Anda di halaman 1dari 22

Transportation Research Part A 94 (2016) 650671

Contents lists available at ScienceDirect

Transportation Research Part A


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

Centralized resource allocation with emission resistance in a


two-stage production system: Evidence from a Taiwans
container shipping company
Ming-Miin Yu, Li-Hsueh Chen
Department of Transportation Science, National Taiwan Ocean University, No. 2 Pei-Ning Road, Keelung 20224, Taiwan

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

Article history: This paper proposes a centralized network data envelopment analysis model that combines
Received 1 April 2016 the centralized data envelopment analysis model and network data envelopment analysis
Received in revised form 26 July 2016 to allocate resources among sub-units. In the proposed model, this paper also considers the
Accepted 4 October 2016
situations in which undesirable outputs are jointly produced with desirable outputs, the
Available online 9 November 2016
reduction of undesirable outputs is associated with the reduction of energy inputs, and
some inputs are dedicated to the specific sub-unit while some inputs are shared among
Keywords:
sub-units. To comprehensively investigate this issue, two cases are discussed. One case
Container shipping company
Resource allocation
explores the situation in which common inputs are shared among the first process of
Centralized data envelopment analysis sub-units, while the other case explores the situation in which common inputs are also
Network data envelopment analysis shared among two processes of sub-units. The proposed model is illustrated in an empirical
example of 14 Asian shipping routes operated by a Taiwanese container shipping company.
In order to avoid organizational resistance, minor and major adjustment policies are
demonstrated. The minor adjustment policy transfers inputs among routes but maintains
the original levels of input resources, whereas the major adjustment policy reduces the
total amount of input resources. The results provide valuable information for the central-
ized decision-maker on how to reallocate resources among the sub-units.
2016 Elsevier Ltd. All rights reserved.

1. Introduction

With the rapid growth of international trade, the demand for and capacity of container shipping services have increased
(Lau et al., 2013). However, globalization also has changed the way shipping companies operate, and caused the higher glo-
bal competition (Yang et al., 2013). Lau et al. (2013) have indicated that Indeed, the effective provision of container facilities
and services is considered to be an important attribute in deciding the competitiveness of ship operators. Therefore, in order
to improve or maintain the competitive advantage of container shipping companies, it is important to allocate limited
resources in an optimal way to individual routes to achieve higher overall performance.
However, previous shipping studies mainly aimed at ship operations strategy, such as network design (e.g., Gelareh et al.,
2010; Asgari et al., 2013), container routing (e.g., Wang et al., 2013), fleet deployment (e.g., Wang and Meng, 2012a), speed
optimization (e.g., Wang and Meng, 2012b), and risk management (Hoorn and Knapp, 2015). Studies in overall resource

Corresponding author.
E-mail addresses: yumm@mail.ntou.edu.tw (M.-M. Yu), yoannachen@gmail.com (L.-H. Chen).

http://dx.doi.org/10.1016/j.tra.2016.10.003
0965-8564/ 2016 Elsevier Ltd. All rights reserved.
M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671 651

allocation of the shipping industry are rather limited. In order to fill this void, this paper constructs resource allocation mod-
els by the data envelopment analysis (DEA) approach.
In the problem of resource allocation, DEA dominates other methods, because it provides the possibility of considering
the effect of feasible allocation plans based on the empirical description of the production possibility set (Li et al., 2009).
Since the development of DEA by Charnes et al. (1978), DEA has been widely used to evaluate the relative efficiency of deci-
sion making units (DMUs). It has also been recently applied to solve the resource allocation problem. In traditional DEA mod-
els, each DMU is independently projected onto the efficient frontier. However, in practice, situations occur when all of the
DMUs are under the control of a centralized decision-maker. The centralized decision-maker may be more interested in
increasing the efficiency of the overall organization instead of enhancing the efficiency of each DMU separately. Hence,
the aim of the centralized decision-maker is to allocate resources to optimize the operations of all the DMUs globally. In
recent years, many authors have developed a number of DEA models for resource allocation in a centralized environment.
Lozano and Villa (2004, 2005) and Lozano et al. (2004) introduced the concept of centralized DEA (CDEA) models to optimize
the total resource consumption of all DMUs in an organization, rather than considering the consumption of each DMU sep-
arately. Korhonen and Syrjnen (2004) proposed a multi-objective linear programming DEA approach to maximize the total
amount of outputs of all DMU simultaneously. Asmild et al. (2009) modified the model proposed by Lozano and Villa (2004)
to only consider adjustments of inefficient units. Lozano et al. (2009) established a CDEA model for resource allocation
involving three objectives: maximizing aggregate desirable production, minimizing the aggregate consumption of input
resources, and minimizing aggregate undesirable emissions. Hosseinzadeh Lotfi et al. (2010) proposed a CDEA model based
on the enhanced Russell measure. The proposed model allowed all DMUs to be easily projected onto the efficient frontier by
solving only one model. Lozano et al. (2011) established a number of non-radial CDEA models to deal with resource reallo-
cation in the Spanish national port system. Hosseinzadeh Lotfi et al. (2012) presented a stochastic centralized model for
resource allocation to deal with the problem of stochastic inputs and outputs. Fang (2013) extended the models proposed
by Lozano and Villa (2004) and Asmild et al. (2009) to a more general form. Yu et al. (2013) proposed a modified CDEA model
with a Russell measure to study the human resource downsizing problem for Taiwans airports. Mar-Molinero et al. (2014)
simplified the model proposed by Lozano and Villa (2004) to make the centralized model easier to implement in many sit-
uations. Chang et al. (2015) applied the CDEA model to analyze the performance of container terminals controlled by one
liner shipping company. Fang (2015) considered the situation in which a dramatic change in the resources might cause orga-
nizational resistance, and then provided step-by-step improvement paths based on the new CDEA model to enhance the effi-
ciency of all DMUs. Fang and Li (2015) developed a CDEA model for resource allocation by maximizing revenue efficiency.
Since the centralized decision-maker of the container shipping company pursues maximal organizational performance, it is
more appropriate to use the CDEA model to allocate resources across routes.
However, container shipping companies possess some particular operational characteristics. First, the capacities provided
by container shipping companies are non-storable. In other words, the quantities of consumed service may be not equal to
the quantities of produced service. Second, the carbon emissions are jointly produced with the freight shipping services.
Since carbon emissions result from fuel consumption, carbon emission reduction associated with energy saving. Third, some
of the inputs, such as labor, are common resources shared among routes. Those common inputs cannot be split up. In order
to adequately allocate resources, those characteristics need to be taken into consideration when the performance of routes is
evaluated.
The non-storable nature of the freight shipping service involves centralized decision-maker planning on how to promptly
sell the services to avoid incurring losses. The non-storable nature of the freight shipping service means that ships must
depart on predetermined schedules, even if these ships are not full. Since the unfilled space cannot be held and used next
time, the capacities provided by a container shipping company cannot be stored. As a result, the linking activities form a net-
work in which the capacities produced in the production process are treated as inputs in the service process. Hence, a con-
tainer shipping company can be treated as a two-stage production system. In the first stage (the production process), inputs
are used to produce shipping capacity. In the second stage (the service process), existing capacity is applied to provide freight
shipping services. Fielding (1987) first introduced the concept of non-storable commodities in a public transit system and
proposed three performance measures: cost efficiency, service effectiveness and cost effectiveness. The researcher defined
cost efficiency as the ratio of capacity to inputs, service effectiveness as the ratio of consumption to capacity, and cost effec-
tiveness as the ratio of consumption to inputs. Kao and Hwang (2008, 2010) further indicated that ignoring the internal
structure of an operational process may misestimate the performance. In addition, the efficiencies in the production and ser-
vice processes may be inconsistent and must be respectively measured. However, the previous studies ignore the need to
separate the production process and service process. In response to the important role of interactions of sub-processes in
measuring the overall performance of a container shipping company, this paper constructs an alternative CDEA model in
a network framework which combines the network DEA (NDEA) model and CEDA model to allocate resources among ship-
ping routes. In addition, common input resources not only are shared among routes, but also contribute to each of these two
processes simultaneously. By dividing liner shipping services into production and service process, this paper can further
investigate how to allocate resources well in the internal structure of a container shipping company.
Carbon emissions are one of the outputs from container shipping companies. It behaves opposite to desirable outputs. It is
undesirable and expected to be minimized. However, carbon emissions are produced together with freight shipping service.
Hence, the trade-off between an increase in freight shipping service and a decrease in carbon emissions should be consid-
ered. Performance measurement with the incorporation of undesirable outputs has received greater attention. Scheel (2001)
652 M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671

