demand (Paulley et al., 2006).
In contrast to fixed values for elasticity, it has been suggested that range values should be used, because no single elas-
ticity can be applied to all situations (Litman, 2004). Paulley et al. (2006) recommended a range of short-term transit price
elasticity of 0.2 to 0.5 (0.15 to 0.3 for peak hours). They also suggested 0.37 can be used for metro in a predicting
scenario. However, it has been observed that it is important to determine the circumstances of the recommended value
before application. In addition, secondary sources of price elasticities should be used with discretion in specific circum-
stances (Hensher, 2008). In the short run, price elasticities of transit demand are low and transit demand can be treated
as an inelastic good, so that a fare increase generally would generate a revenue increase (Paulley et al., 2006).
Regarding the data collection approach, the increasing application of the smart card system in public transit provides a
method to collect consecutive and large amounts of trip data at low expense. Data generated by cards are used widely in trip
pattern analysis and planning. Pelletier et al. (2011) thoroughly reviewed studies on transport smart card data extracted in
the last decade, and categorized the existing research into three areas: strategic, tactical, and operational levels. The majority
of the current research concentrates on trip destination estimation for entry-only smart card systems by setting simple
assumptions of trip patterns (Barry et al., 2007; Chu and Chapleau, 2008; Trépanier et al., 2007), or by incorporating transit
schedule information and station location with geographic information systems (Munizaga and Palma, 2012; Zhao et al.,
2007). In addition, some literature is devoted to investigating transit performance and passengers’ travel behavior. Park
and Kim (2008), Jang (2010) evaluated the travel time, transfers, and time distribution of trips for various modes in Seoul
by extracting smart card data. Morency et al. (2007) calculated the key performance factors of public transit systems, includ-
ing travel speed and time and trip length, with smart card datasets at spatial and temporal levels. Devillaine et al. (2012)
combined smart card databases and land-use information to estimate passengers’ destinations, time duration, and purpose
of activities, and compared people’s travel pattern differences between two cities. The current contribution reveals that
smart card data has great potential for transport evaluation but that exhaustive data have not been used for transit fare pol-
icy assessment.

3. Methodology

As described in the introduction, fare increases would definitely impact metro riders’ trip decisions and the revenue of the
metro operation company. With regard to Beijing Metro, a revenue increase is more important than reducing overcrowding
in the metro system. In addition, demand change analysis generated by fare increases is the first step to assess revenue aug-
mentation. Therefore, only the methodology for revenue evaluation is addressed here.
216 Z.-j. Wang et al. / Transportation Research Part A 77 (2015) 213–224

Currently, Beijing Metro charges a flat fare of 2 RMB for all trips except the Airport Line, which plays a significant role in
the revenue assessment of fare increases, because the new fare policy will apply a distance-based fare structure. Therefore,
the distance between every pair of stations needs to be calculated in order to match the fare structure. In addition, the extent
of fare change for station pairs with different trip lengths is different, and thus the price elasticity varies with the trip length
(Paulley et al., 2006). For instance, in one scenario, the new fare policy charges 10 RMB for a long-distance trip and 3 RMB for
a short-distance trip instead of the original flat fare of 2 RMB. If a fixed value of 0.3 is used as the price elasticity for all
situations, the long-distance demand will shrink to an extremely small value because the fare increases more than three
times. Thus, price elasticity should be treated as a varying factor with respect to trip distance, which is verified by the SP
survey.
Regarding the trip distances used for charging, passengers may not take the shortest path because of the congestion prob-
lem, for example, a less crowded route or a route with fewer transfers. However, if transfers are not recorded by AFC, which
is the case in Beijing, the operation company can only charge in terms of the distance of the shortest path. Therefore, the
shortest path between each pair of stations, rather than the path passengers actually take, should be determined before eval-
uating the impact of fare increases from a flat fare.
By taking into account all these factors, a model can be formulated to analyze the impact of fare increases on revenue for a
large transit network.
X
n X
n
R¼ T 0ij  f ðlij Þ ð1Þ
i¼1 j¼1

where R refers to the predicted revenue after a fare increase for the fare charge alternative f(lij); i and j are the origin station
and destination station of a particular path, respectively, while n is the total station number of the network; lij is the shortest
path length between station i and station j; T 0ij refers to the linked trips from station i to station j after the fare change, which
is an element of the OD matrix at row i and column j; and f(lij) is the distance-based fare function after the fare increase.
The midpoint arc elasticity is defined in Formula 2.
" #, 
T 0ij  T ij f ðlij Þ  f 0
eij ¼ ð2Þ
ðT 0ij þ T ij Þ=2 ðf ðlij Þ þ f 0 Þ=2

where eij denotes the midpoint arc price elasticity, which varies with the fare charged by the new fare policy for the trip from
station i to station j and Tij refers to the current linked trips from station i to station j, which is an element of the OD flow
matrix at row i and column j. f0 is the original flat fare.
Expression T 0ij can be derived from Eq. (2); substituting it into Eq. (1), the predicted revenue can be expressed as

