Available online at [Link].
com
ScienceDirect
Procedia - Social and Behavioral Sciences 96 (2013) 148 – 155
13th COTA International Conference of Transportation Professionals (CICTP 2013)
Evaluation of Public Transportation Operation Based on Data
Envelopment Analysis
Jiabin Lia, Xumei Chenb,c, Xin Lib,c, Xiucheng Guoa*
a
School of Transportation, Southeast University, Nanjing, Jiangsu, 210096, China
b
MOE Key Laboratory for Urban Transportation Complex Systems and Theory and Technology, Beijing Jiaotong University, Beijing,
100044, China
c
School of Traffic and Transportation, Beijing Jiaotong University, Beijing, 100044, China
Abstract
This paper presented method on evaluating the performance of bus routes within a public transportation system using revised
DEA method and sensitivity analysis of indexes. First, based on the analysis of the operation of public transportation,
passenger load rate, service reliability, average dwell time and average running speed were chosen as output indexes, a virtual
ng public
transportation system and the improvement suggestions are put forward. Finally, the results which show that the operation in
off-peak period are better than that in peak period, and average running time and service reliability are the key factors
influencing performance, conform to the reality of the city, thus validating the practicality of the method. In sum, this study is
helpful for improving the operation of public transportation system.
© 2013The
© 2013 TheAuthors.
Authors. Published
Published by Elsevier
by Elsevier B.V.
Ltd. Open access under CC BY-NC-ND license.
Selection and/or
Selection and peer-review
peer-review under responsibility
under responsibility of ChineseofOverseas
ChineseTransportation
Overseas Transportation Association (COTA).
Association (COTA).
Keywords: public transportation; evaluation indexes; data envelopment analysis (DEA)
1. Introduction
Developing public transportation systems is one of the main ways to reduce congestion. A
In general, the level of service of public transportation in China is
relatively low, and there are some significant problems in the public transportation system of metropolis, such as
* Corresponding author. Tel.: +86-25-8379-5528; fax: +86-25-8379-5528.
E-mail address: nbamagic@[Link]
1877-0428 © 2013 The Authors. Published by Elsevier Ltd. Open access under CC BY-NC-ND license.
Selection and peer-review under responsibility of Chinese Overseas Transportation Association (COTA).
doi:10.1016/[Link].2013.08.020
Jiabin Li et al. / Procedia - Social and Behavioral Sciences 96 (2013) 148 – 155 149
Beijing. For instance, during the off-peak period, the passenger load rate is relatively low due to the high
ing time are relatively long, and features frequent
bunching and low speed. Thus, evaluating and analyzing the public transportation operation accurately and
solving existing key problems identified are crucial issues, which is beneficial for the planning, design and
operation of public transportation system.
Currently, the above problems have encouraged researches focusing on the evaluation of public transportation
operation. The main methods include fuzzy evaluation, analytic hierarchy process (AHP), BP artificial neural
network, and clustering analysis, etc. Among them, fuzzy evaluation and AHP may result in inaccurate results
with weighted subjectively. In addition, BP artificial neural network and clustering analysis require a great deal of
statistical data.
Data Envelopment Analysis (DEA) is a nonparametric method, which emphasizes the objectivity and is easy to
use of public
transportation systems due to the problems of data collection, index selection and quantification.
This paper uses the revised DEA method to evaluate the operation of bus routes with GPS and passenger flow
data. Bus routes which need to be improved are identified, and suggestions for improvements s are proposed.
2. Methodology
2.1. Index selection
Index selection, which is the key step to evaluate the performance of bus routes accurately, need to follow
some basic principles as follows:
Select indexes that can be obtained easily in practice.
Select indexes that can reflect the characteristics of performance of bus routes comprehensively and
objectively.
Select key indexes as few as possible to reveal the main features of bus routes.
In addition, the input and output of DEA also should meet the following requirements:
The input/output values which should be positive.
The number of inputs and outputs should not exceed the number of half of the evaluation object.