divided the incorporation of undesirable outputs into two types: direct and indirect approaches. The direct approach uses the
weak disposability assumption of outputs to replace the strong disposability assumption of outputs (Fre et al., 1989, 1993,
2004; Chung et al., 1997; Tone, 2004; Zhou et al., 2006; Zhou et al., 2007; Fukuyama and Weber, 2010; Lozano and Gutirrez,
2011). The indirect approaches are based on the transformation of original data for undesirable outputs and the use of tra-
ditional DEA models (Golany and Roll, 1989; Ali and Seiford, 1990; Hailu and Veeman, 2001; Scheel, 2001; Seiford and Zhu,
2002; Hua et al., 2007). Since carbon emissions result from the use of fuels, carbon emissions are not separable from energy
resource consumption. Guo et al. (2011) proposed an energy conservation performance index to deal with the relationship
between carbon emissions and energy inputs. The researchers assumed that carbon emission reductions were proportional
to the energy reductions. In this paper, the direct approach combined with the concept of the energy conservation perfor-
mance index proposed by Guo et al. (2011) is applied to deal with the problem of undesirable outputs.
As discussed previously, in order to capture the operational characteristics of container shipping companies, this paper
will construct a CDEA model in a network framework, and incorporate the concepts of common inputs and the direct
approach combined with the energy conservation performance index into the proposed model. This paper will apply the idea
of Fukuyama and Weber (2009) to construct a centralized network DEA (CNDEA) model. The proposed model is a non-radial
measure that uses the Russell directional distance function (RDDF),1 which combines the directional distance function (DDF)
developed by Chamders et al. (1996) and slacks-based measure (SBM) proposed by Tone (2001), to allow the different changes
of inefficiency for each input and output. It can simultaneously account for all radial and non-radial slacks, in which non-radial
slacks are ignored in the radial measure. More importantly, it allows this paper to gauge the performance of each route in terms
of increasing desirable outputs and reducing undesirable outputs simultaneously. To sum up, by using the proposed CNDEA
model, this paper can deal with the problems of common inputs, energy inputs and undesirable outputs in a unified network
framework and adequately allocate resources for each route at the same time.
The contributions of this paper are as follows: In method, a novel resource allocation model based on the network struc-
ture is proposed to capture the non-storable characteristic of freight shipping services. The proposed model is based on a
non-radial measure that incorporates both non-radial and radial variables into the model, and classifies inputs into common
inputs, specific inputs and energy inputs, in which common inputs are shared among not only sub-units but also both pro-
cesses. The proposed model can reallocate resources by reduction or expansion of inputs among sub-units under a central-
ized decision-maker. It can also solve the problem of the carbon emissions being jointly produced with the freight shipping
services and their reduction associated with energy saving by linking the carbon emissions with energy reductions. Hence,
the proposed model can more globally describe the operational characteristics of container shipping companies and cover
the omissions of the existing literature. In addition, although the proposed model is used to explore the resource allocation
of a container shipping company, the idea behind the proposed model can be revised and extended to different industries
and fields. In practice, the proposed model provides a performance-based perspective to allocate resources of a container
shipping company. The firm can enhance its operational performance by reallocating input resources. In addition, minor
and major adjustment policies are provided. The decision-maker can choose the suitable policy to allocate resources based
on strategic planning. Hence, the proposed model is more flexible compared with the existing literature dealing with the
problem of resource allocation.
The remainder of this paper proceeds as follows. Section 2 reviews the relevant resource allocation literature in the trans-
portation sector. Section 3 presents a new approach for resource allocation. In Section 4, a case study of 14 shipping routes of
a container shipping company is used to illustrate the proposed approach. Finally, conclusions are summarized in Section 5.

2. Literature review

Resource allocation is a process of assigning scarce resources among various units to produce goods or services. It is a
form of strategic planning for achieving a vision and goals for the future. The vision and goals are accomplished through
the achievement of objectives. A large body of literature has designed various techniques to achieve various objectives of
resource allocation. In the transportation sector, the general transportation resource allocation literature focuses on the
implementation of transportation policies, and the objectives of transportation policies are often to provide an efficient, safe
and environmentally friendly transportation system. Those studies mainly investigated the problems of resource allocation
of transportation infrastructure, such as optimal resource allocation for safety improvement programs (e.g., Melachrinoudis
and Kozanidis, 2002; Mishra and Khasnabis, 2012; Mishra, 2013; Mishra et al., 2015; Murillo-Hoyos et al., 2015), and optimal
resource allocation for roadway infrastructure maintenance projects (e.g., Ahmed, 1983; Boyles, 2015). In addition, funding
allocation mechanisms for units accepting government support have been explored. For example, Mathew et al. (2010),
Mishra et al. (2010) and Mishra et al. (2013) developed various optimal funding allocation models for transit fleet manage-
ment. The above mentioned studies are based on the perspective of social welfare, in which governments provide support in
order to meet societys needs.

1
The Russell measure was first introduced by Fre and Lovell (1978), and then Fre et al. (1985) and Pastor et al. (1999) combined input and output Russell
measures to develop the new Russell graph measure and Enhanced Russell graph measure, respectively. Recently, Fukuyama and Weber (2009) developed a
generalized measure of inefficiency by introducing the DDF technology into SBM approach. This new measure can be thought of as a generalization of the
original Russell measure and is referred to as the directional Russell measure.
M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671 653

However, the transportation sector does not only care about social welfare. In the transportation sector, some firms are
for-profit organizations. The internal controls for those firms are also important. For example, transit firms need to consider
not only how to allocate limited funds for fleet improvement programs, but also how to efficiently utilize internal resources
to enhance performance. Hence, in the transportation sector, the issue of internal resource allocation has also attracted
researchers attention. Relevant literature has covered various problems of resource allocation for transportation systems,
such as airline fleet assignment (e.g., Hane et al., 1995; Rexing et al., 2000; Barnhart and Lohatepanont, 2004; Pita et al.,
2012), empty freight car distribution (e.g., White and Bomberault, 1969; Holmberg et al., 1998; Bektas et al., 2009), and
human resource allocation (Liou and Tzeng, 2009; Yu et al., 2013). In the shipping industry, most studies have mainly
focused on the problem of empty container allocation (e.g., Crainic et al., 1993; Cheung and Chen, 1998; Lam et al., 2007;
Chou et al., 2010). Only Chang et al. (2015) investigated the internal resource reallocation among a number of container ter-
minals supervised by one shipping company. However, according to the authors knowledge, no studies have explored the
problem of internal resource allocation among routes of a shipping company under a centralized perspective. To fill this void,
this paper proposes a CNDEA model to allocate resources among the shipping routes of a container shipping company.

3. Methodology

Since shipping routes are operated under the supervision of a centralized decision-maker, it is more appropriate to allo-
cate resources under the CDEA perspective. Due to the non-storable characteristic of freight shipping services, the operation
of a container shipping company can be divided into production and service processes. In addition, this paper is concerned
with not only how to allocate common and specific inputs efficiently, but also how to save energy resources and reduce car-
bon emissions. Hence, this paper designs a CNDEA model which considers common inputs, specific inputs, energy inputs,
desirable outputs and undesirable outputs in a unified network model, to investigate the allocation of resources among
routes of a container shipping company. In particular, two cases are used to construct the resource allocation model. One
assumes that common inputs are just shared among the production processes of each route. The other allows that common
inputs can be shared among both processes of each route. Both cases consist of two phases. In the first phase, the centralized
decision-maker tries to maximize the production of aggregate desirable outputs and minimize the production of aggregate
undesirable outputs of all routes at current given resources. In the second phase, the centralized decision-maker tries to real-
locate resources to routes in an efficient way by the non-proportional adjustment of inputs at optimal levels of aggregate
desirable outputs and undesirable outputs obtained from the first phase. Since a policy involving rapid change could lead
to organizational resistance and reduced performance (Golembiewski et al., 1976; Argris and Schn, 1978; Yu et al.,
2013), a minor adjustment policy and major adjustment policy are considered in the second phase. The minor adjustment
policy assumes that the centralized decision-maker can only transfer common and specific inputs among routes under a
fixed total amount of common and specific inputs. The major adjustment policy assumes that the centralized decision-
maker can cut the amount of common and specific inputs in each route and transfer common and specific inputs among
routes.
Before formulating the models, the required notation should be introduced:

N number of routes
I1 number of common inputs
I2 number of specific inputs
I3 number of energy inputs
P number of intermediate products
R number of desirable outputs
Q number of undesirable outputs
j, k (j, k = 1, . . . , N) indexes for routes
i1 (i1 = 1, . . . , I1) indexes for common inputs
i2 (i2 = 1, . . . , I2) indexes for specific inputs
i3 (i3 = 1, . . . , I3) indexes for energy inputs
p (p = 1, . . . , P) indexes for intermediate products
r (r = 1, . . . , R) indexes for desirable outputs
q (q = 1, . . . , Q) indexes for undesirable outputs
xi1 the total amount of common input i1
xi2 J the amount of specific input i2 consumed by route j
xi3 J the amount of energy input i3 consumed by route j
zpj the amount of intermediate products p produced by route j
yrj the amount of desirable output r produced by route j
bqj the amount of undesirable output q produced by route j
hr the inflation of aggregate desirable output r

(continued on next page)


654 M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671

dq the deflation of aggregate undesirable output q


s
i1 k the negative slack for common input i1 of route k
s
i1 k
the positive slack for common input i1 of route k
s1
i1 k the negative slack for common input i1 of route k in the production process
s1
i1 k
the positive slack for common input i1 of route k in the production process
s2
i1 k the negative slack for common input i1 of route k in the service process
s2
i1 k
the positive slack for common input i1 of route k in the service process
s
i2 k the negative slack for specific input i2 of route k
s
i2 k
the positive slack for specific input i2 of route k
k11k ; . . . ; k1jk the intensity variables for projecting route k in the production process
k21k ; . . . ; k2jk the intensity variables for projecting route k in the service process
ai1 J the shared rate of the common input quantities xi1 for route j
a1i1 J the shared rate of the common input quantities xi1 for route j in the production process
a2i1 J the shared rate of the common input quantities xi1 for route j in the service process
w1i1 the total shared rate on production process of the common input quantities xi1
w2i1 the total shared rate on service process of the common input quantities xi1

3.1. Case I: common inputs shared among production processes of each route

First, this paper presents a simpler case in which common inputs are only shared among the first process of each route.
Each route uses common inputs, specific inputs and energy inputs to produce intermediate products in the production pro-
cess. In the service process, intermediate products are applied to jointly produce desirable outputs and undesirable outputs.
The operational structure of a container shipping company based on common inputs shared among routes is shown in Fig. 1.
As discussed previously, two phases are used in the proposed CNDEA model. In the first phase, the maximal increase in
desirable outputs and the maximal decrease in undesirable outputs are sought, while additional reduction of any input is
pursued in the second phase. The proposed CNDEA model is formulated as follows:

Phase I:

The Phase I model estimates the average inefficiency of routes by solving the following NDEA model based on the RDDF:
PR PQ
r1 hr q1 dq
Max 1
R Q
Subject to
a. Common input constraint:
X
N
k1jk ai1 J xi1 6 ai1 k xi1 ; i1 1; 2; . . . ; I1 ; k 1; . . . ; N 1:1
j1

b. Specific input constraint:


X
N
k1jk xi2 J 6 xi2 k ; i2 1; 2; . . . ; I2 ; k 1; . . . ; N 1:2
j1

c. Energy input constraint:


X
N
k1jk xi3 J 6 xi3 k ; i3 1; 2; . . . ; I3 ; k 1; . . . ; N 1:3
j1

Eqs. (1.1)(1.3) imply that common, specific and energy input combinations are technologically feasible for all routes,
respectively.

d. Intermediate product constraints:

Regarding the intermediate product z, three rational treatments exist: (i) one can leave it unchanged, (ii) it is better to
increase from the first stage, and simultaneously should be reduced from the second stage, or (iii) it is allowable to increase
or decrease it in an optimal way for the first stage and second stage. In our approach, it seems to be rational to allow the free
M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671 655

Fig. 1. The operational structure of a container shipping company based on common inputs shared among production processes of each route.

adjustment of the intermediate products. Hence, this paper has used the third approach, and the intermediate product con-
straints are shown as follows:
X
N
k1jk zpj P zpk ; p 1; 2; . . . ; P; k 1; . . . ; N 1:4:1
j1

X
N
k2jk zpj 6 zpk ; p 1; 2; . . . ; P; k 1; . . . ; N 1:4:2
j1

Combine Eqs. (1.4.1) and (1.4.2), the intermediate product constraints can be written as follows:
X
N X
N
k1jk zpj P k2jk zpj ; p 1; 2; . . . ; P; k 1; . . . ; N 1:4
j1 j1

e. Desirable output constraints:


X
N X
N X
N
k2jk yrj P 1 hr yrk ; r 1; 2; . . . ; R 1:5
k1 j1 k1

The constraints in Eq. (1.5) seek to non-proportionally increase the total amount of each desirable output as much as pos-
sible, and remain in the feasible aggregate output set.
X
N
k2jk yrj P yrk ; r 1; 2; . . . ; R; k 1; . . . ; N 1:6
j1

Eq. (1.6) implies that desirable output combinations are technologically feasible for all routes.

f. Undesirable output constraints:

To model the phenomenon that the undesirable outputs are jointly produced with desirable output, the null-jointness
and weak disposability assumptions need to be imposed on the production technology. The null-jointness assumption states
that undesirable outputs are unavoidable. The only way to eliminate undesirable outputs is to cease production. Due to the
null-jointness assumption, b can be zero only if y is zero. Weak disposability assumes that undesirable outputs are reduced in
the same proportion as desirable outputs. In order to impose the weak disposability assumption on the production technol-
ogy, the equality constraints on aggregate undesirable outputs and individual undesirable outputs are written as follows
(Fre et al., 1994):
X
N X
N X
N
k2jk bqj 1  dq bqk ; q 1; 2; . . . ; Q 1:7
k1 j1 k1
656 M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671

X
N
k2jk bqj 1  dq bqk ; q 1; 2; . . . ; Q ; k 1; . . . ; N 1:8
j1

Similar to Eqs. (1.5) and (1.6), Eqs. (1.7) and (1.8) are imposed restrictions on the aggregate and individual undesirable
output sets. However, the difference is that Eq. (1.7) pursues a non-proportional decrease in the total amount of each unde-
sirable output as much as possible.

g. Variable return-to-scale constraints:

X
N
k1jk 1; k 1; . . . ; N 1:9
j1

X
N
k2jk 1; k 1; . . . ; N 1:10
j1

Eqs. (1.9) and (1.10) represent the assumption of variable returns to scale in this model.

h. Shared rate constraints:

^ _
a 6 ai1 J 6 ai1 J ; i1 1; 2; . . . ; I1 ; j 1; . . . ; N
i1 J 1:11

X
N
ai1 J 1; i1 1; 2; . . . ; I1 1:12
j1

^ _
Eqs. (1.11) and (1.12) are imposed bounds for common inputs, in which ai1 J ; ai1 J are associated with lower and upper
bounds for the proportion of the common input xi1 J (i1 = 1, 2, . . . , I1) assigned to route j (j = 1, . . . , N), respectively, which
are decided exogenously.

k1jk ; k2jk P 0; j 1; . . . ; N; k 1; . . . ; N 1:13

Eq. (1.13) shows that any intensity variable cannot be less than zero.
Note that Model (1) is non-linear. It should be transformed into linear programming as shown in Appendix A.

Phase II:

In order to avoid organizational resistance and reduced performance, there are two policies for resource adjustment. The
minor adjustment policy maintains the total amount of common and specific inputs unchanged, while the major adjustment
policy cuts the total amount of common and specific inputs. However, situations occur when some inputs need to be
increased to improve performance. In order to consider these situations, the proposed CNDEA model is allowed to increase
the amount of allocable resources.

Minor adjustment policy


If hr r 1; . . . ; R; dq q 1; . . . ; Q and ai1 k i1 1; . . . ; I1 ; k 1; . . . ; N denote the optimal values obtained from Model (1),
then the Phase II model for the minor adjustment policy can be solved by following model:
X
I1 X
N s  s
i k i k
X
I2 X
N s  s
i k i k
Max 1

1
2 2
2
i1 1 k1
a i 1 k xi 1 i2 1 k1
xi2 k

Subject to

a. Common input constraints:


X
N
k1jk ai1 J xi1 ai1 k xi1  si1 k si1 k ; i1 1; 2; . . . ; I1 ; k 1; . . . ; N 2:1
j1

Eq. (2.1) helps determine whether the current amount of individual common inputs is appropriately allocated among routes,
i.e., whether individual routes need to increase or decrease individual common inputs in the optimal solution.
X
N X
N
si1 k si1 k ; i1 1; 2; . . . ; I1 2:2
k1 k1
M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671 657

Eq. (2.2) ensures that the total amount of individual common inputs is equal to the original level.

b. Specific input constraints:

X
N
k1jk xi2 J xi2 k  si2 k si2 k ; i2 1; 2; . . . ; I2 ; k 1; . . . ; N 2:3
j1

X
N X
N
si2 k si2 k ; i2 1; 2; . . . ; I2 2:4
k1 k1

Eqs. (2.3) and (2.4) are similar to Eqs. (2.1) and (2.2). Eq. (2.3) determines whether the current amount of individual speci-
fic inputs is appropriate, and Eq. (2.4) guarantees the total amount of individual specific inputs is unchanged.

c. Energy input constraint:

X
N
k1jk xi3 J 6 1  diq xi3 k ; i3 1; 2; . . . ; I3 ; k 1; . . . ; N 2:5
3
j1

Eq. (2.5) links the relationship between undesirable output reductions and energy resource reductions, and implies that
the amount of energy inputs should decrease in proportion to the amount of undesirable outputs.

d. Intermediate product constraint:

X
N X
N
k1jk zpj P k2jk zpj ; p 1; 2; . . . ; P; k 1; . . . ; N 2:6
j1 j1

Similar to Eq. (1.4) in Model (1), Eq. (2.6) shows the restrictions of the intermediate products.

e. Desirable output constraints:

X
N X
N X
N
k2jk yrj P 1 hr yrk ; r 1; 2; . . . ; R 2:7
k1 j1 k1

X
N
k2jk yrj P yrk ; r 1; 2; . . . ; R; k 1; . . . ; N 2:8
j1

Eq. (2.7) can be seen as using the reallocation perspective to make individual desirable outputs reach their optimal aggre-
gate levels under the CNDEA perspective, while Eq. (2.8) is compared with Eq. (1.6). Eq. (2.8) implies that desirable output
combinations are technologically feasible for all routes.

f. Undesirable output constraints:

X
N X
N X
N
k2jk bqj 1  dq bqk ; q 1; 2; . . . ; Q 2:9
k1 j1 k1

X
N
k2jk bqj 1  dq bqk ; q 1; 2; . . . ; Q ; k 1; . . . ; N 2:10
j1

Eqs. (2.9) and (2.10) are analogous to Eqs. (2.7) and (2.8), respectively. Eq. (2.9) can be seen as using the reallocation per-
spective to make individual undesirable outputs reach their optimal aggregate levels under the CNDEA perspective. Eq.
(2.10) implies that undesirable output combinations are technologically feasible for all routes.

g. Variable return-to-scale constraints:

X
N
k1jk 1; k 1; . . . ; N 2:11
j1
658 M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671

X
N
k2jk 1; k 1; . . . ; N 2:12
j1

k1jk ; k2jk P 0; j 1; . . . ; N; k 1; . . . ; N 2:13

si1 k ; si1 k ; si2 k ; si2 k P 0; i1 1; . . . ; I1 ; i2 1; . . . ; I2 ; k 1; . . . ; N 2:14

Similar to Eqs. (1.9) and (1.10), Eqs. (2.11) and (2.12) represent that this model is based on the assumption of variable
returns to scale. Eq. (2.13) is similar to Eq. (1.13), indicating that any intensity variable cannot be less than zero. Eq.
(2.14) shows that slacks should be equal to or greater than zero.