X
n X
n
ð1 þ eij Þf ðlij Þ þ ð1  eij Þf 0
R¼  T ij  f ðlij Þ ð3Þ
i¼1 j¼1
ð1  eij Þf ðlij Þ þ ð1 þ eij Þf 0

As for lij, there are several algorithms to calculate the shortest path between two nodes in a graph, including Dijkstra’s algo-
rithm, the BellmanFord algorithm, the A* search algorithm, the FloydWarshall algorithm, and Johnson’s algorithm. A suit-
able algorithm should be chosen according to the objective of the analysis and the characteristics of the graph analyzed, such
as whether it is a sparse graph or contains negative edge weight. Generally, these algorithms are applied in road or rail net-
works for path finding (Angeloudis and Fisk, 2006; Latora and Marchiori, 2002). The metro network is actually a sparse
graph, which means that a majority of the nodes do not connect with each other. And all the shortest path lengths of the
network should be calculated in this work. Therefore, Johnson’s algorithm, which is capable of generating the shortest paths
of all pairs of vertices in a weighted sparse graph (Johnson, 1977), is adopted to calculate lij.

4. Data sources

According to the model established, the following data are needed to assess the impact of a fare increase on trip demand
and revenue: price elasticity, OD matrix before the fare increase, the shortest path between all station pairs, and fare alter-
natives. The validity of the results is determined largely by the accuracy of the data. Different types of data are obtained
through different approaches. The approaches for data acquisition and processing are discussed in this section.

4.1. Price elasticity of Beijing’s metro demand

According to the model, travel demand before fare increases can be derived from the card data, and all shortest paths are
determinate for a transit network. Thus, price elasticity is the only uncertain parameter, which plays a significant role in the
analysis. Transit price elasticity is influenced by many factors and has a large variation, so the price elasticity drawn from
different contexts should be used cautiously. With regard to this study, acquiring this factor through a survey is the best
method because the transit price elasticity generally used is obtained from developed countries and is not applicable to
Beijing. Zhang and Zhang (2012) provided a rough analysis of the price elasticities based on the annual ridership before
Z.-j. Wang et al. / Transportation Research Part A 77 (2015) 213–224 217

and after each fare change from 1990 to 2008; the elasticities obtained are in the range of 0.058 to 1.501 (midpoint arc
elasticity). However, the calendar annual demand, rather than the demand immediately before and after the fare change, was
used to calculate the elasticity. In addition, the effect of new lines on demand is not taken into account. Thus, those elastic-
ities may not be appropriate for the current analysis. Therefore, an SP survey was carried out to calibrate the parameters,
although the SP method may overestimate transit price elasticity (Hensher, 2008).
Fare increases for trips between different station pairs may be different because the original flat fare policy is replaced by
a distance-based fare structure. Therefore, passengers’ responses to the fare increase, that is, whether to continue riding the
metro or not, probably vary with the trip length, and price elasticity is not a constant. To address this situation, the SP survey
questionnaire was designed to determine the elasticities for each trip length group with a 5 or 10 km increment for each fare
level with a 1 RMB increment. Considering the variation of possible fare alternatives, only a proportion of fare choices were
provided. For instance, the fare for a trip of 5 km would not be more than 6 RMB; thus, fare levels of more than 6 RMB were
not listed in the questionnaire. A sample of 4210 respondents was collected, then, the midpoint arc elasticities for each trip
distance level and fare level were calculated.
The design and implementation of the survey was as follows. To carry out the survey effectively and to avoid response
bias, it was determined to do the survey both online, and on trains throughout the day. The questionnaire was designed
to include two sets of questions: first, general background information, such as gender, income, and weekly trip frequency,
and second, price sensitivity information about the fare level at which passengers would stop riding the metro for the
corresponding travel distance. The questionnaires for both the online survey and the on-train survey were originally the
same. The number of questions was reduced to seven so each respondent could complete it within two minutes.
A pilot survey was conducted and the results revealed that some passengers were unable to judge distances. For instance,
they could not perceive the length of 10 km or evaluate whether the corresponding fare was acceptable. Therefore, the ques-
tionnaire was entirety redesigned and different versions were used for the on-train survey and the online survey. With
respect to the on-train questionnaire, besides personal information, questions about boarding and alighting stations for
the current trip were added. As a result, there were two questions: at which fare level would passengers shift from the metro
to other modes for this trip (choices provided ranged from 3 RMB to 15 RMB), and at which fare level or trip distance ranges
in Table 1 would passengers shift to other modes from the metro. The passenger could evaluate the current trip more accu-
rately because the current trip distance could be drawn between the boarding and alighting station. Thus, the answer to the
first question could be used as a criterion to delete flawed answers to the second question, provided by passengers who were
not good at estimating distance. In addition, the approximate number of segments was added after each of the distance
ranges in Table 1 to assist passengers to estimate the distance. The online questionnaire was the same as the on-train ques-
tionnaire except that the question about the current trip was replaced by a question about the most frequent trip. At the final
count, 3180 on-train responses and 1030 online responses were collected after data cleaning. To account for the daily and
weekly variation of trips, the survey was conducted both at peak times and off-peak times, and on weekdays and at week-
ends, during April 2014. The samples from different time periods were determined according to the demand variation
derived from smart card data.