The public transportation system is complex, involving vehicles, road conditions, and time-varying passenger
flow. The evaluation index system of public transportation was established based on the current situation of
public transportation system in China. Six aspects safety, speediness, punctuality, comfort, economy and
are included as shown in Fig. 1.
Evaluation Index System of Public Transportation Operation
Safety Speediness Punctuality Comfort Economy Convenience
·Vehicle Accident Rate ·Average Running Speed ·Scheduling Adherence ·Passenger Load Rate ·Travel Cost ·Hours of Service
·Passenger Accident Rate ·Average Dwell Time ·Service Reliability ·Full House Rate ... ·Transfer Time
... ·Average Travel Time Coefficient ·Seat Comfort ...
... ... ·Running Smooth Degree
...
Fig. 1. Evaluation index system of public transportation operation
150 Jiabin Li et al. / Procedia - Social and Behavioral Sciences 96 (2013) 148 – 155
Some of the above indexes can be directly quantified, while others cannot. After a comprehensive
consideration of data acquisition and index calculation, four indexes were selected as follows:
Service reliability. It attracts attention, which refers to headway fluctuations of the
buses. Moreover, even though scheduling adherence is also a key index, it isn't selected since the data is
difficult to obtain and passengers mainly focus on headway fluctuations rather than headways in the situation
that the departure interval of urban buses are generally less than 10 minutes.
Average running speed. speediness.
Average dwell time. It reflects the buses stop time at stations.
Passenger load rate. It reflects the economy of bus operation and the comfort of passengers. A lower one will
be accepted by passengers, and a higher one means more income for operators. It is an important index which
can be obtained easily related with sanitary conditions, seat comfort, and riding comfort.
2.2. Model
DEA is an optimization-based technique widely used to measure relative efficiencies. Due to non-parametric
model Characteristics and its ability to combine multiple inputs and outputs, DEA has been found an effective
tool if used appropriately. A few of the characteristics of DEA are:
DEA does not need the index weight be given in advance, and it can avoid subjective evaluation.
DEA does not require an assumption of a function of inputs and outputs.
It is convenient to deal with the inputs and outputs data, for these data can have different units.
For the traditional DEA method, inputs and outputs can be related with each other by certain economic
relations. However, the indexes of public transportation operation don t have the similar relationship. For this
reason, all indexes were defined as outputs, and a virtual input was defined.
For the self-evaluation of the traditional DEA method, the evaluation objects will search the most favorable
weight combination for themselves. As a result, cross model was introduced in order to make the results more
accurate.
Further, understanding the influence of each index is important. It not only helps to understand the change of
evaluation results due to the change of indexes, but also helps to provide improvement suggestions. Therefore, the
sensitivity analysis of indexes was conducted in this paper.
2.2.1. Operation evaluation model
Assume there are R decision-making units (DMUs), and each DMU has one input and four outputs. The DMUk
N N
the input is denoted by ; L (i=1). Four outputs are denoted by < M (j=1 2
3 and 4). The purpose of DEA is to construct a frontier that envelops all data points representing the efficiency
of all DMUs under consideration and to calculate the efficiency score for each DMUk. The problem of
determining the efficiency score Ek for DMUk can be formulated in the following linear programming problem:
Q
X N< N
M M M
PD[ ( . P
Y N; N
L M L
Q
X MN< MU
V W M
P
U 5 (1)
L
Y MN ; LN
XMN M
N
Y L L
Jiabin Li et al. / Procedia - Social and Behavioral Sciences 96 (2013) 148 – 155 151
where X MN
is a decision variable representing the weight for output j, under the goal of maximizing Ek;
Y LN
is a decision variable representing the weight for input i, under the goal of maximizing Ek;
and is a non-Archimedean number.
The steps of using the cross model are as follows:
U
Step 1: calculate the most favorable weight combination X M Y LU for each DMUr with the above model.
Step 2: calculate R-1 efficiency scores ( NU with other R-1 most favorable weight combinations X MU Y LU
k). The formula is shown as follows:
Q
X MU< MN
( NU M
P
(2)
L
Y LU ; LN
where ( NU is the efficiency score of DMUk , which is calculated using the most favorable weight combinations
X MU Y LU of DMUr.
