Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

Transportation Letters

The International Journal of Transportation Research

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/ytrl20

A dynamic network data envelopment analysis


cross-efficiency evaluation on the benefits of bus
transit services in 33 Chinese cities

Meng Liu, Chunqin Zhang, Wenbin Huang, Mengmeng Wang & Guangnian
Xiao

To cite this article: Meng Liu, Chunqin Zhang, Wenbin Huang, Mengmeng Wang & Guangnian
Xiao (2023): A dynamic network data envelopment analysis cross-efficiency evaluation
on the benefits of bus transit services in 33 Chinese cities, Transportation Letters, DOI:
10.1080/19427867.2023.2198109

To link to this article: https://doi.org/10.1080/19427867.2023.2198109

Published online: 03 Apr 2023.

Submit your article to this journal

Article views: 90

View related articles

View Crossmark data

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


https://www.tandfonline.com/action/journalInformation?journalCode=ytrl20
TRANSPORTATION LETTERS
https://doi.org/10.1080/19427867.2023.2198109

A dynamic network data envelopment analysis cross-efficiency evaluation on the


benefits of bus transit services in 33 Chinese cities
Meng Liua, Chunqin Zhanga, Wenbin Huanga, Mengmeng Wanga and Guangnian Xiaob
a
School of Civil Engineering and Architecture, Zhejiang Sci-Tech University, Hangzhou, China; bCollege of Economics & Management, Shanghai Maritime
University, Shanghai, China

ABSTRACT ARTICLE HISTORY


Our objective in this paper is to evaluate the benefits of bus transit. To obtain the more accurate and highly Received 14 November 2022
discrimination results, this study establishes an evaluation indicator system from a stakeholder perspective Accepted 12 March 2023
and proposes a novel combined method consisting of dynamic network data envelopment analysis, cross- KEYWORDS
efficiency evaluation, and Shannon entropy aggregation method. An empirical study of bus transit systems China; bus transit; benefit
in 33 key Chinese cities from 2016 to 2019 is provided. The results show that the benefit of bus transit in most evaluation; dynamic network
key cities was ineffective. In contrast to production efficiency, focusing on improving service effectiveness is data envelopment analysis;
more conducive to improving the benefits of bus transit. Small cities have higher benefits than large cities. neutral cross-efficiency
Meanwhile, the benefits of bus transit vary geographically. This evaluation outcome provides a more evaluation; Shannon entropy
appropriate basis for decision-making related to the benefits of public transit services, as well as for
operational and management studies.

Introduction Therefore, the external environmental factors should be included


in the evaluation indicator system.
The increased travel demand induced by urbanization has triggered
To obtain more accurate and highly discriminant evaluation
a boost of public transit around the world, which aims to move
results of bus transit systems, this paper proposes a novel com­
urban mobility toward sustainability. Public transit service has
bined method consisting of dynamic network data envelopment
a great impact on society, economy, quality of travel, and citizen
analysis (DNDEA), cross-efficiency evaluation (CREE), and
welfare (Omrani, Shafaat, and Alizadeh 2019; Zhang et al. 2019;
Shannon entropy aggregation method. Additionally, this paper
Zhang, Hu, and Lu 2022). Over the past two decades, the Chinese
establishes a system of evaluation indicator from a stakeholder
government has paid great attention to urban public transit. The
perspective. The advantage of our proposed method is that it
passenger ridership of urban bus transit in 33 key cities of China has
reached 19.7 billion person-time in 2020 (Ministry of Transport of avoids the limitation of self-evaluation of data envelopment
the People’s Republic of China 2020). Over-expansion can result in analysis (DEA) and evaluates the benefits of bus transit opera­
poor quality services for public transit systems, so it is important to tion in terms of individual processes, individual periods, and
assess the benefits of public transit services (Wei et al. 2017). overall operation. First, we use the DNDEA which combines
As the quasi-public products, urban public transit service involves directional distance function (DDF) to obtain the comprehen­
three stakeholders: the operators that provide public transit services, sive benefits score. Second, the neutral CREE is used to obtain
the public who enjoy public transit services, and the government as the cross-efficiency scores (CESs). Finally, the Shannon entropy
the regulator (Huang et al. 2018; Zhang et al. 2016; Zhao et al. 2011). aggregation method is used to obtain the ultimate comprehen­
The operators expect to reduce operational costs as much as possible sive benefits.
while meeting public travel demands (Huang et al. 2018). The public The contributions of this study cover two aspects. First, from the
expects low fares and better services such as high accessibility, clean­ perspectives of three stakeholders (namely, the operators, the pub­
liness, safety, and comfortableness (Ingvardson and Nielsen 2019; lic, and government), we have established an evaluation indicator
Van Lierop, Badami, and El-Geneid 2018). The government expects system that includes subsidies, bus transit operational data, and
to improve bus transit services through financial and regulatory external environmental factors. Second, we propose a novel com­
measures. Accurate measurement of the bus transit benefits requires bined method to evaluate the benefits of bus transit, which com­
a comprehensive consideration of the above stakeholders. bines the cross-sectional data and panel/time series data.
Urban transportation has become a major contributor of energy The remainder of this paper is organized as follows. Section 2
consumption and greenhouse gases emissions (Fan et al. 2017). The gives a brief review of the evaluation model and indicators.
impact of external environmental factors such as congestion and Section 3 constructs the combined method and defines indicators.
greenhouse gases emissions during bus transit operation is Section 4 describes and discusses the empirical results. Finally,
increasing day by day. Meanwhile, traffic congestion is considered Section 5 summarizes the conclusions and the shortcomings of
to be one of the main factors that affect operation efficiency. the study.

CONTACT Chunqin Zhang cqzhang@zstu.edu.cn School of Civil Engineering and Architecture, Zhejiang Sci-Tech University,No. 928, Second Avenue, Xiasha
Higher Education Zone, Hangzhou 310018, China
© 2023 Informa UK Limited, trading as Taylor & Francis Group
2 M. LIU ET AL.