4.2. Current origin–destination matrices

Currently, Beijing’s metro network is equipped with an AFC system, which is capable of recording information about indi-
vidual trips, including line numbers, station numbers, and times of entry and exit. In addition, other information, such as
card number and type (pass or per journey), is recorded. All records made within one day are listed in a database file accord-
ing to the entry time of the corresponding trip. One notable point is that in the AFC system, transfer stations are allocated
with different numbers according to the different lines they link. The numbers should be merged when counting the demand
between station pairs.

Table 1
Price elasticities of demand for Beijing Metro tabulated from SP survey.

Trip length (km) Fare increased to (RMB)


3 4 5 6 7 8 9 10 11 12
0–5 1.172 1.231 1.283 1.143 – – – – – –
6–10 0.691 0.857 0.925 1.038 – – – – – –
11–15 0.438 0.558 0.732 0.821 – – – – – –
16–20 0.366 0.445 0.581 0.683 – – – – – –
21–25 – 0.266 0.428 0.537 0.611 – – – – –
26–35 – 0.234 0.379 0.469 0.556 0.601 – – – –
36–45 – 0.232 0.276 0.372 0.430 0.514 0.568 – – –
46–55 – – 0.254 0.342 0.392 0.442 0.598 0.543 – –
56–65 – – 0.243 0.270 0.328 0.374 0.403 0.499 0.521 –
P66 – – 0.241 0.263 0.269 0.319 0.340 0.419 0.443 0.583
218 Z.-j. Wang et al. / Transportation Research Part A 77 (2015) 213–224

With structured query language (SQL), one week’s card data recorded by the AFC system during the second quarter of
2014 were processed. The average daily OD matrices with a dimension of 233 were extracted from the data. The overall
number of average daily linked trips is 5.053 million, and Dongzhimen Station has the largest origin and destination trips,
which are 94,540 and 104,278, respectively. The OD pair from Dongzhimen Station to T2 Airport Terminal Station on the
Airport Line has the largest daily demand, namely, 5118 trips.

4.3. Shortest path length matrix

All the shortest paths between stations were calculated by Johnson’s shortest path algorithm with the network map and
the segment length provided by the operating company. As shown in Fig. 2, Beijing Metro has 17 lines and 233 individual
stations, 37 of which are transfer stations. The network consists of 263 segments. The lengths of all the segments in two
directions are the same, so the metro system was treated as an undirected graph. The stations were numbered from 1 to
233, and each segment was assigned with the two station numbers it connects and its own length. Then, all the shortest path
lengths were calculated.

4.4. Fare change alternatives

Considering how much Beijing citizens can be expected to be able to afford and the metro fare policies in other metropoli-
tan cities in China, we proposed three possible new fare alternatives for the fare increase in the near future.