Step 3: calculate the arithmetic average of ( NU as the cross efficiency score of DMUk, CEk:
5
( NU
&( N U
U N (3)
5
Step 4: calculate the average efficiency score of DMUk, AEk:
5 &( N ( N
$( N (4)
5
The value of AEk ranges from 0 to 1. And greater value of AEk implies better operation of DMUk. In contrast,
smaller value of AEk means worse operation of DMUk.
2.2.2. Sensitivity analysis of indexes model
Assume D is an evaluation index system, and Dj is defined as the system which removes the index j. Zk(D)
represents the efficiency scores of DMU k which is calculated with D, and Zk(Dj) represents the efficiency scores
of DMUk which is calculated with Dj. Then the influence coefficient of index j, Sk(Dj), can be calculated as
follows:
=N ' = N 'M
6 N 'M N 5 M (5)
= N 'M
152 Jiabin Li et al. / Procedia - Social and Behavioral Sciences 96 (2013) 148 – 155
6 N ' M is obtained by adding the influence coefficient of index j, Sk(Dj). And greater value of 6 N 'M
implies bigger influence of index j (i.e. higher sensitivity). In contrast, smaller value of 6 N ' M means smaller
influence of index j (i.e. lower sensitivity).
3. Case Study
The bus routes 16, 87, 105 in Beijing were chosen as study cases in this paper. The data on each Thursday were
collected from March 8, 2012 to March 22, 2012 by GPS and manual counting. During the study period, no major
weather issues that might have affected the operation of public transportation systems occurred.
3.1. Data
With the processing of GPS data and passenger flow data, corresponding values of operation evaluation
indexes were obtained, as indicated in Table 1.
Table 1. Statistics of operation evaluation indexes
Average Average
Line/ Passenger Service
Period dwell time running speed
Direction load rate reliability
(s) (km/h)
16/A 1.06 1.77 29.18 16.72
Peak
16/B 0.87 5.34 22.53 18.18
16/A 0.50 7.76 18.05 20.06
Off-peak
16/B 0.44 5.82 18.65 21.15
87/A 0.74 2.41 24.41 16.62
Peak
87/B 0.70 2.64 32.60 18.57
87/A 0.64 7.91 23.89 20.93
Off-peak
87/B 0.53 3.54 23.92 22.21
105/A 0.81 2.71 43.50 9.63
Peak
105/B 1.06 1.94 34.63 12.90
105/A 0.68 3.87 33.45 13.65
Off-peak
105/B 0.68 2.78 31.75 14.47
Peak 0.87 2.80 31.14 15.44
Average
Off-peak 0.58 5.28 24.95 18.75
Considering the principles of selecting indexes (i.e. smaller input indexes are better, whereas greater output
indexes are optimal ), two of four indexes were transformed as follows:
Passenger load rate. The problem concerning what kind of passenger load rate is the best has been studied by
previous researches without any firm conclusion being reached. Based on field data, the most appropriate
passenger load rate is 0.6 in this paper. And passenger load rate index was introduced, and its transformation
process was as follows: passenger load rate was defined as a passenger load rate index as A. When a = 0.6, A
= 100; When a = max |a-0.6|, A = 0. Moreover, the values were determined in a linear relationship.
Average dwell time. Average dwell time index was introduced, its transformation process was as follows: we
define t as average dwell time and T as average dwell time index. When t = max t, T = 0; When t = min t, T =
100. The values were determined in a linear relationship as well.
Besides, in this paper defined a virtual input index, and its value is 1. Accordingly, Table 2 displays the values
of input-output indexes.