Literature review evaluation (CREE) based on DEA. The accuracy of CREE is affected
by the non-uniqueness of the cross-efficiency score due to the pre­
Evaluation method
sence of alternative optima in conventional DEA (Wu, Sun, and
The current evaluation of public transit services usually uses the Liang 2012a). Therefore, scholars have proposed the secondary goal
term ‘performance.’ Considering the characteristics of public transit models to address this issue. Unlike the aggressive and benevolent
services, Wang (2015) believes that ‘performance’ should be CREE proposed by Doyle and Green (1994), the neutral CREE
replaced by ‘benefit.’ Parametric and non-parametric evaluation (Wang and Chin 2010) ensures the optimal input and output weights
methods are often used for benefit evaluation. For example, of the DMU without considering the impact on other DMUs. The
Zhang, Juan, and Xiao (2015) used stochastic frontier analysis to main advantage of the neutral CREE is its ability to provide accurate
explore the technical efficiency of public transit in China. Ferreira, evaluation results without having to choose between aggressive and
Marques, and Pedro (2016) used the non-parametric partial fron­ benevolent CREEs.
tier method to compare the efficiency of the holding business model Several scholars have recognized the limitations of conventional
with the individual management model of the airport. Besides, the CREE and secondary goal models. They began to combine the
non-parametric evaluation method represented by DEA was also CREE and NDEA for evaluation. Kao and Liu (2018) and Orkcu
used for the benefit evaluation. As the DEA is suitable for the et al. (2019) combined the NDEA with aggressive CREE and neutral
situation of multi-inputs and multi-outputs and eliminates the CREE, respectively. It is shown that their proposed model has
influence of subjective factors in the evaluation process (Aldamak higher discriminative power. In this paper, we present a new
and Zolfaghari, 2017). Therefore, scholars prefer the DEA. approach that combines the DNDEA with the neutral CREE.
The DEA was first proposed by Charnes, Cooper, and Rhodes
(1978). Based on the efficiency evaluation theory of Farrell (1957),
they constructed a CCR-DEA (CCR is named by the combination of Evaluation indicator
the initials of the three authors, and BCC is the same in the following) We refer to the literature related to the DEA and get the following
model with constant returns to scale (CRS). The limitation of con­ findings: (1) From the perspective of indicator selection, existing
ventional DEA such as CCR-DEA and BCC-DEA is that the decision studies focus on public transit operators and neglect many stake­
making units (DMUs) are treated as a black box in the evaluation holders, which results in the evaluation indicators that are mostly
process (Kao 2014), which may lead to incorrect evaluation results. operational data. (2) From the types of indicators, basic production
To break the ‘black box,’ Färe and Grosskopf (2000) proposed the factors (capital, fuel consumption, and labor) are usually selected as
network data envelopment analysis (NDEA). The NDEA takes into input indicators (e.g. operating vehicles, number of employees, total
account the effects of individual processes, but it ignores the linkage length of operating roads) (Boame 2004; Fitzová, Matulová, and
of the processes (Kao 2009). Considering the linkage of the sub- Tomeš 2018; Huang et al. 2018; Karlaftis and Tsamboulas 2012;
processes within the whole process, Kao (2009) proposed the rela­ Omrani, Shafaat, and Alizadeh 2019; Yu, Chen, and Hsiao 2016,
tional NDEA model of series and parallel structure. NDEA related 2018); benefit indicators are selected as output indicators, such as
models have been widely used in banking (Fukuyama and Weber supply indicators (e.g. vehicle kilometers traveled, seat kilometers
2010), medical treatment (Pereira et al. 2021), public health (Pereira traveled) (Karlaftis and Tsamboulas 2012; Sheth, Triantis, and
et al. 2022), water and sanitation (Ferreira et al. 2021), and eco- Teodorovic 2007; Viton 1997) and indicators that reflect demand
innovation (Kiani Mavi, Saen, and Goh 2019). However, there are (e.g. passenger kilometers traveled, passenger ridership) (Fitzová,
relatively few applications in the urban public transit benefit evalua­ Matulová, and Tomeš 2018; Karlaftis and Tsamboulas 2012; Link
tion (Mahmoudi et al. 2018). Saeedi et al. (2019) proposed a modified 2019; Omrani, Shafaat, and Alizadeh 2019; Sheth, Triantis, and
model based on Slacks-based NDEA to evaluate the efficiency of Teodorovic 2007; Viton 1997; Yu, Chen, and Hsiao 2016). In
intermodal freight transport chains with different number of divi­ order to obtain realistic evaluation results, it is necessary to take
sions. Kang et al. (2020) applied a NDEA to consider the impact of into account the presence of some undesirable by-products gener­
undesirable outputs, non-storable features and transport policy fac­ ated during the operation. As a proxy for greenhouse gases, CO2 has
tors on the efficiency evaluation of Taipei’s bus transit system. a negative impact on production activities and people’s daily lives.
Stefaniec et al. (2020) used the NDEA to evaluate the sustainability Therefore, CO2 has been studied as an undesirable output in the
efficiency of transportation in China. transportation research field (Kang et al. 2020; Omrani, Shafaat,
The above studies neglect the effect of the linkage between and Alizadeh 2019; Park et al. 2018; Tian et al., 2019; Wang and He
individual periods. Tone and Tsutsui (2014) combine NDEA 2017; Wei et al. 2021). Therefore, this paper selects the operational
(Tone and Tsutsui 2009) and dynamic DEA (Tone and Tsutsui data of bus companies, government subsidies (Zhang et al. 2018;
2010) to construct a DNDEA. The advantage of DNDEA is that it Link 2019), road occupancy index (Zhang, Hu, and Lu 2022), and
can assess the benefits in terms of individual processes, individual CO2 emissions to constitute the evaluation indicator system.
periods, and overall operation. DNDEA has been used in container
shipping (Chao, Yu, and Hsieh 2018), hospitals (Khushalani and
Ozcan 2017), banking (Avkiran 2015). To the best of our knowl­ Methodology
edge, few scholars have used this model in the bus transit domain.
Combined evaluation method
For example, Yu, Chen, and Hsiao (2016) proposed a multi-activity
DNDEA to evaluate the efficiency of bus transit companies in A DNDEA for neutral CREE
Taiwan province. The results show that all bus transit companies In this study, we use bus transit systems in 33 key Chinese cities as the
provide inefficient services. study object to evaluate the benefits during 2016–2019. From the point
Although DEA related models are perfect, there are still some of view of cross-sectional data, bus transit operation can be divided
disadvantages, such as the inability to fully rank DMUs, and obtain into two sub-processes: production process (PP) and service process
the inaccuracy of efficiency values (Sun et al. 2017; Tian et al., 2019). (SP), which correspond to the operation preparation stage and the
To solve the above problems and obtain more discrimination results, operation implementation stage, respectively. From the point of view
Sexton, Silkman, and Hogan (1986) proposed a cross efficiency of panel data, bus transit operation is a long-term and continuous
TRANSPORTATION LETTERS 3

process, where some operational activities in a previous period will J


X � �
have a positive or negative impact on the next period. Therefore, the λtj;P xijt � 1 βtk;P xtik ; i ¼ 1; . . . ; n; t ¼ 1; . . . ; T (1:1)
DNDEA is used to evaluate the bus transit operation. The structure of j¼1

the DNDEA is shown in Figure 1. As the first sub-process, PP not only


transforms the initial inputs indicator into intermediate products, but J
X
also delivers the carry-over item from the previous period to the next. λtj;P zqj
t t
¼ zqk ; q ¼ 1; . . . ; e; t ¼ 1; . . . ; T (1:2)
SP transforms intermediate products into final outputs that can be j¼1
divided into desirable outputs and undesirable outputs. Intermediate
products connect PP and SP, and carry-over items connect the differ­
ent periods. X
J
ðt;tþ1Þ
X
J
ðt:tþ1Þ
λtj;P maj ¼ λtþ1
j;P maj ; a ¼ 1; . . . ; d; t ¼ 1; . . . ; T 1
The bus transit system of 33 key cities in China were chosen as j¼1 j¼1
the DMUs. Suppose that there are J DMUs operating in period
tðt ¼ 1; . . . ; T Þ. DMU can be divided into two sub-processes, i.e. PP (1:3)
and SP. DMUk is the DMU that under evaluation, where
k ¼ 1; . . . ; J. xtij ði ¼ 1; . . . ; nÞ and zqj
t
ðq ¼ 1; . . . ; eÞ denotes the X
J
ðt;tþ1Þ ðt;tþ1Þ ðt;tþ1Þ
initial input and intermediate products in period t, respectively. λtj;P maj ¼ mak Sak ; a ¼ 1; . . . ; d; t ¼ 1; . . . ; T 1
Since there are undesirable outputs, let ytrj ðr ¼ 1; . . . ; sÞ denotes s j¼1

desirable output and ytbj ðb ¼ 1; . . . ; s0 Þ denotes s0 undesirable out­ (1:4)


ðt;tþ1Þ
put in period t. maj ða
¼ 1; . . . ; dÞ denotes the carry-over items b.SP:
from period t to period t þ 1. The initial carry-over item is denoted J
ð0;1Þ
X
by maj . λtj;S zqj
t t
¼ zqk ; q ¼ 1; . . . ; e; t ¼ 1; . . . ; T (1:5)
Yu, Chen, and Hsiao (2016) and Chao, Yu, and Hsieh (2018) j¼1
proposed a multi-activity DNDEA based on the non-oriented DDF.
This model can accurately evaluate DMUs with multi-period net­ � �
X
J
work structure. We modify the model by deleting the shared input λtj;S ytrj � 1 þ βtk;S ytrk ; r ¼ 1; . . . ; s; t ¼ 1; . . . ; T (1:6)
constraints and adding undesirable output constraints. j¼1
Additionally, the enterprise is under variable returns to scale
(VRS) in the short-term operation, but should aim to achieve CRS
in the long-term operation (Cummins and Xie 2013; Yu, Chen, and X
J � �
Hsiao 2016). Considering the multi-period operation of bus transit λtj;S ytgj � 1 þ βtk;S ytgk ; g ¼ 1; . . . ; s0 ; t ¼ 1; . . . ; T (1:7)
j¼1
system, the CRS assumption is chosen in this paper. The model is as
follows: c.Initial{\rm{ }}conditions:

! X �T � X
J
ð0;1Þ ð0;1Þ
D ðxk ; zk ; mk ; yk Þ¼ max βk ¼ W t wp βtk;p þ ws βtk;s (1) λ1j;P maj ¼ mak ; a ¼ 1; . . . ; d (1:8)
t¼1 j¼1

Subject to
a.PP: λtj;P ; λtj;S � 0; j ¼ 1; . . . ; J; t ¼ 1; . . . ; T; (1:9)

Figure 1. The structure of DNDEA.