Major adjustment policy


Similarly, if hr r 1; . . . ; R; dq q 1; . . . ; Q and ai1 k i1 1; . . . ; I1 ; k 1; . . . ; N denote the optimal values obtained from
Model (1), then the Phase II model for the major adjustment policy can be solved by the following model:

X
I1 X
N s  s
i k i k
X
I2 X
N s  s
i k i k
Max 1 1
2 2
3
i1 1 k1
ai1 k xi1 i2 1 k1
xi2 k

Subject to

a. Common input constraints:


X
N
k1jk ai1 J xi1 ai1 k xi1  si1 k si1 k ; i1 1; 2; . . . ; I1 ; k 1; . . . ; N 3:1
j1

X
N X
N
si1 k > si1 k ; i1 1; 2; . . . ; I1 3:2
k1 k1

b. Specific input constraints:


X
N
k1jk xi2 J xi2 k  si2 k si2 k ; i2 1; 2; . . . ; I2 ; k 1; . . . ; N 3:3
j1

X
N X
N
si2 k > si2 k ; i2 1; 2; . . . ; I2 3:4
k1 k1

and
Constraints (2.5)(2.14).
Similar to Eqs. (2.1) and (2.3), Eqs. (3.1) and (3.3) determine whether the current amounts of individual common inputs
and individual specific inputs are appropriately allocated among routes, respectively. The only difference between Model (3)
and Model (2) lies in Eqs. (3.2) and (3.4) versus Eqs. (2.2) and (2.4). Eqs. (3.2) and (3.4) allow that the total amounts of indi-
vidual common and specific inputs after resource reallocation can be less than the original levels, respectively.

3.2. Case II: common inputs shared among both processes of each route

In this case, common inputs are shared among routes and processes. Each route uses specific inputs, energy inputs and
part of common inputs to produce the intermediate products in the production process. In the service process, intermediate
products and the other part of common inputs are used to jointly produce the desirable outputs and undesirable outputs. The
operational structure of a container shipping company can be modified as shown in Fig. 2.
There are also two phases in the second CNDEA model. The objective function in the first phase is the same as Model (1),
i.e., to maximize increases in desirable outputs and decreases in undesirable outputs given the current resource allocation.
On the other hand, the objective function in the second phase is to maximize the net slacks of common inputs in the pro-
duction and service processes and specific inputs. The second CNDEA model is formulated as follows:

Phase I:

PR PQ
r1 hr q1 dq
Max 4
R Q
Subject to
M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671 659

Fig. 2. The operational structure of a container shipping company based on common inputs shared among both processes of each route.

a. Common input constraint in the production process:


X
N
k1jk a1i1 J xi1 6 a1i1 k xi1 ; i1 1; 2; . . . ; I1 ; k 1; . . . ; N 4:1
j1

b. Specific input constraint:


X
N
k1jk xi2 J 6 xi2 k ; i2 1; 2; . . . ; I2 ; k 1; . . . ; N 4:2
j1

c. Energy input constraint:


X
N
k1jk xi3 J 6 xi3 k ; i3 1; 2; . . . ; I3 ; k 1; . . . ; N 4:3
j1

d. Intermediate product constraint:


X
N X
N
k1jk zpj P k2jk zpj ; p 1; 2; . . . ; P; k 1; . . . ; N 4:4
j1 j1

e. Common input constraint in the service process:


X
N
k2jk a2i1 J xi1 6 a2i1 k xi1 ; i1 1; 2; . . . ; I1 ; k 1; . . . ; N 4:5
j1

f. Desirable output constraints:


X
N X
N X
N
k2jk yrj P 1 hr yrk ; r 1; 2; . . . ; R 4:6
k1 j1 k1

X
N
k2jk yrj P yrk ; r 1; 2; . . . ; R; k 1; . . . ; N 4:7
j1

g. Undesirable output constraints:


X
N X
N X
N
k2jk bqj 1  dq bqk ; q 1; 2; . . . ; Q 4:8
k1 j1 k1

X
N
k2jk bqj 1  dq bqk ; q 1; 2; . . . ; Q ; k 1; . . . ; N 4:9
j1

h. Variable return-to-scale constraints:


X
N
k1jk 1; k 1; . . . ; N 4:10
j1
660 M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671

X
N
k2jk 1; k 1; . . . ; N 4:11
j1

i. Shared rate constraints:


^ _
a 6 a1i1 J 6 a1i1 J ; i1 1; 2; . . . ; I1 ; j 1; . . . ; N
1
i1 J 4:12

^ _
a 6 a2i1 J 6 a2i1 J ; i1 1; 2; . . . ; I1 ; j 1; . . . ; N
2
i1 J 4:13

X
N
a1i1 J w1i1 ; i1 1; 2; . . . ; I1 4:14
j1

X
N
a2i1 J w2i1 ; i1 1; 2; . . . ; I1 4:15
j1

w1i1 w2i1 1; i1 1; 2; . . . ; I1 4:16

k1jk ; k2jk P 0; j 1; . . . ; N; k 1; . . . ; N 4:17


^ _ ^ _
where a1i1 J ; a1i1 J and a2i1 J ; a2i1 J are lower and upper bounds of shared rates of common input xi1 J (i1 = 1, 2, . . . , I1) for the produc-
tion and service processes assigned to route j (j = 1, . . . , N), respectively, which are decided exogenously. Eqs. (4.1)(4.4),
(4.6)(4.11) and (4.17) are analogous to Eqs. (1.1)(1.10) and (1.13), respectively. The main difference between Model (4)
and Model (1) is Eq. (4.5), which shows common inputs are also allocated to the service process. The shared rates of common
inputs in the production and service processes should be simultaneously considered. Eqs. (4.12) and (4.14) are imposed
bounds for common inputs for the production process, in which Eq. (4.14) indicates the total shared rate on the production
process. Eqs. (4.13) and (4.15) are imposed bounds for common inputs for the service process, in which Eq. (4.15) represents
the total shared rate on the service process. Eq. (4.16) shows that the total shared rate is equal to one.
Similarly, Model (4) is non-linear. It should be transformed into a linear one as represented in Appendix B.

Phase II:

If hr r 1; . . . ; R; dq q 1; . . . ; Q ; a1


i1 k i1 1; . . . ; I 1 ; k 1; . . . ; N and ai1 k i1 1; . . . ; I1 ; k 1; . . . ; N denote the optimal
2

values obtained from Model (4), then the Phase II model for minor and major adjustment policies can be respectively solved
by the following models:

Minor adjustment policy

X
I1 X
N s1  s1
i k i k
X
I1 X
N s2  s2
i k i k
X
I2 X
N s  s
i k i k
Max 1 1
1 1
2 2
5
i1 1 k1
a1
i 1 k xi 1 i1 1 k1
a2
i 1 k xi 1 i2 1 k1
xi 2 k

Subject to

a. Common input constraints in the production process:


X
N
k1jk a1
i1 J xi1 ai1 k xi1  si1 k si1 k ;
1 1 1
i1 1; 2; . . . ; I1 ; k 1; . . . ; N 5:1
j1

X
N X
N

i1 k
s1 i1 k ;
s1 i1 1; 2; . . . ; I1 5:2
k1 k1

b. Specific input constraints:


X
N
k1jk xi2 J xi2 k  si2 k si2 k ; i2 1; 2; . . . ; I2 ; k 1; . . . ; N 5:3
j1

X
N X
N
si2 k si2 k ; i2 1; 2; . . . ; I2 5:4
k1 k1
M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671 661

c. Energy input constraint:


X
N  
k1jk xi3 J 6 1  diq xi3 k ; i3 1; 2; . . . ; I3 ; k 1; . . . ; N 5:5
3
j1

d. Intermediate product constraint:


X
N X
N
k1jk zpj P k2jk zpj ; p 1; 2; . . . ; P; k 1; . . . ; N 5:6
j1 j1

e. Common input constraints in the service process:


X
N
k2jk a2
i1 J xi1 ai1 k xi1  si1 k si1 k ;
2 2 2
i1 1; 2; . . . ; I1 ; k 1; . . . ; N 5:7
j1

X
N X
N

i1 k
s2 i1 k ;
s2 i1 1; 2; . . . ; I1 5:8
k1 k1

f. Desirable output constraints:


X
N X
N X
N
k2jk yrj P 1 hr yrk ; r 1; 2; . . . ; R 5:9
k1 j1 k1

X
N
k2jk yrj P yrk ; r 1; 2; . . . ; R; k 1; . . . ; N 5:10
j1

g. Undesirable output constraints:


X
N X
N X
N
k2jk bqj 1  dq bqk ; q 1; 2; . . . ; Q 5:11
k1 j1 k1

X
N
k2jk bqj 1  dq bqk ; q 1; 2; . . . ; Q ; k 1; . . . ; N 5:12
j1

h. Variable return-to-scale constraints:


X
N
k1jk 1; k 1; . . . ; N 5:13
j1

X
N
k2jk 1; k 1; . . . ; N 5:14
j1

k1jk ; k2jk P 0; j 1; . . . ; N; k 1; . . . ; N 5:15

1 2 2 
i1 k ; si1 k ; si1 k ; si1 k ; si2 k ; si2 k P 0;
s1 i1 1; . . . ; I1 ; i2 1; . . . ; I2 ; k 1; . . . ; N 5:16

Eqs. (5.1)(5.6) and (5.9)(5.16) are analogous to Eqs. (2.1)(2.14), respectively. The difference between Model (5) and
Model (2) is that Model (5) adds Eqs. (5.7) and (5.8) to show that common inputs are also allocated to the service process.
Eq. (5.7) determines whether the current amount of individual common inputs is appropriately allocated among routes in
the service process. Eq. (5.8) ensures that, in the service process, the total amount of individual common inputs is equal
to the original level.