Table 2. Data of input output indexes
Jiabin Li et al. / Procedia - Social and Behavioral Sciences 96 (2013) 148 – 155 153
Route/
Period Y1* Y2* Y3* Y4* X1*
Direction
16/A 1.50 1.77 56.28 16.72 1.00
Peak
16/B 42.06 5.34 82.41 18.18 1.00
16/A 77.06 7.76 100.00 20.06 1.00
Off-peak
16/B 64.50 5.82 97.64 21.15 1.00
87/A 69.32 2.41 75.00 16.62 1.00
Peak
87/B 79.61 2.64 42.81 18.57 1.00
87/A 92.04 7.91 77.04 20.93 1.00
Off-peak
87/B 84.48 3.54 76.94 22.21 1.00
105/A 54.79 2.71 0.00 9.63 1.00
Peak
105/B 0.00 1.94 34.87 12.90 1.00
105/A 83.66 3.87 39.50 13.65 1.00
Off-peak
105/B 83.44 2.78 46.16 14.47 1.00
*denotes that Y1 is the Passenger load rate index, Y2 is the Service reliability, Y3 is the Average dwell time
index, Y4 is the Average running speed, and X1 is the virtual input index.
3.2. Results and analysis
The software MATLAB was used to solve the formulated linear programming problems. Table 3 displays the
results indicating the operational performance and the sensitivity of indexes for each route/direction. It can be
observed that the operation for each route/direction in its off-peak period is better than that in its peak period.
Route 87/A is the best during the off-peak period, and its AEk is 0.954, which is close to 1. Whereas route 105/B
is the worst during the peak period, and its AE k is 0.268. Therefore, in depth analysis on route 105 is conducted
and suggestions for improvement are proposed. Furthermore, indexes Y4 and Y2 are relatively larger for route
105, and their 6 N ' M are 1.146 and 0.872. Accordingly, average running speed and service reliability, which
represented by indexes Y4 and Y2, can be identified as the key factors that influence performance.
According to the calculated results (Table 3) and statistics of operation evaluation indexes (Table 1),
conclusions can be obtained as follows.
Firstly, during the off-peak period, the average running speed of route, which is 13.65km/h and 14.47km/h for
105/A and route 105/B respectively, are far below the average of all bus routes, 18.75 km/h. During the peak
period, the index values of this route, which is 9.63km/h and 12.90km/h respectively, are below the average of all
bus routes, 15.44 km/h.
Secondly, during the off-peak period, the service reliability which are 3.87 and 2.78 for line 105/A and route
105/B respectively, are smaller than the average of bus routes, 5.28. During the peak period, the index values of
this route, which are 2.71 and 1.94 respectively, are below the average of bus routes, 2.80.
Thirdly, i value of service reliability is lower than 2, bunching would occur, because the
fluctuation of gap between arrive interval and departure interval has reached 50% of the departure interval.
Bunching may easily happen for route 105 during the peak period.
Fourthly, the reason leading to the above results has two aspects. One reason is that the layout of route 105 has
nearly one-third of its route on collector streets and minor roads, where the road conditions are bad, with narrow
roads and serious mixed traffic flow. Another reason is that the layout of route 105 is mainly inside the 2nd Ring
and goes through the political and financial center in Xicheng district, where there are many signal intersections.
Table 3. The evaluation results for bus routes
Route/ Sensitivity analysis of indexes Operation evaluation
Period
Direction Remove Y1 Remove Y2 Remove Y3 Remove Y4 AEk Rank
16/A 0.000 0.000 0.000 0.338 0.364 11
Peak
16/B 0.008 0.000 0.017 0.048 0.658 9
16/A 0.000 0.000 0.019 0.000 0.911 2
Off-peak
16/B 0.000 0.000 0.017 0.024 0.844 4
154 Jiabin Li et al. / Procedia - Social and Behavioral Sciences 96 (2013) 148 – 155
87/A 0.137 0.062 0.070 0.000 0.687 6
Peak
87/B 0.129 0.050 0.000 0.015 0.744 5
87/A 0.000 0.000 0.000 0.000 0.954 1
Off-peak
87/B 0.000 0.000 0.000 0.057 0.879 3
105/A 0.000 0.328 0.000 0.000 0.398 10
Peak
105/B 0.000 0.000 0.000 0.665 0.268 12
105/A 0.011 0.235 0.000 0.000 0.672 8
Off-peak
105/B 0.011 0.197 0.000 0.000 0.678 7
6 N 'M 0.296 0.872 0.123 1.146 -- --
Therefore, improving the running environment is essential, and can be realized through the following two
measures:
Provide the bus route with transit signal priority (TSP) and fewer intersections if possible.