(1) Low fare alternative. There are fares of 2, 3, 4, and 5 RMB for all the trips according to trip length. Specifically, for trips
no longer than 5 km, the fare will not change; for trips from 6 km to 25 km, the fare will be 3 RMB; for the range of 25–
45 km and longer than 45 km, the fares will be 4 and 5 RMB, respectively.
(2) Medium fare alternative. For this scheme, trips within 5 km will be charged the current fare; for trips from 6 km to
15 km and from 16 km to 25 km, the fares will be 3 RMB and 4 RMB, respectively, while for trips longer than
25 km, the fares will be 1 RMB extra for each additional 10 km, with a cap of 8 RMB.
(3) High fare alternative. For this scheme, trips within 5 km will be charged 3 RMB; for trips between 6 km and 25 km, 1
RMB will be added for each additional 5 km, while for trips longer than 25 km, 1 RMB will be charged for each addi-
tional 10 km, with a cap of 12 RMB.

Fig. 2. Map of Beijing Metro. Source: Beijing Metro homepage, for public use, http://www.bjsubway.com/.
Z.-j. Wang et al. / Transportation Research Part A 77 (2015) 213–224 219

Whichever alternative is chosen, the Airport Line fare will not change.
According to the fare level and trip length categories, the price elasticities of demand for each group can be obtained
based on the results of the SP survey, after which the impact on the demand and revenue of the new fare policies can be
evaluated.

5. Results and discussion

We obtained results on elasticities, shortest path length, and trip length distribution, and the impact of fare increases on
demand and revenue. Aspects of these results are discussed in this section.

5.1. Price elasticities of demand for various fare levels and trip distances

Price elasticities of demand are obtained by SP survey in this study. Using responses to questionnaires by passengers on
trains and online, the elasticities at various fare levels and trip distances are tabulated (Table 1).
Regardless of the trip length and fare increase, the price elasticities of demand for Beijing Metro fall in the range from
0.232 to 1.272 with an average of 0.529, which is narrower than the previous results (Zhang and Zhang, 2012).
Generally speaking, metro demand in Beijing is more sensitive to fares for short trips (no longer than 5 km) than for
medium and long ones. For trips longer than 45 km, the demand is relatively inelastic because the metro is a much
better choice for long-distance trips when taking fare and travel time into account. However, the elasticities of long distance
trips seem similar to those of mid-length ones because midpoint arc elasticities are used here. Furthermore, for most
of the trip length groups, absolute values of elasticities are larger for small fare increases than those for large fare increases.
This is also due to the characteristic of midpoint arc elasticity. In fact, the marginal demand decrease is larger for low
prices than for high prices. For instance, the demand decrease is 16.7% when the fare is increased to 3 RMB from 2 RMB
for trips between 11 km and 15 km, while it is 10.14% when the fare is increased to 6 RMB from 5 RMB for trips of the same
length.
According to the literature, the metro price elasticity of demand generally referred to is 0.3 or less (Paulley
et al., 2006), which is much lower (absolute value) than what this study finds. This can be ascribed partially to over-
estimation of the SP survey, especially for Beijing Metro, because citizens of Beijing have enjoyed the low flat fare for
a long time and any fare increase would be strongly opposed. Even though the purpose of this survey as a scientific
study was emphasized, some respondents regarded the survey as a government initiative. They claimed they would
rather stop using the metro at a low price so as to influence government policymaking on metro fares. Therefore, the
elasticity was overestimated. In addition, the larger elasticity can be attributed to lower income in Beijing than in
developed countries because a higher fare level relative to people’s income results in a larger price elasticity
(Collins, 1982).