Major adjustment policy

X
I1 X
N s1  s1
i k i k
X
I1 X
N s2  s2
i k i k
X
I2 X
N s  s
i k i k
Max 1
1
1
1
2
1
2 2
6
i1 1 k1
a i 1 k xi 1 i1 1 k1
a i1 k xi1 i2 1 k1
xi 2 k
662 M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671

Subject to
a. Common input constraints in the production process:
X
N
k1jk a1
i1 J xi1 ai1 k xi1  si1 k si1 k ;
1 1 1
i1 1; 2; . . . ; I1 ; k 1; . . . ; N 6:1
j1

X
N X
N

i1 k >
s1 i1 k ;
s1 i1 1; 2; . . . ; I1 6:2
k1 k1

b. Specific input constraints:


X
N
k1jk xi2 J xi2 k  si2 k si2 k ; i2 1; 2; . . . ; I2 ; k 1; . . . ; N 6:3
j1

X
N X
N
si2 k > si2 k ; i2 1; 2; . . . ; I2 6:4
k1 k1

c. Common input constraints in the service process:


X
N
k2jk a2
i1 J xi1 ai1 k xi1  si1 k si1 k ;
2 2 2
i1 1; 2; . . . ; I1 ; k 1; . . . ; N 6:5
j1

X
N X
N

i1 k >
s2 i1 k ;
s2 i1 1; 2; . . . ; I1 6:6
k1 k1
and
Constraints (5.5), (5.6) and (5.9)(5.16).
Similar to Eqs. (5.1) and (5.7), Eqs. (6.1) and (6.5) determine whether the current amounts of individual common inputs in
the production and service processes are appropriately allocated among routes, respectively. Eq. (6.3) is similar to Eq. (5.3),
determining whether the current amount of individual specific inputs is appropriate. The only difference between Model (6)
and Model (5) lies in Eqs. (6.2), (6.4) and (6.6) versus Eqs. (5.2), (5.4) and (5.8). Eqs. (6.2), (6.4) and (6.6) allow that the total
amounts of individual common inputs in the production process, individual specific inputs and individual common inputs in
the service process after resource reallocation can be less than the original levels, respectively.

4. Empirical results

4.1. Data and variables

The data consists of 14 Asian shipping routes belonging to a Taiwanese container shipping company for the fourth quarter
in 2014. Since container shipping companies adjust shipping routes and calling ports depending on transportation demand,
and liner shipping must comply with schedules after the announcement, it is not suitable to adopt long-term data in the
container shipping industry. In practice, container shipping companies regularly announce new schedules. Hence, in order
to conform to the actual operating state, this paper adopts one quarters data to analyze the issue. These routes are super-
vised by a centralized decision-maker with the authority and power to allocate resources among these 14 routes. However,
because of confidentiality reasons, this paper is unable to release the name of this container shipping company from which
the data were obtained for this paper.
Following Shintani et al. (2007), Lun and Marlow (2011) and Bang et al. (2012), ten variables are chosen, including one
desirable output, one undesirable output, three specific inputs, one energy input, two intermediate products and two com-
mon inputs. The desirable output and undesirable output are revenue (R) and carbon emissions (CE), respectively. Because
combustion of fuel by marine diesel engines produces carbon emissions, such emissions are produced jointly with the freight
shipping services, and any carbon emission reduction is associated with fuel reduction. Although many methods are used to
calculate carbon emissions, this paper directly adopts the values provided by this company. The three specific inputs include
vessel capacity (VC), handling cost (HC) and other cost (OC), because these variables belong to the specific route. The energy
input is fuel cost (FC). The two intermediate products consist of TEU-nautical miles (TNM) and the number of destination
ports (DP). TEU-nautical miles and the number of destination ports are used to measure capacity. TEU-nautical miles is
the sum of the products obtained by multiplying the number of TEUs carried on each ship by the shipping distance. The
two common inputs consist of crew salary (CS) and empty container repositioning cost (ECRC), because vessels are moved
between routes in a shipping network (Kelareva et al., 2014) and the crew members do not only serve on a single route.2 In
addition, since the trade volume on some routes is imbalanced, the decision-maker needs to reposition empty containers

2
In general, labor can be measured by the number of employees or the total cost of labor. Due to reasons of data availability and the heterogeneity of labor,
the crew salary is used to measure labor in this paper.
M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671 663

(Agarwal and Ergun, 2008). A reasonable distribution of empty containers can not only provide sufficient capacity in the pro-
duction process, but also satisfy customer demand in the service process. Crew members are responsible not only for managing
and operating vessels, but also for providing quality customer services. Hence, these two common inputs are shared among
routes and processes, simultaneously. In summary, vessel capacity, handling cost, other cost, fuel cost, part of crew salary
and part of empty container repositioning cost are used to generate TEU-nautical miles and the number of destination ports
in the production process. In the service process, TEU-nautical miles, the number of destination ports, the other part of crew
salary and the other part of empty container repositioning cost are utilized to produce revenue and carbon emissions.3 In
response to this characteristic, this paper adopts Case II to make the empirical study. Table 1 displays the summary statistics
of all variables used in the paper.

4.2. Empirical result

Since these routes belong to the same company, a centralized decision-maker can simultaneously control these 14 routes
to minimize the total input consumption of all routes. In the empirical analysis, this paper assumes that the common inputs
are not only shared among the production process of routes, but also allocated to the service process. Hence, a route con-
sumes some of crew salary, some of empty container repositioning cost, vessel capacity, handling cost, other cost and fuel
cost to provide TEU-nautical miles and the number of destination ports in the production process. In the service process, it
uses remaining crew salary, remaining empty container repositioning, TEU-nautical miles and the number of destination
ports to obtain revenue and emit exhaust.

4.2.1. Minor adjustment policy analysis


First, this paper analyzes the minor adjustment policy. Following Models (4) and (5), the amount of the common and
specific inputs that can be transferred from one route to another route and the amount of the energy input that can be
reduced are calculated for maximum output performance. The results of the optimal values of shared rates and resource real-
location based on the minor adjustment policy are shown in Table 2. The values of a1 2 1 2
CS , aCS , aECRC and aECRC for each route are
reported in the second to fifth columns, the changes of common, specific and energy inputs are reported from the sixth to
thirteenth columns, and the optimal values of the inflation of the desirable output and the deflation of the undesirable out-
put are reported in the last row. The value of h is 0.269, and the value of d is 0.211, indicating that there is ample room for
output improvement in this company. The total amount of revenue of 14 routes can be increased by USD $20,768,752,
whereas the total amount of carbon emissions can be reduced by 15.843 million kg when the whole company becomes effi-
cient. Since the common inputs are also allocated between the production and service processes, four shared rates need to be
calculated. These rates are the shared rate of crew salary in the production process, the shared rate of crew salary in the ser-
vice process, the shared rate of empty container repositioning cost in the production process and the shared rate of empty
container repositioning cost in the service process. The results show that the percentage of crew salary apportioned to the
production process and service process are 74.7% and 25.3%, and the percentage of empty container repositioning cost appor-
tioned to the production process and service process are 76.5% and 23.5%. This means that this container shipping company
consumes more common inputs to produce more TEU-nautical miles and destination ports. In the production process, Routes
A and H should be assigned more common inputs than other routes, whereas Route N should be assigned more common
inputs than other routes in the service process.
In terms of resource reallocation, decreases in crew salaries of Routes A, D, H, I and M in the production process and those
of Routes G, H and N in the service process can be transferred to Routes B, C, E, F, G, J, K, L and N in the production process and
Routes A, B, C, D, E, F, J and M in the service process. Decreases in empty container repositioning costs of Routes A, C, D, H, I
and J in the production process and Routes C, G, H, J and N in the service process can be transferred to Routes B, E, F, G, K, L, M
and N in the production process and Routes A, B, D, E, F and M in the service process. Routes A, E, F, L and M can increase their
vessel capacities, which could be transferred from Routes B, C, D, G, H, I, J, K and N. Routes A, C, D, H, I, J, K and N can transfer
their handling costs to Routes B, E, F, G, L and M. Routes B, D, E, F, L and M can increase their other costs, which could be
transferred from Routes A, C, G, H, I, J, K and N. In addition, in order to reduce carbon emissions, the usage of fuel must
be reduced. The result indicates that the total amount of fuel costs of 14 routes should be decreased by USD $5,068,710.
Under the minor adjustment policy, this company can simultaneously increase its revenue and decrease emission levels
by transferring excess input resources among routes, sharing common inputs among routes and decreasing fuel cost. An
increase in revenue and a decrease in fuel cost can further increase profit, which is the primary concern of owners and stock-
holders. Since international shipping is a significant contributor to global greenhouse gas emissions (Bittner et al., 2012),
shipping companies have been under pressure to pay attention to carbon emission reduction to reduce greenhouse gas emis-
sions. This performance-based resource allocation approach can provide strong incentives for the decision-maker of this
company to reallocate input resources, because it simultaneously achieves the above mentioned benefits. In addition, the
decision-maker only needs to reallocate input resources, and does not need to decrease these resources except the fuel cost.
Hence, this policy can avoid organizational resistance, and is feasible in practice. However, the contributions of individual

3
Although some influence factors are also involved in the service process, such as the on-shore employees, these factors are excluded due to reasons of data
availability. However, the variables selected in this paper are sufficient to illustrate the operational process of the container shipping company.
664 M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671

Table 1
Descriptive statistics of variables from the 14 routes.

Total Mean Std. Dev. Max Min


Common inputs
CS (1000 USD) 625.13 44.65 44.91 132.76 0.00
ECRC (1000 USD) 5,579.21 398.52 156.92 725.43 191.02
Specific inputs
VC (TUEs) 151,959.00 10,854.21 4,385.68 18,200.00 5,200.00
HC (1000 USD) 22,828.34 1,630.60 758.06 2,889.58 647.20
OC (1000 USD) 34,768.28 2,483.45 1,097.05 3,966.61 1,078.70
Energy input
FC (1000 USD) 24,022.32 1,715.88 810.04 2,729.90 739.73
Intermediate products
TNM (TEU-nautical miles) 914.44 65.32 34.96 122.43 25.35
DP (Ports) 252.00 18.00 7.25 30.00 12.00
Desirable output
R (1000 USD) 77207.26 5514.80 2815.57 10285.25 2617.08
Undesirable output
CE (Million kg) 75.09 5.36 3.68 15.82 1.73

Note: TEU represents the twenty-foot equivalent unit.

inputs for the enhancement of performance are different. In order to increase revenue by 27% and reduce carbon emissions
by 21%, the decision-maker of this company should reallocate 28% of crew salary, 18% of empty container repositioning cost,
13% of vessel capacity, 14% of handling cost and 12% of other cost and decrease fuel cost by 21% among 14 routes, among
which the adjustment range of crew salary is the biggest. This means that the extent of employees being idle is more severe
than other situations. The transformation of manpower among these routes is important for this company.