4 M. LIU ET AL.

ðt;tþ1Þ first proposed by Wang and Chin (2010), the model we use can
Sak : free; a ¼ 1; . . . ; d; t ¼ 1; . . . ; T 1; (1:10)
ðt;tþ1Þ
reduce the number of zero weights.
where βtk;P ; βtk;S ; λtj;P ; λtj;S ; Sak , and t ¼ 1; . . . ; T, j ¼ 1; . . . ; J,
i ¼ 1; . . . ; n, a ¼ 1; . . . ; d, q ¼ 1; . . . ; e, r ¼ 1; . . . ; s, g ¼ 1; . . . ; s0 , Max αδ þ βγ
k ¼ 1; . . . ; J are variables of this model. βis an inefficiency measure Subject to
in the DDF. If β ¼ 0, the DMUk is efficient; if β ¼ 1, the DMUk is Xn
inefficient. βtk;P is the measure of inefficiency of the PP for DMUk in vik xik ¼ 1;
i¼1
period t. Similarly, βtk;S is the measure of inefficiency of the SP for
X
r
DMUk in period t. λ denotes intensity variable. λtj;P and λtj;S are ðsÞ�
ulk ylk ¼ Ek ;
intensity vector of DMUj associated with the PP and SP in period t, l¼1 (6)
ðt;tþ1Þ X
r X
n
respectively. Sak denotes slack variable of the discretionary (free) ulk ylj vik xij � 0; j ¼ 1; . . . ; J; j�k;
carry-over item (Tone and Tsutsui 2010). W t is the weight on l¼1 i¼1
period t; wp and ws are the weights of PP and SP on period t,
ulk ylk δ � 0; l ¼ 1; . . . ; r;
respectively. These weights represent the relative importance of
PT t vik xik γ � 0; i ¼ 1; . . . ; n;
different periods and processes, where t¼1 W ¼ 1 and
wP þ wS ¼ 1. Since the weights are exogenously pre-assigned sca­ vik ; ulk ; δ; γ � 0; i ¼ 1; . . . ; n; l ¼ 1; . . . ; r:
lar, all weights are assumed to be equal in this study, i.e. W 1 ¼ where xij ði ¼ 1; . . . ; nÞ and ylj ðl ¼ 1; . . . ; rÞ denotes initial inputs
W 2 ¼ . . . ¼ W T ¼ 1=T and wP ¼ wS ¼ 0:5 (Yu, Chen, and Hsiao and final outputs, respectively. Undesirable output is also treated by
2016). Eq. (2). δ and γ in the objective function can be understood as the
The treatment of undesirable outputs has been a hot topic in minimum relative efficiency of the initial input and final output of the
study. By comparing the four methods for dealing with undesirable DMUk . α, β, vij , and ulj are the weighting coefficients for δ, γ, xij and
output summarized by Halkos and Petrou (2019), we determine to yij , respectively. In addition, satisfy the condition α þ β ¼ 1. In this
apply the linear monotone decreasing transformation method pro­ paper, it is assumed that α ¼ β ¼ 0:5 can reduce the number of zero
posed by Seiford and Zhu (2002).nBy ofinding a proper translation ðsÞ�
vector wh (wh � maxj¼1;...;J;t¼1;...;T ytbj ) minus the undesired out­ weights for inputs and outputs (Wang, Chin, and Jiang 2011). Ek is
put, the undesirable output ytbj can be transformed into the desir­ the comprehensive benefits score of DMUk obtained by Eq. (5) in the
DNDEA. Model (6) needs to be solved T times to get the optimal
able output ytgj . The conversion process is formulated as:
weights v�ik and u�lk of DMUk in different period. Finally, the cross-
ytgj ¼ wh ytbj (2) efficiency matrix (CEM) (see Table 1) is composed of the cross-
efficiency score (peer-evaluation score) calculated by Eq. (7) and
The production efficiency of DMUk in period t is defined as Et1 , the DNDEA efficiency score (self-evaluation score) calculated by
which is calculated from βtk;P in the Model (1), and the formula is: Eq. (5).
,
Et1 ¼ 1 βtk;p ; (3) X r X
n
Ekj ¼ u�lk ylj v�ik xij ; k; j ¼ 1; . . . ; J: (7)
Similarly, the service effectiveness in period t is Et2 ¼ 1 þ βtk;S . Since l¼1 i¼1
the effectiveness score meets the constraint range [0,1], we define
the formula as:
.� � Aggregation for CREE
Et2 ¼ 1 1 þ βtk;s ; (4) Conventional methods aggregate all CESs using arithmetic or
weighted averages (Ruiz and Sirvent 2012; Wu et al., 2020). These
Finally, benefits (self-evaluation) in period t can be calculated based methods have many defects, such as loss of correlation between the
on Et1 and Et2 , which are defined as Et , and the formula is as follows: weight and CESs, the weight obtained for calculating the cross
� �.� � efficiency may not be accepted by all DMUs, and the non-Pareto
Et ¼ Et1 Et2 ¼ 1 βtk;p 1 þ βtk;s : (5) optimal CESs be obtained (Omrani, Shafaat, and Alizadeh 2019;
Wu et al., 2020). In this paper, the Shannon entropy aggregation
Based on the principle of cross-efficiency evaluation, self-evaluation method (Wu et al. 2011; Wu, Sun, and Liang 2012b) is used to
results from the DNDEA can be used for peer-evaluation. In the obtain the ultimate CESs. This approach can overcome the limita­
peer-evaluation process, we applied the neutral CREE proposed by tions of the above methods and reduce the inconsistency between
Wang, Chin, and Jiang (2011). Different from the neutral CREE self-evaluation and peer-evaluation. The model is as follows:

Table 1. Generalized cross-efficiency matrix.


Rated DMUj
Rating DMUk 1 2 ... J
1 E11 E12 ... E1J
2 E21 E22 ... E2J
... ... ... ... ...
J EJ1 EJ2 ... EJJ
Mean E1 E2 ... EJ
Ejj denotes self-evaluation efficiency scores, Eij denotes peer-evaluation effi­
ciency scores.
TRANSPORTATION LETTERS 5

. Xn
Dij ¼ Eij Eij ; i ¼ 1; . . . ; n; j ¼ 1; . . . ; J; the assets of bus transit company, so the length of operating lines
i¼1 (Lol) is selected as the carry-over item (Yu, Chen, and Hsiao 2016).
hij ¼ Dij ln Dij ; i ¼ 1; . . . ; n; j ¼ 1; . . . ; J; (3) Output indicators
bij ¼ hij hjj ; i ¼ 1; . . . ; n; j ¼ 1; . . . ; J; Passenger ridership (Passenger), road occupancy index (Roi),
h XJ ��X �X � CO2 emissions (CO2 ) are selected as the final output indicators.
�2 i 1 n J �2 � 1 (8)
λi ¼ bij bij ; Passenger is one of the data that reflects the operation of public
j¼1 i¼1 j¼1
transit and is related to operating income. Therefore, Passenger is
X
n
taken as the desirable output. Conversely, Roi and CO2 are una­
ECross
j ¼ λi Eij ; j ¼ 1; . . . ; J: voidable by-products, so these indicators are treated as undesirable
i¼1
outputs. We made reasonable changes based on the formula pro­
The formula of Dij means the normalization of CEM. hij is defined posed by Lin and Du (2018). The formula is as follows:
as the entropy value of cross-efficiency (Eij ) in the CEM. Similarly,
Roi ¼ Vehicle � Average mileage=Lol; (10)
hjj denotes the entropy value of the DNDEA efficiency score (Ejj ).
bij denotes the entropy distance between Eij and Ejj . λi denotes where Averagemileage denotes the average operating mileage and
weight for DMUs. ECross is the ultimate CES, i.e. the ultimate com­ Averagemileage ¼ Mileage=8. The higher the value of Roi, the more
j
prehensive benefits. urban road resources are occupied by public transit vehicles. CO2 is
According to the efficiency classification standard proposed by an important reason why the carbon emissions of the transportation
Lao and Liu (2009), DMUs can be divided into three categories: industry are higher than other industries (Fan et al. 2017; Wang and
Efficient (E ¼ 1), Fairly efficient (0:6 � E < 1), and Inefficient He 2017). Therefore, CO2 is considered as an undesirable output in
(E < 0:6). E denotes the efficiency score. this study. The formula for CO2 emissions is as follows:
X
CO2 ¼ Mileage � Rated � EFd ; (11)
d
Definition of indicator where Rated denotes the proportion of the number of buses that
The availability of data determines the choice of indicators, and the meeting the dth (d ¼ 1; . . . ; 3) emission standards to the total number
accuracy of the data also affects the accuracy of the evaluation of buses. EFd denotes the CO2 emission factor of buses with dth
results. Different from previous studies which only select opera­ emission standard. According to Limits and measurement methods
tional data as indicators, the indicators in this study are chosen with for emissions from light-duty vehicles (CHINA 5) (Ministry of Ecology
the relevant data from operators, the public, and the government. and Environment of the People’s Republic of China 2018), buses can
be divided into three categories: National II and below, National III,
National IV and above. The emission factors are 1082.5g � km 1 ,
Input indicators 1129.7g � km 1 and 1072.8g � km 1 , respectively (Cai and Xie 2010).
On the input side, the number of routes (Route), the number of The selection of indicators is shown in Table 3.
standard vehicles (Vehicle), and government subsidies (Subsidy) are
selected as initial input indicators. Subsidy refers to the sum of the
reform subsidy for bus oil price and the operating subsidy for bus Empirical analysis
transit. Vehicle is calculated by the vehicle conversion formula Data description
(Ministry of Transport of the People’s Republic of China 2016-2019):
X � We evaluate the benefits of bus transit systems in 33 key Chinese
Vehicle ¼ Vf � CCf ; (9) cities. The panel data used in this study were collected from the
f China Transport Statistical Yearbook (Ministry of Transport of the
where Vf denotes the number of vehicles of f th (f ¼ 1; . . . ; 8) type. People’s Republic of China 2016-2019), China City Statistical
CCf denotes the conversion coefficient for f th type of vehicles, Yearbook (National Bureau of Statistics of the People’s Republic
which is determined by the vehicle length or the number of layers of China 2017-2020), China Urban Construction Statistical
of the vehicle (see Table 2). Yearbook (Ministry of Housing and Urban-Rural Development of
the People’s Republic of China2016–2019), the data provided by
bus transit companies, and the official websites of the government
Link indicator and carry-over item finance bureaus. The statistical data for the 33 key Chinese cities are
Operating mileage (Mileage) is not only the operating capacity shown in Tables 4 and 5. According to the statistical data of Roi and
produced in the production preparation stage but also the basic CO2 , the wh in Eq. (2) is selected as 35,000 and 14,000, respectively.
premise for providing services in the service process, so Mileage is After the Spearman correlation test (see Table 6), there is
used as an intermediate product (link indicator) which connects the a significant correlation between initial inputs, intermediate outputs,
two sub-processes (Yu, Chen, and Hsiao 2016). The road network is and final outputs, which means these indicators are isotonicity related
(Bang et al. 2012). Thus, these indicators can be used in the evaluation
of DEA.
Table 2. Bus classification standards and conversion coefficients.
Categories Vehicle length range L (meter) Conversion coefficient
Result
1 L≤5 0.5
2 5<L≤7 0.7 Results of dynamic network data envelopment analysis
3 7<L≤10 1.0
4 10<L≤13 1.3
The 33 DMUs were numbered sequentially using Arabic numerals
5 13<L≤16 1.7 (see Table 5). LINGO 17.0 software is used to calculate βtk;P and βtk;S
6 16<L≤18 2.0 in different periods. Then bring βtk;P and βtk;S in the same period into
7 L>18 2.5 Eqs. (3)-(5), and can get the production efficiency (E1 ), service
8 Double vehicle 1.9
effectiveness (E2 ) and comprehensive benefits (E).
6 M. LIU ET AL.