5.2. Shortest path and trip distance distribution

All the shortest paths of Beijing Metro network are calculated using the method and data illustrated in Sections 3 and 4.
There are 52,056 shortest paths. The minimum path is 0.42 km, between Nanlishilu Station and Fuxingmen Station on Line 1,
and the maximum path is 86.13 km with five transfers, between Fengbo Station on Line 15 and Suzhuang Station on
Fangshan Line. The average shortest path length is 23.56 km. The distribution of the shortest path length is illustrated in
Fig. 3. More than 50% of the shortest paths have lengths in the range of 8–26 km. The 75 percentile is 32 km. Only 4.17%
of the overall station pairs have a shortest path length of more than 50 km, although the range of 50–86 km is 41.9% of
the total length span.
Combining the shortest path length matrix and the demand matrix, the trip length distribution of Beijing Metro was
generated, as illustrated in Fig. 4. In contrast to the shortest path length distribution, the trip length distribution is more
concentrated on short–medium distances than the shortest path length distribution. More than 70% of the ridership has a
trip length in the range of 6–22 km. Only 10% of the total daily ridership has a trip distance longer than 30 km. In addition,
the average trip distance is 15.39 km, which is much less than the average shortest path length. This finding indicates that
specific consideration must be given to short and medium distance trips when launching new fare policies. The actual trip
length distribution may slightly differ from the figure because some passengers may not take the shortest paths due to
congestion or inconvenient interchange.
The trip distance distribution pattern of Beijing Metro is very close to that of London rail transit, as extracted from Oyster
card data (Chan, 2007) and that of Seoul subway derived from trip data (Kim et al., 2009). Even though the urban form and
transit network are very different in the aforementioned three cities, the distances of the metro trips are primarily 10–15 km
in each city. In fact, this feature of metro trip distance distribution is consistent with the function of mass transit: focusing on
medium–long distance trips rather than short distance trips. It is different from buses, which catering for short distance trips
or act as a feeder to mass transit.
220 Z.-j. Wang et al. / Transportation Research Part A 77 (2015) 213–224

1800

1600

1400

1200

Frequency
1000

800

600

400

200

0
0 10 20 30 40 50 60 70 80 0
Shortest path length (km)

Fig. 3. Shortest path length distribution of Beijing Metro.

5
x 10
6

4
Ridership

0
0 10 20 30 40 50 60 70 80 0
Trip length (km)

Fig. 4. Trip length distribution of Beijing Metro.

5.3. The effect of the three new fare alternatives on demand and revenue

Using data of ridership, shortest path lengths, elasticities, and new fare alternatives, ridership and revenue after fare
change on a daily basis can be predicted by Eq. (3) (Table 2). The results show that the low fare alternative can produce a
26.26% revenue increase with a 15.24% ridership decrease. For the medium fare scheme, the revenue increase is 36.21%,
which is considerably higher than the low fare scheme, while the ridership decrease is only slightly greater than that of
the low fare scheme. The outcome of the high fare alternative is not desirable as the metro system would lose almost half
of its ridership while the revenue would increase only 1% more than that of the medium scheme. The differences can be
explained by the elasticities and the distribution of trip length. As illustrated in Fig. 4, trip lengths of the Beijing metro
are primarily in the range of 6–22 km. For this range, the elasticities increase rapidly with fare levels higher than 4 RMB.
Thus, it is not a good choice to increase charges significantly in this trip length range.
If the fare increase is measured by average fare per trip, the average midpoint arc elasticities of demand for the three
alternatives can be established. The elasticities for the low and medium fare policies are very close, while those of the high
scheme are much larger. All the averaged elasticities are larger than the generally quoted metro price elasticity, which is
0.3 or less.
Z.-j. Wang et al. / Transportation Research Part A 77 (2015) 213–224 221

Table 2
Change of daily ridership and revenue under the three fare alternatives.

Alternatives Ridership (106 Revenue (106 Ridership decrease Revenue increase Average fare Average
trip) RMB) (percentage) (percentage) (RMB) elasticity
Low 4.693 14.86 15.24 26.26 3.17 0.419
Medium 4.449 16.03 19.65 36.21 3.60 0.422
High 2.919 16.23 47.29 37.89 5.56 0.693

As discussed in this section, the medium fare alternative is the best choice. If implemented, annual revenue would be
5.851 billion RMB, which is 1.555 billion RMB more than that under the current fare policy. In addition, a ridership decrease
of about one fifth could mitigate the overcrowding problem but would not jeopardize the efficiency of the mass transit sys-
tem. Under this new fare policy, the predicted annual demand would be 1.623 billion linked trips, and Beijing Metro would
continue to be one of the most efficient systems in the world in terms of linked trip load.