Separate the mixed traffic flow. Deploy exclusive bus lanes when road conditions allow.
Additionally, implementing real-time scheduling actively is also helpful to ensure better service quality.
4. Conclusions
In this paper, the evaluation index system of public transportation operation is established by selecting four key
indexes, including passenger load rate, service reliability, average dwell time, and average running speed. Then
operation evaluation model are built and sensitivity analysis of indexes model are conducted. The case study
shows that the method has good applicability, and can be used to improve the operation of public transportation
system. Nevertheless, the DEA method has relative evaluation result. Hence, the route with a good evaluation
result may still require improvements. This problem can be solved by adding evaluation objects or determining
the index expectations using the Delphi method.
Acknowledgements
This research was supported by Program for New Century Excellent Talents in University: NCET-12-0763.
References
Jiu, B. X. (2007). Outlook of urban transportation development in China. Urban Transport of China, 5(5): 6-12.
Chu, X., G. J. Fielding, & B. W. Lamar. (1996). Measuring transit performance using data envelopment analysis. Transportation Research
Part A , 26(3): 223-230.
Viton, P. A. (1998). Changes in multi-mode bus transit efficiency, 1988-1992. Transportation, 25(1): 1-21.
Nakanishi, Y. J. & J. R. Norsworthy (2000). Assessing efficiency of transit service. 2000 IEEE International Engineering Management
Society Conference. Albuquerque, US: the IEEE Inc.
Matthew, G. K. (2003). Investigating transit production and performance a programming approach. Transportation Research Part A, 37:225-
250.
Matthew, G. K. (2004). A DEA approach for evaluating the efficiency and effectiveness of urban transit systems. European Journal of
Operational Research, 152: 354-364.
Attah, K. B. (2004). The technical efficiency of Canadian urban transit systems. Transportation Research Part E, 40: 401-416.
Chintan, S., T. Konstantinos, & T. Dusan (2007). Performance evaluation of bus routes: A provider and passenger perspective.
Transportation Research Part E, 43: 453-487.
Yong, L. & L. Lin (2009). Performance evaluation of bus lines with data envelopment analysis and geographic information systems.
Computers, Environment and Urban Systems, 33(4): 247-255.
Shen, X. J. Chen, J. & Wang, C. (2008). Evaluation for efficiency of city public transportation based on DEA model. Modern
Transportation Technology, 5(6):76-79.
Jiabin Li et al. / Procedia - Social and Behavioral Sciences 96 (2013) 148 – 155 155
Zhang, B., Chen, X. W., & Ge, H. W. (2008). Study on planning and design theory for external passenger transport hub of metropolis. Urban
Public Transport, 3:47-50.
Yan, Y. D., Guo, X. C., & Liu, C. P. (2010). Performance evaluation of bus transit routes based on the BCC model. Proceedings of the
Conference on Traffic and Transportation Studies. Kunming, China: the ASCE Inc.
Jian, Z. F. (2007). Decision analysis and management. Beijing: Tsinghua University Press.
Lin, J, Wang, P, & Barnum, D. T. (2008).A quality control framework for bus schedule reliability. Transportation Research Part E, 33(4):
1086-1098.
Zhang, H. J. & Zhao, H. L. (2007). Urban public traffic evaluation index system analysis. Private science and technology, 5, 134.
Cooper, W. W., SeifordL, M., & Tone K (2000). Data envelopment analysis. Boston: Kluwe Academic Publishers.
Lv, S., Tian, F., & Wang, J. Y. (2010). Assessment of performance of feeder buses based on DEA technique. Journal of Highway and
Transportation Research and Development, 27(8):96-102.
Jiang, X. (2011). Evaluation of public transportation operation in metropolis based on multi-source data. Beijing Jiaotong University.