Table 3. Description of input and output indicators.


Type Indicators Description
Inputs Number of routes (Route) The number of fixed operation lines set for operation vehicles, including fixed lines such as branch lines and
dedicated lines.
Number of standard vehicles Different types of operating vehicles are converted into a unified standard number of operating vehicles according
(Vehicle) to the conversion coefficients.
Government subsidies Financial subsidies provided by the government to the operators. This study includes two parts: fuel price subsidies
(Subsidy) and operating subsidies.
Intermediate Operating mileages (Mileage) The total mileage of operating vehicles during the operation period.
product
Carry-over items Length of operating lines (Lol) Total length of all operating lines owned by the operating enterprise.
Outputs Passenger ridership The total number of passengers transported by operating vehicles during the operation period. Desirable output.
(Passenger)
Road occupancy index (Roi) It reflects the impact of bus operations on road congestion and the occupation of social resources. Undesirable
output.
CO2 emissions (CO2 ) Volume of CO2 in the exhaust emissions from operating vehicles during the operation period. Undesirable output.

Table 4. Statistics of the indicators (2016–2019).


Carry-over Intermediate
Initial inputs items product Final outputs
Subsidy
Statistical Route Vehicle (million Lol Mileage (104 Passenger (104 Roi as desirable CO2 CO2 as desirable
Characteristic (lines) (vehicles) USD) (km) km) person-times) Roi (104) output (104) (100 tons) output (100 tons)
Max 1575 36572 4224.18 27632 133630 369019 31367.03 34243.11 13474.65 13864.34
Min 94 1836 0.29 1399 6581 19847 756.89 3632.97 135.66 525.35
Mean 497.67 9613.42 261.16 9289.86 42180.90 103287.21 6034.52 28965.48 3268.22 10731.78
SD 368.64 6774.67 603.31 6997.10 31671.13 71409.69 5157.35 5157.35 2606.21 2606.21

Table 5. Socioeconomic statistics of 33 DMUs.


Number 1 2 3 4 5 6 7 8 9 10 11
DMU Beijing Tianjin Shijiazhuang Taiyuan Hohhot Shenyang Changchun Harbin Shanghai Nanjing Hangzhou
Population Density 1140 4218 5728 3742 8084 2932 1695 10495 3821 1537 3799
(person/square kilometer)
Per capital GDP (million) 203.28 164.09 86.11 123.65 141.92 108.69 125.68 99.59 197.30 216.28 208.46
Number 12 13 14 15 16 17 18 19 20 21 22
DMU Hefei Fuzhou Nanchang Jinan Zhengzhou Wuhan Changsha Guangzhou Haikou Chongqing Chengdu
Population Density 3693 3197 7511 2398 11236 5901 3423 6102 3195 2002 6169
(person/square kilometer)
Per capital GDP (million) 154.39 153.39 139.53 152.50 144.85 190.13 203.91 222.77 94.38 96.40 139.62
Number 23 24 25 26 27 28 29 30 31 32 33
DMU Guiyang Kunming Xi’an Lanzhou Xining Yinchuan Urumqi Qingdao Ningbo Shenzhen Xiamen
Population Density 2319 2232 6989 7404 3514 1346 2500 1810 2840 6371 9166
(person/square kilometer)
Per capital GDP (million) 113.32 119.55 122.69 106.09 89.06 118.17 122.96 183.81 198.03 249.34 172.38

Table 6. The Spearman correlation test of indicators.


Indicators Route Vehicle Subsidy Lol Mileage Passenger Roi CO2
Route 1
Vehicle 0.899** 1
Subsidy 0.766** 0.803** 1
Lol 0.981** 0.890** 0.741** 1
Mileage 0.920** 0.953** 0.806** 0.915** 1
Passenger 0.836** 0.901** 0.794** 0.805** 0.899** 1
Roi 0.677** 0.886** 0.747** 0.644** 0.857** 0.865** 1
CO2 0.814** 0.841** 0.746** 0.825** 0.896** 0.874** 0.762** 1
** significant correlated at 0.01 level (two-sided).

Production efficiency. Figure 2 shows that the production effi­ and carry-over of operating lines that cannot be converted into high
ciency scores of most DMUs fluctuate between 0.6 and 1 during operational mileage. Compared with Chengdu, Harbin has
2016–2019, which indicates that the production process of bus achieved a higher operating mileage than Chengdu with three
transit in 33 key cities is generally fairly efficient. E1 of Harbin inputs far lower than this city. In addition, there are some DMUs
(DMU8, the same below), Hangzhou (11), Haikou (20), Xi’an (25) that are efficient in certain years, but the efficiency scores decrease
are equal to 1 in four years, which indicates that the input of these gradually, such as Tianjin (2) and Hohhot (5).
cities has reached a balance with intermediate products and carry- Overall, it is consistent with common sense that cities with high
over items. It is worth noting that Chengdu (22) has the lowest per capita GDP have higher production efficiency than other cities.
efficiency score, which may be caused by excessive resource input The high per capita GDP of the city means a more developed
TRANSPORTATION LETTERS 7

2016 2017 2018 2019


1.0

0.8

Production efficiency
0.6

0.4

0.2

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33
Cities

Figure 2. Production efficiency (The number stands for DMU).

economy and the higher investment in the bus transit industry, so travel by bus. On the other hand, smaller cities have fewer alter­
the production is more efficient. native travel modes than other cities, so the bus sharing rate is
relatively high. Additionally, the bus operation of small cities also
Service effectiveness. From Figure 3, E2 has a minimum of 0.259 occupies less urban road resources, and the CO2 emission of vehi­
and a maximum of 1 among the 33 cities, indicating that there are cles remains at a low level. The aforementioned advantages that
great differences in bus transit services between different cities. E2 is above-mentioned enable bus operations in small cities to achieve
lower than E1 in the same city from 2016 to 2019, and the service higher service effectiveness.
effectiveness of the city varies in different years. It should be noted
that the four cities that are efficient in the production process have Comprehensive benefit. As shown in Figure 4, the E of 33 cities is
low service effectiveness. Hangzhou (11) is the city with the largest inefficient as a whole and large fluctuations due to the effect of E1
difference in production efficiency and service effectiveness among and E2 . The cities with the largest gap between the two processes are
the four cities mentioned above. The reason for the large difference those with large-scale bus transit systems, such as Beijing (1),
is that higher inputs achieve higher operating mileage but do not Tianjin (2), Guangzhou (19) and Shenzhen (32). The large input
lead to higher passenger ridership. The city with the highest E2 is of subsidies and operational resources in these cities results in high
Xining (27), which has been efficient for four years, followed by passenger ridership, but passenger growth has not yet reached the
Yinchuan (28) and Lanzhou (26). The three cities also maintained optimal level, resulting in low comprehensive benefits. In general,
high efficiency scores in the production process. improving resource utilization and reducing undesirable output is
According to the City Size Classification of China (State Council one of the effective measures to improve the comprehensive benefit.
of the People’s Republic of China 2014), it is found that cities with We use SPSS 25 to perform regression analysis on E1 , E2 , and E.
high service effectiveness scores all belong to small cities with low Before regression analysis, the Pearson correlation test was used to
per capita GDP, such as Taiyuan (4), Hohhot (5), Lanzhou (26), analyze the correlation between them. The result shows E1 and E2
Xining (27) and Yinchuan (28). Cities with high service effective­ have positive correlation with E, respectively. In addition, the
ness are small cities with low per capita GDP. On the one hand, low Variance Inflation Factor between E1 and E2 is less than 5, which
per capita GDP means low income for residents, so people prefer to indicates that there is no multicollinearity between them. The sub-

2016 2017 2018 2019


1.0

0.8
Service effectiveness

0.6

0.4

0.2

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33
Cities

Figure 3. Service effectiveness (The number stands for DMU).