6. Conclusions

This research conducted an ex-ante assessment of the impact of fare increases on demand and revenue with exhaustive
smart card data, an SP survey, and the network approach for transferring a flat fare strategy to a distance-based fare struc-
ture. The methodology can be applied to other transit systems equipped with an AFC system.
The analysis presented in this work demonstrated that the use of a smart card dataset has great potential for transit
demand management and operational policymaking. The advantage of smart card data is that it is exhaustive and records
all trip origins. In addition, trip ends are archived by some systems or can be estimated. Thus, accurate trip distances and
boarding times are available to allow more detailed analysis of demand management, for instance, assessing the influence
of distance-based fare structure change. We can address each trip distance group with the corresponding elasticity rather
than using a general elasticity for the total demand, in which case the range of error could be large.
The comparison of trip distance distribution between Beijing Metro and the mass transit systems in other metropolitan
cities illustrated that the trip distance distribution of established metro networks shares a common feature: a large propor-
tion of trip distances tend to be around or a little longer than 10 km. This characteristic should be considered when designing
fare policy for specific purposes. In the case of Beijing, trips with a distance of around 10 km are very sensitive to fare
increases, with elasticities of about 0.8, and so, a fare of 3 RMB for trips with a distance from 6 km to 15 km could yield
a good outcome for both revenue boosting and crowding mitigation, which is the purpose of fare change for Beijing Metro.
Projection analysis of transit fare adjustment is always confronted with the difficulty of determining feasible elasticities,
especially when there is an absence of historical data. Often, an SP survey is adopted as an alternative. However, with regard
to fare increase elasticity of transit demand, participants of an SP survey would exaggerate their sensitivity because of pas-
sengers’ aversion to price increases. Additionally, underestimation of the difficulty of shifting to other modes may also con-
tribute to the overestimation. In this case, the range of the elasticities is 0.232 to 1.272, the absolute values of which are
relatively bigger than the generally referenced value of 0.3 or lower.
The fare increase for Beijing Metro will be implemented in the near future. We plan to collect and compare the short-term
and long-term elasticities of fare increases to those of this SP survey.
This study did not address the variety of elasticity during different periods of the day or week, although the SP survey
sample covered all time periods. In fact, the elasticities of peak-hour and weekday demand are smaller than the elasticities
of demand during other times. This implies that a complex fare structure, including a congestion fare, could be implemented
for demand management and revenue increases. Future studies could be dedicated to this topic. That will help the govern-
ment to implement an optimum fare structure according to different trip times and purposes.

Acknowledgements

This work is supported by the Fundamental Research Funds for Beijing Jiaotong University (grant numbers: 2014JBM077)
and National Natural Science Foundation of China (grant numbers: 51408029). We thank Beijing Subway Operation
Corporation Limited and Beijing MTR Corporation Limited for data supporting. In addition, we are grateful to the China
Scholarship Committee for enabling the author to visit University College London for one year, which has been very bene-
ficial for this work. All this support is greatly appreciated.
222 Z.-j. Wang et al. / Transportation Research Part A 77 (2015) 213–224

Appendix A
Z.-j. Wang et al. / Transportation Research Part A 77 (2015) 213–224 223

Appendix B
224 Z.-j. Wang et al. / Transportation Research Part A 77 (2015) 213–224