4.2.2. Major adjustment policy analysis


Second, the major adjustment policy is analyzed. Following Models (4) and (6), the amount of the common and specific
inputs that can be transferred or reduced and the amount of the energy input that can be reduced are calculated for max-
imum output performance. The results of the optimal values of shared rates and resource reallocation based on the major
 
adjustment policy are reported in Table 3. Since the optimal values of a1 CS , aCS , aECRC , aECRC , h ; d and DxFC are calculated
2 1 2

by Model (4), these values of individual routes in the major adjustment policy are equal to those of individual routes in
the minor adjustment policy.
In regard to resource reallocation, the results indicate that the crew salaries in the production and service processes,
empty container repositioning costs in the production and service processes, vessel capacity, handling cost and other cost
for the whole company can be respectively reduced by USD $229,820, USD $58,420, USD $1,256,940, USD $10, USD
$510,620,000, USD $8,451,670 and USD $12,190,500, without affecting the production of the maximum desirable output
and the minimum undesirable output. Route J shows no changes in all common and specific inputs. In the production pro-
cess, it is suggested that on eight routes (Routes A, B, D, E, F, G, H and I) and seven routes (Routes A, B, D, G, H, I and K), crew
salaries and empty container repositioning costs be reduced respectively, whereas on five routes (Routes C, K, L, M and N)
and six routes (Routes C, E, F, L, M and N) the decision-maker should increase crew salaries and empty container reposition-
ing costs respectively. In the service process, six routes (Routes B, G, H, L, M and N) and four routes (Routes A, G, H and N)
need decreases in crew salaries and empty container repositioning costs respectively, whereas five routes (Routes B, D, E, F
and K) and seven routes (Routes B, D, E, F, K, L and M) should have crew salaries and empty container repositioning costs
respectively be increased. Eight routes (Routes B, C, D, G, H, I, K and N) need decreases in vessel capacities, whereas four
routes (Routes A, E, F and L) need increases in vessel capacities. For ten routes (Routes A, B, C, D, G, H, I, K, M and N) decreases
in handling costs are recommended, whereas only three routes (Routes E, F and L) should increase handling costs. Finally,
nine routes (Routes A, B, C, D, G, H, I, K and N) require decreases in other costs, whereas four routes (Routes E, F, L and
M) need increases in other costs.
Under the major adjustment policy, this company can also simultaneously increase its revenue and decrease emission
levels, and the proportions are the same as those under the minor adjustment policy. However, the enhancement of perfor-
mance is achieved by simultaneously reducing and transferring excess input resources among routes. Since profit can be fur-
ther increased, this policy provides stronger incentives for the decision-maker of this company to reallocate input resources.
Although the major adjustment policy can raise more profit for this company, it may cause stronger resistance, especially in
terms of manpower. This result leads to the conclusion that although this policy is also feasible in practice, the decision-
maker should not easily implement it without careful consideration. Similarly, the contributions of individual inputs for
the enhancement of performance are different under the major adjustment policy. Besides reallocating input resources,
the decision-maker of this company can reduce crew salary by 46%, empty container repositioning cost by 23%, vessel capac-
ity by 34%, handling cost by 37%, other cost by 35% and fuel cost by 21% among 14 routes. It is worth noting that crew salary
M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671
Table 2
Results of minor adjustment policy.

Route a1
CS a2
CS a1
ECRC a2
ECRC Dx1CS Dx2CS Dx1ECRC Dx2ECRC DxSC DxHC DxOC DxFC
(1000 USD) (1000 USD) (1000 USD) (1000 USD) (TUEs) (1000 USD) (1000 USD) (1000 USD)
A 0.189 (118.15) 0.042 (26.26) 0.118 (658.35) 0.027 (150.64) 90.79 1.48 421.73 54.40 2162.92 1223.06 903.27 507.13
B 0.041 (25.63) 0.009 (5.63) 0.068 (379.39) 0.015 (83.69) 34.67 8.66 30.96 11.71 2502.94 25.67 234.06 467.94
C 0.004 (2.50) 0.001 (0.63) 0.038 (212.01) 0.009 (50.21) 0.71 3.10 45.63 6.84 1584.31 53.44 182.39 188.12
D 0.040 (25.01) 0.009 (5.63) 0.065 (362.65) 0.014 (78.11) 3.90 5.73 106.40 9.95 2132.36 363.46 195.80 452.87
E 0.023 (14.38) 0.005 (3.13) 0.038 (212.01) 0.008 (44.63) 1.61 5.01 90.55 37.63 7063.80 714.54 978.23 18.6.21
F 0.029 (18.13) 0.006 (3.75) 0.045 (251.06) 0.010 (55.79) 58.72 2.20 186.63 24.60 5144.95 1292.93 1664.41 156.08
G 0.118 (73.77) 0.026 (16.25) 0.078 (435.18) 0.017 (94.85) 22.46 7.02 46.84 4.84 6616.12 399.48 658.54 576.01
H 0.190 (118.77) 0.042 (26.26) 0.097 (541.18) 0.021 (117.16) 43.68 8.03 125.66 7.47 806.85 582.37 628.37 507.13
I 0.039 (24.38) 0.010 (6.25) 0.088 (490.97) 0.020 (111.58) 5.99 0 95.28 0 1205.29 537.09 617.06 467.94
J 0.001 (0.63) 0.001 (0.63) 0.046 (256.64) 0.010 (55.79) 1.72 3.21 48.31 13.09 360.17 80.30 199.01 188.12
K 0.015 (9.38) 0.010 (6.25) 0.033 (184.11) 0.014 (78.11) 22.15 0 145.59 0 2444.71 52.88 481.98 452.87
L 0.009 (5.63) 0.006 (3.75) 0.010 (55.79) 0.007 (39.05) 1.00 0 137.11 0 2589.46 490.73 642.13 186.21
M 0.009 (5.63) 0.006 (3.75) 0.015 (83.69) 0.010 (55.79) 3.15 1.17 139.42 11.18 3525.00 187.00 563.25 156.08
N 0.040 (25.01) 0.080 (50.01) 0.026 (145.06) 0.053 (295.70) 4.46 15.49 65.89 117.23 2833.38 217.73 607.24 576.01
Total 0.747 (466.97) 0.253 (158.16) 0.765 (4268.10) 0.235 (1311.11) 0 0 0 0 0 0 0 5068.71
h = 0.269 d = 0.211

Note: Amounts of allocated common inputs are provided in parentheses.

665
666
M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671
Table 3
Results of major adjustment policy.

Route a1
CS a2
CS a1
ECRC a2
ECRC Dx1CS Dx2CS Dx1ECRC Dx2ECRC DxSC DxHC DxOC DxFC
(1000 USD) (1000 USD) (1000 USD) (1000 USD) (TUEs) (1000 USD) (1000 USD) (1000 USD)
A 0.189 (118.15) 0.042 (26.26) 0.118 (658.35) 0.027 (150.64) 96.43 0.48 312.05 0.48 430.62 1216.51 1532.16 507.13
B 0.041 (25.63) 0.009 (5.63) 0.068 (379.39) 0.015 (83.69) 18.72 0.60 293.24 24.40 7103.51 1170.96 1362.65 467.94
C 0.004 (2.50) 0.001 (0.63) 0.038 (212.01) 0.009 (50.21) 13.67 0 16.98 0 3570.00 212.41 473.11 188.12
D 0.040 (25.01) 0.009 (5.63) 0.065 (362.65) 0.014 (78.11) 12.70 0.80 267.03 28.07 8155.22 1133.28 877.49 452.87
E 0.023 (14.38) 0.005 (3.13) 0.038 (212.01) 0.008 (44.63) 0.11 2.83 41.79 55.30 1136.40 23.08 357.94 186.21
F 0.029 (18.13) 0.006 (3.76) 0.045 (251.06) 0.010 (55.79) 0.19 0.46 0.02 9.59 37.34 3.25 6.99 156.08
G 0.118 (73.77) 0.026 (16.25) 0.078 (435.18) 0.017 (94.85) 48.96 10.40 61.49 16.40 5473.59 79.88 1392.01 576.01
H 0.190 (118.77) 0.042 (26.26) 0.097 (541.18) 0.021 (117.16) 110.08 20.26 414.72 18.85 5221.76 1724.72 2173.45 507.13
I 0.039 (24.38) 0.010 (6.25) 0.088 (490.97) 0.020 (111.58) 17.28 0 426.36 0 6259.43 1769.30 2170.52 467.94
J 0.001 (0.63) 0.001 (0.63) 0.046 (256.64) 0.010 (55.79) 0 0 0 0 0 0 0 188.12
K 0.015 (9.38) 0.010 (6.25) 0.033 (184.11) 0.014 (78.11) 1.14 0.20 98.76 28.40 8199.78 971.34 1419.36 452.87
L 0.009 (5.63) 0.006 (3.75) 0.010 (55.79) 0.007 (39.05) 12.60 1.63 195.60 15.62 400.00 144.00 172.00 186.21
M 0.009 (5.63) 0.006 (3.75) 0.015 (83.69) 0.010 (55.79) 12.60 0.23 168.00 3.05 0 51.00 55.00 156.08
N 0.040 (25.01) 0.080 (50.01) 0.026 (145.06) 0.053 (295.70) 34.64 30.31 194.32 128.71 9083.07 292.60 1381.66 576.01
Total 0.747 (465.13) 0.253 (158.16) 0.765 (4268.10) 0.235 (1311.11) 229.82 58.42 1256.94 0.01 51062.00 8451.67 12190.50 5068.71
h = 0.269 d = 0.211

Note: Amounts of allocated common inputs are provided in parentheses.