8 M. LIU ET AL.

2016 2017 2018 2019


1.0

Comprehensive benefits
0.8

0.6

0.4

0.2

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33
Cities

Figure 4. Comprehensive benefits (The number stands for DMU).

process efficiencies of all cities from 2016 to 2019 are plotted in bus transit in China are low. From the spatial distribution point of
a Figure 5 which comprehensive benefits as the horizontal coordi­ view, the distribution of benefits in the key cities is not balanced. Most
nate. As the R2 of E1 and E2 are 0.044 and 0.753 respectively, it can of the fairly efficient cities in 2016 are located in the western region,
be concluded that E is mainly affected by E2 . Therefore, improving while in 2019 they are spread across the western and northeastern
service effectiveness should be considered as the main measure to regions. Key cities in the central and eastern regions have been ineffi­
improve the comprehensive benefits of bus transit services. cient for four years. As can be seen from the changes in the compre­
hensive benefits of bus transit companies over the past four years, it can
be seen that the improvement of the benefits of bus transit is closely
Results of ultimate cross-efficiency related to the geographical location, the scale of the bus transit system,
In terms of cross-efficiency results, Lanzhou (26) ranked first in the and the distribution of population. Compared with the central and
four years with high benefit scores of 1.001, 0.993, 0.954, and 0.972. eastern regions, it is easier to improve the benefits of bus transit in the
Xining (27) ranked second with 0.785, 0.814, 0.869, and 0.936. western and northeastern regions. The possible explanation is that
Large cities such as Beijing (1), Tianjin (2), Shanghai (9), Nanjing lower per capita GDP in the west and northeast means lower economic
(10), and Shenzhen (32) ranked the lowest among the 33 cities. It is development and less choice of travel modes, so residents prefer to
clear that small cities rank higher than large cities, which is con­ travel by bus. The central and eastern regions have well-developed
sistent with the conclusion obtained by the DNDEA. Most of the 33 economies and infrastructure, so residents can choose multiple modes
cities were in the range of 0.3 to 0.6. of travel, resulting in low bus transit sharing. In addition, bus opera­
Figure 6 shows the ultimate comprehensive benefits for 33 cities in tions in the eastern and central regions have a large negative impact on
2016 and 2019. It can be seen that only one city (Lanzhou, 26) is the external environment, such as road congestion and CO2 emissions,
efficient, a few cities are fairly efficient, and most cities are inefficient. resulting in lower benefits.
The number of efficient and fairly efficient cities has increased from 5
to 8 during 2016–2019, indicating that the comprehensive benefits of
Comparison of methods
The comprehensive benefits and rankings calculated based on the
1.0 three methods of CCR, DNDEA, and combined method are
shown in Table A (see Appendix 1). As can be seen from
Figure 7, CCR achieves higher evaluation results than DNDEA
and the combined method, while the results of DNDEA and the
0.8 combined method are close. Without considering the impact of
DMUs’ sub-processes and carry-over items, the benefit scores of
Scores

CCR are higher and close to 1. For efficient DMUs, the CCR
0.6 cannot be further classified, but the DNDEA and the combined
method solve this problem. Thus, CCR has the weakest resolu­
tion and the largest error. The difference between DNDEA and
the combined method is mainly caused by the peer-evaluation.
0.4 E1, R2=0.044 For example, the CCR and DNDEA rankings of Beijing (1) are
both 14, while the combined method rankings are 30 in 2016.
E2, R2=0.753 Such a large difference is caused by the lower efficiency of peer-
0.2 evaluation and vice versa. With the exception of a few cities, the
0.2 0.4 0.6 0.8 1.0 ranking results of DNDEA and the combined method are gen­
Comprehensive benefits (E) erally close.
The Spearman correlation test is applied to the results obtained
Figure 5. Regression Analysis of Sub-process Efficiency and Comprehensive by CCR, DNDEA and the combined method. The correlation
Efficiency. coefficient results are shown in Table 7. The results show that
TRANSPORTATION LETTERS 9

(a)2016 (b)2019

0
± 375 750 Miles
E<0.600
0.600≤E<1.000
E≥1.000

Figure 6. Comprehensive benefits of 33 cities in 2016(a) and 2019(b) (The number stands for DMUs).

CCR DNDEA Combined method


1.0
Comprehensive benefits

0.8

0.6

0.4

0.2

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33
Cities

Figure 7. Comparison of the average benefit scores obtained by three methods (The number stands for DMU).

Table 7. The Spearman correlation test among different methods.


Passenger is removed, the benefit score is greatly reduced. Such
Model CCR DNDEA Our method
a large reduction in the benefit scores could be related to Passenger
CCR 1 being the only desirable output.
DNDEA 0.923** 1
Our method 0.886** 0.868** 1
Note: ** significant correlated at 0.01 level (two-sided). Discussion
Using the combined evaluation method, this paper conducts an
empirical study on the benefits of bus transit service in 33 key cities
there is a significant positive correlation between the results of the of China. Based on the self-evaluation results obtained by the
three methods. It shows the accuracy of the method used in this DNDEA, different sub-processes can be evaluated. We find that
study. production processes are more efficient than service processes. The
main reason may be that the excessive resource input increases
operating mileage and the length of the operating line, but the
Sensitivity analysis
desired passenger ridership is not achieved. From the relationship
Sensitivity analysis is an indispensable step to test the reliability of between comprehensive benefits and the efficiency of the two sub-
evaluation results (Caulfield, Bailey, and Mullarkey 2013). The processes, it can be seen that comprehensive benefits are more
reliability of our method is evaluated by removing one or more affected by service effectiveness. Therefore, the measures to
indicators while keeping the others unchanged. Since the two improve the benefits should mainly focus on the service process,
undesired outputs (Roi and CO2 ) have a negative impact on the i.e. increase in desirable output and decrease in undesirable output.
benefits, they are deleted at the same time. By comparing the benefit scores of different cities, it can be found
Figure 8 shows a comparison of the average benefit scores after that small cities have higher benefits than large cities. On the one
removing the indicators. It can be seen that the benefit score hand, large cities need more resources to sustain bus operations, but
changes little before and after removing the indicators except too many resources input cannot achieve stable growth of passenger
Passenger, indicating that the combined method is reliable. When ridership. On the other hand, travel modes such as metro and tram
10 M. LIU ET AL.

Route Vehicle Subsidy Passenger Roi, CO2


1.0

Combined method benefit scores


0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33
Cities

Figure 8. Comparison of average benefit scores after removed indicator (The number stands for DMU).

in large cities have reduced the sharing rate of bus transit. governments in central and eastern of China need to improve
Additionally, the benefits in the west and northeast region are the bus ride experience to increase the attractiveness of buses.
higher than the central and eastern region, which is similar to the There are several flaws need to discussed in this paper:
conclusions of Li et al. (2020). To ensure balanced bus transit W t , wP , and wS in Model (1) can be reasonably determined in
development in China, the next phase aims to take measures to the future by using methods such as expert scoring method, entro­
improve the benefits of large cities located in the central and eastern pic method, etc.
regions. Neutral CREE does not evaluate the efficiency of the sub-
None-the-less some indicators for cities are changing quite a lot processes. It is necessary to construct the CREE with network
over these few years, which indicate faster changes of services and/ structure and dynamics factors to evaluate the ultimate compre­
or population. Prior to 2016, some indicators had changed drama­ hensive efficiency in the future.
tically. After 2020, the epidemic has had a great impact on public This study ignores competition or cooperation between bus
transport transit operations. Therefore, we choose 2016–2019 as the transit companies in the same city and different cities. Corporate
study period. Since the study period is only four years and the relationships are known to have some impact on benefits. Similarly,
benefits of bus transit do not change much over the period 2016– the effect of other public transit modes such as the metro/rail and
2019. However, since data only cover 4 years, the empirical con­ private cars on bus transit operations has been overlooked. In the
tribution is smaller than if more years were available. future, the DNDEA can be combined with the game-related CREE
to comprehensively evaluate the benefits of bus transit.

Conclusion
Acknowledgments
City construction is inextricably linked to the development of the
The authors thank the managers and respondents at the sector of PT for providing
transportation industry. However, bus transit operations have been data and information that were essential for this work. This paper is largely
greatly affected by increased investment and the impact of external supported by National Natural Science Foundation of China (Grant
environmental factors. This study divides the operation of bus No. 71901196, 72274178), Major Humanities and Social Sciences Project of
transit process into production and service processes. It selects Universities in Zhejiang Province, China (Grant No. 2018QN028), and MOE
indicators from the perspective of operators, governments, and (Ministry of Education in China) Project of Humanities and Social Sciences
(Grant No. 19YJCZH238).
the public. The combined method consisting of DNDEA, CREE,
and Shannon entropy aggregation method is used to evaluate the
bus benefits in 33 key Chinese cities from 2016 to 2019. Finally, we Disclosure statement
compare the results of three methods: CCR, DNDEA, and our
No potential conflict of interest was reported by the authors.
proposed method. Empirical results show that most of China’s 33
key cities have poor benefits, but the comprehensive benefits are
slowly increasing. Production efficiency is higher than service effec­ Funding
tiveness. The benefits of bus transit are higher in the western and
The work was supported by the National Natural Science Foundation of China
northeastern regions than in the central and eastern regions. Small [71901196]; National Natural Science Foundation of China [72274178]; MOE
cities have higher comprehensive benefits than large cities. (Ministry of Education in China) Project of Humanities and Social Sciences
The study results show that the benefits can be improved by [19YJCZH238]; Major Humanities and Social Sciences Project of Universities in
increasing passenger ridership while maintaining the growth of Zhejiang Province, China [2018QN028]
operating mileage, minimizing road occupancy and CO2 emis­
sions. These measures require cooperation between companies References
and governments. For bus enterprises, the operator responds to
the government carbon reduction policies such as promoting Aldamak, A., and S. Zolfaghari. 2017. “Review of Efficiency Ranking Methods in
new energy buses and replacing vehicles that do not meet emis­ Data Envelopment Analysis.” Measurement 106: 161–172. doi:10.1016/j.mea
surement.2017.04.028.
sion standards. For the government, the government provides Avkiran, N. K. 2015. “An Illustration of dynamic network DEA in commercial
subsidies for bus enterprises to purchase new energy buses and banking including robustness tests.” Omega 55: 141–150. doi:10.1016/j.
strengthen the construction of transportation. Additionally, omega.2014.07.002.
TRANSPORTATION LETTERS 11