References

Angeloudis, P., Fisk, D., 2006. Large subway systems as complex networks. Physica A 367, 553–558.
Association of Train Operating Companies (ATOC), 2002. Passenger Demand Forecasting Handbook, fourth ed. Passenger Demand Forecasting Council,
London.
Barry, J., Newhouser, R., Rahbee, A., Sayeda, S., 2007. Origin and destination estimation in New York City with automated fare system data. Transport. Res.
Rec. 1817, 183–187.
Bresson, G., Dargay, J., Madre, J., Pirotte, A., 2003. The main determinants of demand for public transport: a comparative analysis of England and France
using shrinkage estimators. Transport. Res. Part A 37, 605–627.
Chan, J., 2007. Rail Transit OD Matrix Estimation and Journey Time Reliability Metrics using Automated Fare Data. Master’s thesis, Massachusetts Institute of
Technology.
Chen, X., Lin, G., 2005. Evaluation analysis on an integrated fare initiative in Beijing. J. Public Transport. 8 (3), 33–55.
Chu, K.K., Chapleau, R., 2008. Enriching archived smart card transaction data for transit demand modeling. Transport. Res. Rec. 2063, 63–72.
Collins, P.H., 1982. The effects of recent London transport fares changes. In: Proceedings of Seminar Held at the PTRC 10th Summer Annual Meeting, 12–15
July 1982, University of Warwick, PTRC, London.
Dargay, J., Hanly, M., 2002. The demand for local bus services in England. J. Transport. Econ. Policy 36, 73–91.
Devillaine, F., Munizaga, M., Trépanier, M., 2012. Detection of activities of public transport users by analyzing smart card data. Transport. Res. Rec. 2276, 48–
55.
Dygert, P., Holec, J., Hill, D., 1977. Public Transportation Fare Policy. Sponsored by Office of the Secretary, US Department of Transportation. Peat, Marwick,
Mitchell and Co, Washington DC.
Evans, J.E. IV, 2004. Transit Pricing and Fares: Traveler Response to Transportation System Changes. Publication TCRP Rep. 95, Chapter 12, Transportation
Research Board, Washington, DC.
Finance Bureau of Beijing (FBB), 2014. Report on Budget Execution of 2013 and Budget Draft of 2014 of Beijing. <http://www.mof.gov.cn/zhuantihuigu/
2014yshb/201402/t20140219_1044590.html> (19.02.14).
Goodwin, P.B., 1992. A review of new demand elasticities with special reference to short and long run effects of price charges. J. Transport. Econ. Policy 26,
155–170.
Hensher, D., 2008. Assessing systematic sources of variation in public transport elasticities: some comparative warnings. Transport. Res. Part A 42, 1031–
1042.
Holmgren, J., 2007. Meta-analysis of public transport demand. Transport. Res. Part A 41, 1021–1035.
Jang, W., 2010. Travel time and transfer analysis using transit smart card data. Transport. Res. Rec 2144, 142–149.
Johnson, D.B., 1977. Efficient algorithms for shortest paths in sparse networks. J. ACM 24 (1), 1–13.
Kim, C., Choi, C.G., Cho, S., Kim, D., 2009. A comparative study of aggregate and disaggregate gravity models using Seoul metropolitan subway trip data.
Transport. Plan. Technol. 32 (1), 59–70.
Kremers, H., Nijkamp, P., Rietveld, P., 2002. A meta-analysis of price elasticities of transport demand in a general equilibrium framework. Econ. Model. 19,
463–485.
Latora, V., Marchiori, M., 2002. Is the Boston subway a small-world network? Physica A 314, 109–113.
Litman, T., 2004. Transit price elasticities and cross-elasticities. J. Public Transport. 7 (2), 37–58.
Morency, C., Trépanier, M., Agard, B., 2007. Measuring transit use variability with smart-card data. Transport Policy 14 (3), 193–203.
Munizaga, M.A., Palma, C., 2012. Estimation of a disaggregate multimodal public transport origin–destination matrix from passive smartcard data from
Santiago, Chile. Transport. Res. Part C 24, 9–18.
Park, J.Y., Kim, D.J., 2008. The potential of using the smart card data to define the use of public transit in Seoul. Transport. Res. Rec. 2063, 3–9.
Paulley, N., Balcombe, R., Mackett, R., Titheridge, H., Preston, J.M., Wardman, M.R., Shires, J.D., White, P., 2006. The demand for public transport: the effects of
fares, quality of service, income and car ownership. Transport Policy 13, 295–306.
Pelletier, M., Trépanier, M., Morency, C., 2011. Smart card data use in public transit: a literature review. Transport. Res. Part C 19, 557–568.
Sharaby, N., Shiftan, Y., 2012. The impact of fare integration on travel behavior and transit ridership. Transport Policy 26, 63–70.
Trépanier, M., Tranchant, N., Chapleau, R., 2007. Individual trip destination estimation in a transit smart card automated fare collection system. J. Intell.
Transport. Syst. 11 (1), 1–14.
Webster, F.V., Bly, P.H., 1980. The Demand for Public Transport. Report of an International Collaborative Study. Transport and Road Research Laboratory,
Crowthorne, Berkshire.
Zhang, H., Zhang, W., 2012. Analysis of price elasticity for Beijing’s urban rail transit. Foreign Invest. China 2012 (8), 210–211.
Zhang, G., Liu, Y., Xu, X., Chen, F., 2012. Innovation of Investment and Financing Theory and Practice of Beijing’s Urban Rail Transit. Tsing Hua University
Press, Beijing.
Zhao, J., Rahbee, A., Wilson, N.H.M., 2007. Estimating a rail passenger trip origin–destination matrix using automatic data collection systems. Comput.-Aid.
Civil Infrastruct. Eng. 22 (5), 376–387.
Zureiqat, H.M., 2008. Fare Policy Analysis for Public Transport: A Discrete–Continuous Modeling Approach using Panel Data. Master’s Thesis. Massachusetts
Institute of Technology.

You might also like