M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671 667

needs a considerable reduction. However, whether the decision-maker uses pay cuts or layoffs, slashing crew salary by
nearly half may cause strong protest. Hence, the decision-maker of this company should be cautious in the adoption of
the major adjustment policy.

5. Conclusions

This paper develops a non-radial CNDEA model for resource allocation to consider the two-stage production system,
which is not covered by the existing centralized resource allocation models. This paper incorporates the common inputs,
specific inputs, energy inputs, desirable outputs and undesirable outputs into the proposed model. Based on the framework
of the non-radial measure, it can simultaneously account for radial and non-radial slacks. Two cases are presented for cen-
tralized resource allocation. One case considers the situation in which common inputs are shared among sub-units, while the
other case considers the situation in which common inputs are not only shared among sub-units but also allocated between
two processes. The proposed model is illustrated in an empirical example of 14 Asian shipping routes operated by a Tai-
wanese container shipping company for the fourth quarter in 2014. The empirical results show that the aggregate desirable
output of the container shipping company can be expanded, but the aggregate undesirable output can be contracted at cur-
rent given input resource levels. In order to improve the output performance, the container shipping company can adopt
minor and major adjustment policies. Its decision-maker can choose the minor adjustment policy, which transfers inputs
among routes but maintains the original levels of input resources, to avoid organizational resistance. Or the decision-
maker can choose the major adjustment policy, which reduces the total amount of input resources, to further cut cost.
However, this paper has some limitations. First, the non-adjustable input variables are ignored in this paper. In the actual
production environment, some allocation of input resources may be restricted. Second, this paper assumes that the alloca-
tion cost is zero. However, the allocation of resources may be costly among sub-units in practice. Future research can take
the above-mentioned limitations into consideration.

Acknowledgments

The research is supported by the Ministry of Science and Technology of the Republic of China under project number MOST
103-2410-H-019-011.

Appendix A

Let k1jk ai1 J l1i1 jk , k1jk l1i1 jk m1i1 jk , m1i1 jk 1  ai1 J k1jk , where i1 1; 2; . . . ; I1 , j 1; . . . ; N, k 1; . . . ; N. Since k1jk l1i1 jk m1i1 jk ,
8i1 ; j; k, this paper can arbitrarily select each of i01 1; 2; . . . ; I1 and impose l1i0 jk m1i0 jk l1i1 jk m1i1 jk (i1 2; . . . ; I1 , j 1; . . . ; N,
1 1

k 1; . . . ; N) on Phase I. Without loss of generality, this paper imposes k1jk l1jk m1jk l1i1 jk m1i1 jk , 8i1 ; j; k to preserve the lin-
earity and convexity in the DEA model. Then model (1) can be transformed into the following linear programming model
(A.1)
PR PQ
r1 hr q1 dq
Max A:1
R Q
Subject to
a. Common input constraint:
X
N
l1i1 jk xi1 6 ai1 k xi1 ; i1 1; 2; . . . ; I1 ; k 1; . . . ; N A:1:1
j1

b. Specific input constraint:


X
N
l1jk m1jk xi2 J 6 xi2 k ; i2 1; 2; . . . ; I2 ; k 1; . . . ; N A:1:2
j1

c. Energy input constraint:


X
N
l1jk m1jk xi3 J 6 xi3 k ; i3 1; 2; . . . ; I3 ; k 1; . . . ; N A:1:3
j1

d. Intermediate product constraints:


X
N
l1jk m1jk zpj P zpk ; p 1; 2; . . . ; P; k 1; . . . ; N A:1:4:1
j1
668 M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671

X
N
k2jk zpj 6 zpk ; p 1; 2; . . . ; P; k 1; . . . ; N A:1:4:2
j1

Combining Eqs. (A.1.4.1) and (A.1.4.2), the intermediate product constraints can be shown as follows:
X
N X
N
l1jk m1jk zpj P k2jk zpj ; p 1; 2; . . . ; P; k 1; . . . ; N A:1:4
j1 j1

e. Desirable output constraints:


X
N X
N X
N
k2jk yrj P 1 hr yrk ; r 1; 2; . . . ; R A:1:5
k1 j1 k1

X
N
k2jk yrj P yrk ; r 1; 2; . . . ; R; k 1; . . . ; N A:1:6
j1

f. Undesirable output constraints:


X
N X
N X
N
k2jk bqj 1  dq bqk ; q 1; 2; . . . ; Q A:1:7
k1 j1 k1

X
N
k2jk bqj 1  dq bqk ; q 1; 2; . . . ; Q ; k 1; . . . ; N A:1:8
j1

g. Variable return-to-scale constraints:


X
N
l1jk m1jk 1; k 1; . . . ; N A:1:9
j1

X
N
k2jk 1; k 1; . . . ; N A:1:10
j1

h. Shared rate constraints:


^ _
a 6 ai1 J 6 ai1 J ; i1 1; 2; . . . ; I1 ; j 1; . . . ; N
i1 J A:1:11

X
N
ai1 J 1; i1 1; 2; . . . ; I1 A:1:12
j1

l1jk ; m1jk ; k2jk P 0; j 1; . . . ; N; k 1; . . . ; N A:1:13

Since l1jk m1jk ai1 J l1i1 jk , multiplying ljk + mjk on Eq. (A.1.11) to ensure the range of l1i1 jk is related to ai1 J , this paper thus
has
^ _
ai1 J ljk mjk 6 li1 jk 6 ai1 J ljk mjk ; i1 1; 2; . . . ; I1 ; j 1; . . . ; N; k 1; . . . ; N A:1:11:1

Furthermore, this paper imposes Eq. (A.1.11.1) on Model (A.1).

Appendix B

Let k1jk a1i1 J l1i1 jk , k2jk a2i1 J l2i1 jk , k1jk l1i1 jk m1i1 jk , k2jk l2i1 jk m2i1 jk , m1i1 jk 1  ai1 J k1jk , and m2i1 jk 1  a2i1 J k2jk , where
i1 1; 2; . . . ; I1 , j 1; . . . ; N, k 1; . . . ; N. Since k1jk l 1
i1 jk m 1
i1 jk , 8i1 ; j; k and k2jk l2i1 jk m2i1 jk , 8i1 ; j; k, this paper imposes
k1jk l1jk m1jk , 8i1 ; j; k and k2jk l2jk m2jk , 8i1 ; j; k. Then Model (4) can be transformed into the following linear programming
model (B.1).
PR PQ
r1 hr q1 dq
Max B:1
R Q
Subject to
a. Common input constraint in the production process:
X
N
l1i1 jk xi1 6 a1i1 k xi1 ; i1 1; 2; . . . ; I1 ; k 1; . . . ; N B:1:1
j1
M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671 669

b. Specific input constraint:


X
N
l1jk m1jk xi2 J 6 xi2 k ; i2 1; 2; . . . ; I2 ; k 1; . . . ; N B:1:2
j1

c. Energy input constraint:


X
N
l1jk m1jk xi3 J 6 xi3 k ; i3 1; 2; . . . ; I3 ; k 1; . . . ; N B:1:3
j1

d. Intermediate product constraint:


X
N X
N
l1jk m1jk zpj P l2jk m2jk zpj ; p 1; 2; . . . ; P; k 1; . . . ; N B:1:4
j1 j1

e. Common input constraint in the service process:


X
N
l2i1 jk xi1 6 a2i1 k xi1 ; i1 1; 2; . . . ; I1 ; k 1; . . . ; N B:1:5
j1

f. Desirable output constraints:


X
N X
N X
N
l2jk m2jk yrj P 1 hr yrk ; r 1; 2; . . . ; R B:1:6
k1 j1 k1

X
N
l2jk m2jk yrj P yrk ; r 1; 2; . . . ; R; k 1; . . . ; N B:1:7
j1

g. Undesirable output constraints:


X
N X
N X
N
l2jk m2jk bqj 1  dq bqk ; q 1; 2; . . . ; Q B:1:8
k1 j1 k1

X
N
l2jk m2jk bqj 1  dq bqk ; q 1; 2; . . . ; Q ; k 1; . . . ; N B:1:9
j1

h. Variable return-to-scale constraints:


X
N
l1jk m1jk 1; k 1; . . . ; N B:1:10
j1

X
N
l2jk m2jk 1; k 1; . . . ; N B:1:11
j1

i. Shared rate constraints:


^ _
a 6 a1i1 J 6 a1i1 J ; i1 1; 2; . . . ; I1 ; j 1; . . . ; N
1
i1 J B:1:12

^ _
a 6 a2i1 J 6 a2i1 J ; i1 1; 2; . . . ; I1 ; j 1; . . . ; N
2
i1 J B:1:13

X
N
a1i1 J w1i1 ; i1 1; 2; . . . ; I1 B:1:14
j1

X
N
a2i1 J w2i1 ; i1 1; 2; . . . ; I1 B:1:15
j1

w1i1 w2i1 1; i1 1; 2; . . . ; I1 B:1:16

l1jk ; m1jk ; l2jk ; m2jk P 0; j 1; . . . ; N; k 1; . . . ; N B:1:17

With multiplication of l1jk m1jk on Eq. (B.1.12) and l2jk m2jk on Eq. (B.1.13), it becomes
^ _
i1 J l1jk m1jk 6 l1i1 jk 6 a1i1 J l1jk m1jk ; i1 1; 2; . . . ; I1 ; j 1; . . . ; N; k 1; . . . ; N B:1:12:1
1
a
670 M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671

^ _
i1 J l2jk m2jk 6 l2i1 jk 6 a2i1 J l2jk m2jk ; i1 1; 2; . . . ; I1 ; j 1; . . . ; N; k 1; . . . ; N B:1:13:1
2
a
Replacing Eq. (B.1.12) by Eq. (B.1.12.1) and Eq. (B.1.13) by Eq. (B.1.13.1), Model (B.1) becomes a linear programming model.