Bang, H. S., H. W. Kang, J. Martin, and S. H. Woo. 2012. “The impact of Analysis (DEA).” Socio-Economic Planning Sciences 60 (DEC): 15–23.
operational and strategic management on liner shipping efficiency: A doi:10.1016/j.seps.2017.01.009.
two-stage DEA approach.” Maritime Policy & Management 39 (7): Kiani Mavi, R., R. F. Saen, and M. Goh. 2019. “Joint Analysis of Eco-Efficiency
653–672. doi:10.1080/03088839.2012.740165. and Eco-Innovation with Common Weights in Two-Stage Network DEA:
Boame, A. K. 2004. “The technical efficiency of Canadian urban transit systems.” A Big Data Approach.” Technological Forecasting and Social Change
Transportation Research Part E: Logistics and Transportation Review 40 (5): 144 (JUL): 553–562. doi:10.1016/j.techfore.2018.01.035.
401–416. doi:10.1016/j.tre.2003.09.002. Lao, Y., and L. Liu. 2009. “Performance Evaluation of Bus Lines with Data
Cai, H., and S. D. Xie. 2010. “determination of emission factors from motor Envelopment Analysis and Geographic Information Systems.” Computers,
vehicles under different emission standards in China.” Acta Scientiarum Environment and Urban Systems 33 (4): 247–255. doi:10.1016/j.compen
Naturalium Universitatis Pekinensis 46 (3): 319–326. in chinese. doi:10. vurbsys.2009.01.005.
13209/j.0479-8023.2010.046. Li, Q., P. R. Bai, Y. Chen, and X. Wei. 2020. “Efficiency Evaluation of Bus
Caulfield, B., D. Bailey, and S. Mullarkey. 2013. “Using data envelopment Transport Operations Given Exogenous Environmental Factors.” Journal of
analysis as a public transport project appraisal tool.” Transport Policy 29: Advanced Transportation 2020: 1–13. doi:10.1155/2020/8899782.
74–85. doi:10.1016/j.tranpol.2013.04.006. Lin, B. Q., and Z. L. Du. 2018. “A Study of China’s Urban Vehicle Energy
Chao, S. L., M. M. Yu, and W. F. Hsieh. 2018. “Evaluating the Efficiency of major Consumption and Public Transport Efficiency.” Economic Research Journal
container shipping companies: a framework of dynamic network DEA with 53 (06): 142–156. in chinese. https://kns.cnki.net/kcms/detail/detail.aspx?
Shared inputs.” Transportation Research Part A: Policy and Practice 117: FileName=JJYJ201806011&DbName=CJFQ2018.
44–57. doi:10.1016/j.tra.2018.08.002. Link, H. 2019. “The Impact of Including Service Quality into Efficiency Analysis:
Charnes, A., W. W. Cooper, and E. Rhodes. 1978. “Measuring the efficiency of The Case of Franchising Regional Rail Passenger Serves in Germany.”
decision making units.” European Journal of Operational Research 2 (6): Transportation Research Part A: Policy and Practice 119: 284–300. doi:10.
429–444. doi:10.1016/0377-2217(78)90138-8. 1016/j.tra.2018.11.019.
Cummins, J. D., and X. Y. Xie. 2013. “Efficiency, productivity, and scale econo­ Mahmoudi, R., A. Emrouznejad, S. N. Shetab-Boushehri, and S. R. Hejazi. 2018.
mies in the U.S. property-liability insurance industry.” Journal of Productivity “The Origins, Development and Future Directions of Data Envelopment
Analysis 39 (2): 141–164. doi:10.1007/s11123-012-0302-2. Analysis Approach in Transportation Systems.” Socio-Economic Planning
Doyle, J., and R. Green. 1994. “Efficiency and cross-efficiency in DEA: deriva­ Sciences 69: 100672. doi:10.1016/j.seps.2018.11.009.
tions, meanings and uses.” The Journal of the Operational Research Society Ministry of Ecology and Environment of the People’s Republic of China, 2018.
45 (5): 567–578. doi:10.1057/jors.1994.84. “Limits and Measurement Methods for Emissions from Light-Duty Vehicles
Fan, J. L., J. X. Wang, F. Y. Li, H. Yu, and X. Zhang. 2017. “Energy demand and (CHINA 5)”. https://www.mee.gov.cn/ywgz/fgbz/bz/bzwb/dqhjbh/dqy
greenhouse gas emissions of urban passenger transport in the internet Era: dywrwpfbz/201309/t20130917_260352.shtml.
A Case Study of Beijing.” Journal of Cleaner Production 165: 177–189. doi:10. Ministry of Housing and Urban-Rural Development of the People’s Republic of
1016/j.jclepro.2017.07.106. China, 2016-2019. China Urban Construction Statistical Yearbook. China
Färe, R., and S. Grosskopf. 2000. “Network DEA.” Socio-Economic Planning Planning Press.
Sciences 34 (1): 35–49. doi:10.1016/S0038-0121(99)00012-9. Ministry of Transport of the People’s Republic of China. 2016-2019. China
Farrell, M. J. 1957. “The measurement of productive efficiency.” Journal of the
Transport Statistical Yearbook. Beijing China: China Communications Press.
Royal Statistical Society 120 (3): 253–290. doi:10.2307/2343100.
Ministry of Transport of the People’s Republic of China, 2020. “Passenger
Ferreira, D. C., I. Graziele, R. C. Marques, and J. Gonçalves. 2021. “Investment in
Traffic in China’s Central Cities in 2020”. https://xxgk.mot.gov.cn/2020/
drinking water and sanitation infrastructure and its impact on waterborne
jigou/zhghs/202101/t20210121_3517380.html.
diseases dissemination: The Brazilian Case.” The Science of the Total
National Bureau of Statistics of the People’s Republic of China. 2017-2020.
Environment 779: 146279. doi:10.1016/j.scitotenv.2021.146279.
China City Statistical Yearbook. Beijing China: China Statistics Press.
Ferreira, D. C., R. C. Marques, and M. I. Pedro. 2016. “Comparing efficiency of
Omrani, H., K. Shafaat, and A. Alizadeh. 2019. “Integrated Data Envelopment
holding business model and individual management model of airports.”
Analysis and Cooperative Game for Evaluating Energy Efficiency of
Journal of Air Transport Management 57: 168–183. doi:10.1016/j.jairtra
Transportation Sector: A Case of Iran.” Annals of Operations Research
man.2016.07.020.
274 (1–2): 471–499. doi:10.1007/s10479-018-2803-5.
Fitzová, H., M. Matulová, and Z. Tomeš. 2018. “Determinants of Urban Public
Transport Efficiency: Case Study of the Czech Republic.” European Transport Orkcu, H. H., V. S. Ozsoy, M. Orkcu, and H. Bal. 2019. “A Neutral Cross
Research Review 10 (2): 42. doi:10.1186/s12544-018-0311-y. Efficiency Approach for Basic Two Stage Production Systems.” Expert
Fukuyama, H., and W. L. Weber. 2010. “A Slacks-Based Inefficiency Measure for Systems with Applications 125: 333–344. doi:10.1016/j.eswa.2019.01.067.
a Two-Stage System with Bad Outputs.” Omega 38 (5): 398–409. doi:10.1016/ Park, Y. S., S. H. Lim, G. Egilmez, and J. Szmerekovsky. 2018. “Environmental
j.omega.2009.10.006. Efficiency Assessment of US Transport Sector: A Slack-Based Data
Halkos, G., and K. N. Petrou. 2019. “Treating Undesirable Outputs in DEA: Envelopment Analysis Approach.” Transportation Research Part D:
A Critical Review.” Economic Analysis and Policy 62: 97–104. doi:10.1016/j. Transport and Environment 61: 152–164. doi:10.1016/j.trd.2016.09.009.
eap.2019.01.005. Pereira, M. A., D. C. Dinis, D. C. Ferreira, J. R. Figueira, and R. C. Marques.
Huang, W. C., B. Shuai, Y. Sun, Y. Wang, and E. Antwi. 2018. “Using 2022. “A Network Data Envelopment Analysis to Estimate nations’ Efficiency
Entropy-TOPSIS Method to Evaluate Urban Rail Transit System Operation in the Fight Against SARS-CoV-2.” Expert Systems with Applications 210:
Performance: The China Case.” Transportation Research Part A: Policy and 118362. doi:10.1016/j.eswa.2022.118362.
Practice 111: 292–303. doi:10.1016/j.tra.2018.03.025. Pereira, M. A., D. C. Ferreira, J. R. Figueira, and R. C. Marques. 2021.
Ingvardson, J. B., and O. A. Nielsen. 2019. “The Relationship Between Norms, “Measuring the Efficiency of the Portuguese Public Hospitals: A Value
Satisfaction and Public Transport Use: A Comparison Across Six European Modelled Network Data Envelopment Analysis with Simulation.” Expert
Cities Using Structural Equation Modelling.” Transportation Research Part Systems with Applications 181 (1): 115169. doi:10.1016/j.eswa.2021.115169.
A: Policy and Practice 126 (AUG): 37–57. doi:10.1016/j.tra.2019.05.016. Ruiz, J. L., and I. Sirvent. 2012. “On the DEA Total Weight Flexibility and the
Kang, C. C., C. M. Feng, B. R. Liao, and H. A. Khan. 2020. “Accounting for Air Aggregation in Cross-Efficiency Evaluations.” European Journal of
Pollution Emissions and Transport Policy in the Measurement of the Operational Research 223 (3): 732–738. doi:10.1016/j.ejor.2012.06.011.
Efficiency and Effectiveness of Bus Transits.” Transportation Letters 12 (5): Saeedi, H., B. Behdani, B. Wiegmans, and R. Zuidwijk. 2019. “Assessing the
349–361. doi:10.1080/19427867.2019.1592369. Technical Efficiency of Intermodal Freight Transport Chains Using
Kao, C. 2009. “Efficiency Decomposition in Network Data Envelopment a Modified Network DEA Approach.” Transportation Research Part E:
Analysis: A Relational Model.” European Journal of Operational Research Logistics and Transportation Review 126 (JUN): 66–86. doi:10.1016/j.tre.
192 (3): 949–962. doi:10.1016/j.ejor.2007.10.008. 2019.04.003.
Kao, C. 2014. “Network Data Envelopment Analysis: A Review.” European Seiford, L. M., and J. Zhu. 2002. “Modeling Undesirable Factors in Efficiency
Journal of Operational Research 239 (1): 1–16. doi:10.1016/j.ejor.2014.02.039. Evaluation.” European Journal of Operational Research 142 (1): 16–20. doi:10.
Kao, C., and S. T. Liu. 2018. “Cross Efficiency Measurement and Decomposition 1016/S0377-2217(01)00293-4.
in Two Basic Network Systems.” Omega 83 (MAR): 70–79. doi:10.1016/j. Sexton, T. R., R. H. Silkman, and A. J. Hogan. 1986. “Data Envelopment
omega.2018.02.004. Analysis: Critique and Extensions.” New Directions for Program Evaluation
Karlaftis, M. G., and D. Tsamboulas. 2012. “Efficiency Measurement in Public 1986 (32): 73–105. doi:10.1002/ev.1441.
Transport: Are Findings Specification Sensitive?” Transportation Research Sheth, C., K. Triantis, and D. Teodorovic. 2007. “Performance Evaluation of Bus
Part A: Policy and Practice 46 (2): 392–402. doi:10.1016/j.tra.2011.10.005. Routes: A Provider and Passenger Perspective.” Transportation Research Part
Khushalani, J., and Y. A. Ozcan. 2017. “Are Hospitals Producing Quality Care E: Logistics and Transportation Review 43 (4): 453–478. doi:10.1016/j.tre.
Efficiently? An Analysis Using Dynamic Network Data Envelopment 2005.09.010.
12 M. LIU ET AL.

State Council of the People’s Republic of China, 2014. “Notice of the State Equity.” Journal of Transport Geography 65 (DEC): 70–79. doi:10.1016/j.
Council of China on Adjusting the Criteria for City Size Classification”. jtrangeo.2017.10.010.
https://www.gov.cn/zhengce/content/2014-11/20/content_9225.htm. Wei, F. Q., X. Q. Zhang, J. F. Chu, F. Yang, and Z. Yuan. 2021. “Energy and
Stefaniec, A., K. Hosseini, J. H. Xie, and Y. J. Li. 2020. “Sustainability Assessment Environmental Efficiency of China’s Transportation Sectors Considering
of Inland Transportation in China: A Triple Bottom Line-Based Network CO2 Emission Uncertainty.” Transportation Research Part D: Transport
DEA Approach.” Transportation Research Part D: Transport and and Environment 97 (11): 102955. doi:10.1016/j.trd.2021.102955.
Environment 80: 102258. doi:10.1016/j.trd.2020.102258. Wu, J., J. S. Sun, and L. Liang. 2012a. “Cross Efficiency Evaluation Method Based
Sun, J., Y. Yuan, R. Yang, X. Ji, and J. Wu. 2017. “Performance Evaluation of on Weight-Balanced Data Envelopment Analysis Model.” Computers &
Chinese Port Enterprises Under Significant Environmental Concerns: An Industrial Engineering 63 (2): 513–519. doi:10.1016/j.cie.2012.04.017.
Extended DEA-Based Analysis.” Transport Policy 60: 75–86. doi:10.1016/j. Wu, J., J. S. Sun, and L. Liang. 2012b. “DEA Cross-Efficiency Aggregation Method
tranpol.2017.09.001. Based Upon Shannon Entropy.” International Journal of Production Research
Tian, N., S. S. Tang, A. Che, and P. Wu. 2019. “Measuring Regional Transport 50 (23): 6726–6736. doi:10.1080/00207543.2011.618150.
Sustainability Using Super-Efficiency SBM-DEA with Weighting Wu, J., J. S. Sun, L. Liang, and Y. C. Zha. 2011. “Determination of Weights for
Preference.” Journal of Cleaner Production 242: 118474. doi:10.1016/j.jcle Ultimate Cross Efficiency Using Shannon Entropy.” Expert Systems with
pro.2019.118474. Applications 38 (5): 5162–5165. doi:10.1016/j.eswa.2010.10.046.
Tone, K., and M. Tsutsui. 2009. “Network DEA: A Slacks-Based Measure Yu, M. M., L. H. Chen, and B. Hsiao. 2016. “Dynamic Performance Assessment
Approach.” European Journal of Operational Research 197 (1): 243–252. of Bus Transit with the Multi-Activity Network Structure.” Omega 60 (APR):
doi:10.1016/j.ejor.2008.05.027. 15–25. doi:10.1016/j.omega.2015.06.003.
Tone, K., and M. Tsutsui. 2010. “Dynamic DEA: A Slacks-Based Measure Yu, M. M., L. H. Chen, and B. Hsiao. 2018. “A Performance-Based Subsidy
Approach.” Omega 38 (3–4): 145–156. doi:10.1016/j.omega.2009.07.003. Allocation of Ferry Transportation: A Data Envelopment Approach.”
Tone, K., and M. Tsutsui. 2014. “Dynamic DEA with Network Structure: A Transport Policy 68: 13–19. doi:10.1016/j.tranpol.2018.04.004.
Slacks-Based Measure Approach.” Omega 42 (1): 124–131. doi:10.1016/j. Zhang, C. Q., Y. T. Hu, and W. T. Lu. 2022. “Evaluating the Comprehensive
omega.2013.04.002. Benefit of Public Transport Service–The Perspective of Three Stakeholders.”
Van Lierop, D., M. G. Badami, and A. M. El-Geneid. 2018. “What Influences Promet-Traffic & Transportation 34 (2): 179–193. doi:10.7307/ptt.v34i2.3855.
Satisfaction and Loyalty in Public Transport? A Review of the Literature.” Zhang, C. Q., Z. C. Juan, Q. Y. Luo, and G. N. Xiao. 2016. “Performance Evaluation
Transport Reviews 38 (1): 52–72. doi:10.1080/01441647.2017.1298683. of Public Transit Systems Using a Combined Evaluation Method.” Transport
Viton, P. A. 1997. “Technical Efficiency in Multi-Mode Bus Transit: Policy 45 (JAN): 156–167. doi:10.1016/j.tranpol.2015.10.001.
A Production Frontier Analysis.” Transportation Research Part B: Zhang, C. Q., Z. C. Juan, and G. N. Xiao. 2015. “Do Contractual Practices Affect
Methodological 31 (1): 23–39. doi:10.1016/S0191-2615(96)00019-7. Technical Efficiency? Evidence from Public Transport Operators in China.”
Wang, H. M. 2015. Research on the Benefit of Public Bus Service Based on Transportation Research Part E: Logistics and Transportation Review
Cooperative Governance: A Case Study of Shanghai Public Bus Service. 80 (AUG): 39–55. doi:10.1016/j.tre.2015.05.001.
Shanghai China: Tongji University Press. Zhang, C. Q., Y. Liu, W. Lu, and G. Xiao. 2019. “Evaluating Passenger
Wang, Y. M., and K. S. Chin. 2010. “A Neutral DEA Model for Cross-Efficiency Satisfaction Index Based on PLS-SEM Model: Evidence from Chinese
Evaluation and Its Extension.” Expert Systems with Applications 37 (5): Public Transport Service.” Transportation Research Part A: Policy and
3666–3675. doi:10.1016/j.eswa.2009.10.024. Practice 120 (FEB): 149–164. doi:10.1016/j.tra.2018.12.013.
Wang, Y. M., K. S. Chin, and P. Jiang. 2011. “Weight Determination in the Zhang, C. Q., G. N. Xiao, Y. Liu, and F. Yu. 2018. “The Relationship Between
Cross-Efficiency Evaluation.” Computers & Industrial Engineering 61 (3): Organizational Forms and the Comprehensive Effectiveness for Public
497–502. doi:10.1016/j.cie.2011.04.004. Transport Services in China?” Transportation Research Part A: Policy and
Wang, Z. H., and W. J. He. 2017. “CO2 Emissions Efficiency and Marginal Practice 118: 783–802. doi:10.1016/j.tra.2018.10.025.
Abatement Costs of the Regional Transportation Sectors in China.” Zhao, Y., K. Triantis, P. Murray-Tuite, and P. Edara. 2011. “Performance
Transportation Research Part D: Transport and Environment 50: 83–97. Measurement of a Transportation Network with a Downtown Space
doi:10.1016/j.trd.2016.10.004. Reservation System: A Network-DEA Approach.” Transportation Research
Wei, R., X. Y. Liu, Y. J. Mu, L. M. Wang, A. Golub, and S. Farber. 2017. Part E: Logistics and Transportation Review 47 (6): 1140–1159. doi:10.1016/j.
“Evaluating Public Transit Services for Operational Efficiency and Access tre.2011.02.008.
Appendix

Table A1. The comprehensive benefits of three methods and their ranking (Sorting in parentheses).
2016 2017 2018 2019
Number DMU CCR DNDEA Our method CCR DNDEA Our method CCR DNDEA Our method CCR DNDEA Our method
1 Beijing 0.680(14) 0.469(14) 0.256(29) 0.613(14) 0.444(15) 0.165(33) 0.774(13) 0.613(13) 0.273(33) 0.715(15) 0.515(15) 0.266(31)
2 Tianjin 0.421(30) 0.533(8) 0.301(28) 0.426(29) 0.522(8) 0.258(28) 0.389(33) 0.524(21) 0.330(29) 0.436(32) 0.370(28) 0.320(30)
3 Shijiazhuang 0.667(15) 0.481(12) 0.415(16) 0.577(19) 0.415(17) 0.336(19) 0.722(16) 0.395(26) 0.396(25) 0.530(24) 0.368(29) 0.385(25)
4 Taiyuan 0.762(7) 0.613(5) 0.399(18) 0.783(8) 0.574(7) 0.448(11) 0.746(15) 0.563(18) 0.521(17) 0.614(18) 0.398(21) 0.412(23)
5 Hohhot 0.962(4) 0.553(6) 0.746(3) 0.859(6) 0.756(3) 0.706(3) 0.885(10) 0.609(15) 0.610(10) 0.645(17) 0.432(18) 0.506(15)
6 Shenyang 0.708(12) 0.546(7) 0.524(9) 0.592(17) 0.465(12) 0.466(10) 0.992(6) 0.770(3) 0.854(4) 0.967(6) 0.762(3) 0.813(4)
7 Changchun 0.687(13) 0.434(20) 0.543(6) 0.720(11) 0.419(16) 0.439(12) 0.648(20) 0.615(12) 0.579(14) 0.754(12) 0.606(11) 0.603(7)
8 Harbin 0.751(10) 0.387(21) 0.530(8) 0.755(9) 0.401(19) 0.517(8) 0.938(7) 0.742(6) 0.728(6) 0.901(7) 0.648(7) 0.657(6)
9 Shanghai 0.476(23) 0.343(27) 0.243 (30) 0.453(25) 0.275(30) 0.230(30) 0.511(25) 0.423(23) 0.382(26) 0.517(25) 0.389(23) 0.352(28)
10 Nanjing 0.342(33) 0.239(33) 0.222(32) 0.375(32) 0.215(33) 0.221(31) 0.431(31) 0.348(31) 0.323(30) 0.491(29) 0.376(27) 0.348(29)
11 Hangzhou 0.539(19) 0.453(17) 0.410(17) 0.594(16) 0.446(14) 0.398(13) 0.800(11) 0.661(10) 0.598(12) 0.531(23) 0.416(20) 0.382(26)
12 Hefei 0.587(17) 0.483(11) 0.434(14) 0.533(20) 0.376(21) 0.396(14) 0.585(23) 0.418(24) 0.492(19) 0.549(22) 0.397(22) 0.445(18)
13 Fuzhou 0.464(25) 0.378(24) 0.343(22) 0.453(25) 0.376(21) 0.315(23) 0.479(29) 0.368(28) 0.398(24) 0.499(27) 0.367(30) 0.388(24)
14 Nanchang 0.528(20) 0.278(32) 0.365(21) 0.510(22) 0.284(29) 0.370(16) 0.508(26) 0.343(32) 0.415(23) 0.556(21) 0.377(26) 0.416(21)
15 Jinan 0.461(27) 0.444(18) 0.439(13) 0.401(31) 0.398(20) 0.369(17) 0.648(20) 0.543(19) 0.544(15) 0.494(28) 0.389(23) 0.445(18)
16 Zhengzhou 0.473(24) 0.353(26) 0.308(27) 0.477(24) 0.329(25) 0.326(21) 0.720(17) 0.576(17) 0.635(8) 0.739(14) 0.578(12) 0.551(11)
17 Wuhan 0.523(21) 0.335(28) 0.369(19) 0.582(18) 0.346(23) 0.328(20) 0.654(19) 0.626(11) 0.526(16) 0.704(16) 0.577(13) 0.544(12)
18 Changsha 0.514(22) 0.316(29) 0.326(25) 0.443(27) 0.319(27) 0.318(22) 0.483(27) 0.357(30) 0.369(27) 0.516(26) 0.335(32) 0.376(27)
19 Guangzhou 0.586(18) 0.379(23) 0.327(24) 0.597(15) 0.327(26) 0.270(26) 0.663(18) 0.534(20) 0.505(18) 0.744(13) 0.499(16) 0.565(10)
20 Haikou 1.000(1) 0.713(4) 0.698(4) 1.000(1) 0.654(6) 0.563(6) 0.893(8) 0.499(22) 0.465(21) 0.882(8) 0.489(17) 0.461(17)
21 Chongqing 0.759(9) 0.454(15) 0.456(12) 0.754(10) 0.410(18) 0.356(18) 0.893(8) 0.741(7) 0.628(9) 0.881(9) 0.649(6) 0.543(13)
22 Chengdu 0.459(28) 0.354(25) 0.338(23) 0.435(28) 0.238(32) 0.266(27) 0.540(24) 0.412(25) 0.439(22) 0.487(30) 0.389(23) 0.421(20)
23 Guiyang 0.710(11) 0.514(10) 0.522(10) 1.000(1) 0.664(5) 0.565(5) 1.000(1) 0.821(2) 0.777(5) 1.000(1) 0.790(2) 0.659(5)
24 Kunming 0.464(25) 0.480(13) 0.368(20) 0.528(21) 0.472(10) 0.304(24) 0.615(22) 0.598(16) 0.485(20) 0.582(19) 0.574(14) 0.468(16)
25 Xi’an 0.848(6) 0.531(9) 0.536(7) 1.000(1) 0.518(9) 0.637(4) 1.000(1) 0.747(5) 0.867(3) 1.000(1) 0.681(5) 0.822(3)
26 Lanzhou 1.000(1) 0.996(1) 1.001(1) 1.000(1) 0.900(1) 0.993(1) 1.000(1) 0.972(1) 0.954(1) 1.000(1) 0.856(1) 0.972(1)
27 Xining 1.000(1) 0.745(3) 0.785(2) 1.000(1) 0.778(2) 0.814(2) 1.000(1) 0.761(4) 0.869(2) 1.000(1) 0.740(4) 0.936(2)
28 Yinchuan 0.865(5) 0.810(2) 0.603(5) 0.803(7) 0.718(4) 0.546(7) 1.000(1) 0.710(8) 0.607(11) 1.000(1) 0.617(9) 0.599(9)
29 Urumqi 0.762(7) 0.454(15) 0.489(11) 0.714(12) 0.472(10) 0.470(9) 0.787(12) 0.681(9) 0.652(7) 0.860(10) 0.645(8) 0.601(8)
30 Qingdao 0.442(29) 0.387(21) 0.315(26) 0.498(23) 0.341(24) 0.286(25) 0.480(28) 0.387(27) 0.343(28) 0.568(20) 0.432(18) 0.414(22)
31 Ningbo 0.345(32) 0.307(30) 0.214(33) 0.420(30) 0.288(28) 0.235(29) 0.432(30) 0.310(33) 0.303(32) 0.370(33) 0.277(33) 0.261(32)
32 Shenzhen 0.406(31) 0.284(31) 0.240(31) 0.345(33) 0.251(31) 0.172(32) 0.416(32) 0.363(29) 0.306(31) 0.458(31) 0.346(31) 0.257(33)
33 Xiamen 0.597(16) 0.438(19) 0.418(15) 0.663(13) 0.452(13) 0.384(15) 0.760(14) 0.612(14) 0.589(13) 0.823(11) 0.614(10) 0.533(14)
Mean 0.630 0.469 0.439 0.633 0.450 0.408 0.709 0.565 0.538 0.691 0.512 0.507
TRANSPORTATION LETTERS
13

You might also like