References

Agarwal, R., Ergun, ., 2008. Ship scheduling and network design for cargo routing in liner shipping. Transp. Sci. 42 (2), 175196.
Ahmed, N.U., 1983. An analytical decision model for resource allocation in highway maintenance management. Transp. Res. Part A 17A (2), 133138.
Ali, A.I., Seiford, L.M., 1990. Translation invariance in data envelopment analysis. Oper. Res. Lett. 9 (6), 403405.
Argris, C., Schn, D., 1978. Organizational Learning: A Theory of Action Perspective. Addison Wesley, Reading, MA.
Asgari, N., Farahani, R.Z., Goh, M., 2013. Network design approach for hub ports-shipping companies competition and cooperation. Transp. Res. Part A 48, 1
18.
Asmild, M., Paradi, J.C., Pastor, J.T., 2009. Centralized resource allocation BCC models. Omega 37, 4049.
Bang, H.S., Kang, H.W., Martin, J., Woo, S.H., 2012. The impact of operational and strategic management on liner shipping efficiency: a two-stage DEA
approach. Maritime Policy Manage. 39 (7), 653672.
Barnhart, C., Lohatepanont, M., 2004. Airline schedule planning: integrated models and algorithms for schedule design and fleet assignment. Transp. Sci. 38
(1), 1932.
Bektas, T., Crainic, T.G., Morency, V., 2009. Improving the performance of rail yards through dynamic reassignments of empty cars. Transp. Res. Part C 17,
259273.
Bittner, J., Baird, T., Adams, T., 2012. Impacts of Panama Canal expansion on US greenhouse gas emissions. Transp. Res. Rec. 2273, 3844.
Boyles, S.D., 2015. Equity and network-level maintenance scheduling. EURO J. Transp. Logist. 4, 175193.
Chamders, R., Chung, Y., Fre, R., 1996. Benefit and distance functions. J. Econ. Theory 70, 407419.
Chang, S.M., Wang, J.S., Yu, M.M., Shang, K.C., Lin, S.H., Hsiao, B., 2015. An application of centralized data envelopment analysis in resource allocation in
container terminal operations. Maritime Policy Manage. 42 (8), 776788.
Charnes, A., Cooper, W.W., Rhodes, E., 1978. Measuring efficiency of decision making units. Eur. J. Oper. Res. 2, 429444.
Cheung, R.J., Chen, C.Y., 1998. A two-stage stochastic network model and solution methods for the dynamic empty container allocation problem. Transp. Sci.
32 (2), 142162.
Chou, C.C., Gou, R.H., Tsai, C.L., Tsou, M.C., Wong, C.P., Yu, H.L., 2010. Application of a mixed fuzzy decision making and optimization programming model to
the empty container allocation. Appl. Soft Comput. 10 (4), 10711079.
Chung, Y.H., Fre, R., Grosskopf, S., 1997. Productivity and undesirable outputs: a directional distance function approach. J. Environ. Manage. 51 (3), 229
240.
Crainic, T.G., Gendreau, M., Dejax, P., 1993. Dynamic and stochastic models for the allocation of empty containers. Oper. Res. 41 (5), 102126.
Fang, L., 2013. A generalized DEA model for centralized resource allocation. Eur. J. Oper. Res. 228, 405412.
Fang, L., 2015. Centralized resource allocation based on efficiency analysis for step-by-step improvement paths. Omega 51, 2428.
Fang, L., Li, H.C., 2015. Centralized resource allocation based on the cost-revenue analysis. Comput. Ind. Eng. 85, 395401.
Fre, R., Grosskopf, S., Lovell, C.A.K., 1985. The Measurement of Efficiency of Production. Kluwer Academic Publishers, Boston.
Fre, R., Grosskopf, S., Lovell, C.A.K., 1989. Multilateral productivity comparisons when some outputs are undesirable: a nonparametric approach. Rev. Econ.
Stat. 71 (1), 9098.
Fre, R., Grosskopf, S., Lovell, C.A.K., 1994. Production Frontiers. Cambridge University Press, Cambridge.
Fre, R., Grosskopf, S., Lovell, C.A.K., Yaiswarng, S., 1993. Deviation of shadow prices for undesirable outputs: a distance function approach. Rev. Econ. Stat.
75 (2), 374380.
Fre, R., Grosskopf, S., Hernandez-Sancho, F., 2004. Environmental performance: an index number approach. Resour. Energy Econ. 26 (4), 343352.
Fre, R., Lovell, C.A.K., 1978. Measuring the technical efficiency of production. J. Econ. Theory 19 (1), 150162.
Fielding, G.J., 1987. Managing Public Transit Strategically. Jossey-Bass, San Francisco.
Fukuyama, H., Weber, W.L., 2009. A directional slacks-based measure of technical inefficiency. Socio Econ. Plan. Sci. 43 (4), 274287.
Fukuyama, H., Weber, W.L., 2010. A slacks-based inefficiency measure for a two-stage system with bad outputs. Omega 38 (4), 398409.
Gelareh, S., Nickel, S., Pisinger, D., 2010. Liner shipping hub network design in a competitive environment. Transp. Res. Part E 46 (6), 9911004.
Golany, B., Roll, Y., 1989. An application procedure for DEA. Omega 17 (3), 237250.
Golembiewski, R., Billingsley, K., Yeager, S., 1976. Measuring change and persistence in human affairs: types of change generated by OD designs. J. Appl.
Behav. Sci. 12 (2), 133157.
Guo, X.D., Zhu, L., Fan, Y., Xie, B.C., 2011. Evaluation of potential reductions in carbon emissions in Chinese provinces based on environmental DEA. Energy
Policy 39, 23522360.
Hailu, A., Veeman, T., 2001. Non-parametric productivity analysis with undesirable outputs: an application to Canadian pulp and paper industry. Am. J.
Agric. Econ. 83 (3), 605616.
Hane, C.A., Barnhart, C., Johnson, E.L., Marsten, R.E., Nemhauser, G.L., Sigismondi, G., 1995. Fleet assignment problem: solving a large-scale integer program.
Math. Program. 70 (13), 211232.
Holmberg, K., Joborn, M., Lundgren, J.T., 1998. Improved empty freight car distribution. Transp. Sci. 32, 163173.
Hoorn, S.V., Knapp, S., 2015. A multi-layered risk exposure assessment approach for the shipping industry. Transp. Res. Part A 78, 2133.
Hosseinzadeh Lotfi, F., Noora, A.A., Jahanshahloo, G.R., Gerami, J., Mozaffari, M.R., 2010. Centralized resource allocation for enhanced Russell models. J.
Comput. Appl. Math. 235, 110.
Hosseinzadeh Lotfi, F., Nematollahi, N., Behzadi, M.H., Mirbolouki, M., Moghaddas, Z., 2012. Centralized resource allocation with stochastic data. J. Comput.
Appl. Math. 236, 17831788.
Hua, Z.S., Bian, Y.W., Liang, L., 2007. Eco-efficiency analysis of paper mills along the Huai River: an extended DEA approach. Omega 35 (5), 578587.
Kao, C., Hwang, S.N., 2008. Efficiency decomposition in two-stage data envelopment analysis: an application to non-life insurance companies in Taiwan. Eur.
J. Oper. Res. 185 (1), 418429.
Kao, C., Hwang, S.N., 2010. Efficiency measurement for network systems: IT impact on firm performance. Decis. Support Syst. 48, 437446.
Kelareva, E., Tierney, K., Kilby, P., 2014. CP methods for scheduling and routing with time-dependent task costs. EURO J. Comput. Optim. 2 (3), 147194.
Korhonen, P., Syrjnen, M., 2004. Resource allocation based on efficiency analysis. Manage. Sci. 50 (8), 11341144.
Lam, S.W., Lee, L.H., Tang, L.C., 2007. An approximate dynamic programming approach for the empty container allocation problem. Transp. Res. Part C 15 (4),
265277.
Lau, Y.Y., Ng, A.K.Y., Fu, X., Li, K.X., 2013. Evolution and research trends of container shipping. Maritime Policy Manage. 40 (7), 654674.
Li, Y., Yang, F., Liang, L., Hua, Z., 2009. Allocating the fixed cost as a complement of other cost inputs: a DEA approach. Eur. J. Oper. Res. 197, 389401.
Liou, J.J.H., Tzeng, G.H., 2009. Airline maintenance manpower optimization from the De Novo perspective. Commun. Comput. Inf. Sci. 35, 806814.
Lozano, S., Gutirrez, E., 2011. Slack-based measure of efficiency of airports with airplanes delays as undesirable outputs. Comput. Oper. Res. 38, 131139.
Lozano, S., Villa, G., 2004. Centralized resource allocation using data envelopment analysis. J. Prod. Anal. 22, 143161.
Lozano, S., Villa, G., 2005. Centralized DEA models with the possibility of downsizing. J. Oper. Res. Soc. 56, 357364.
Lozano, S., Villa, G., Adenso-daz, B., 2004. Centralized target setting for regional recycling operations using DEA. Omega 32, 101110.
Lozano, S., Villa, G., Brnnlund, R., 2009. Centralised reallocation of emission permits using DEA. Eur. J. Oper. Res. 193, 752760.
M.-M. Yu, L.-H. Chen / Transportation Research Part A 94 (2016) 650671 671

Lozano, S., Villa, G., Canca, D., 2011. Application of centralized DEA approach to capital budgeting in Spanish ports. Comput. Ind. Eng. 60, 455465.
Lun, V.Y.H., Marlow, P., 2011. The impact of capacity on firm performance: a study of the liner shipping industry. Int. J. Ship. Transp. Logist. 3 (1), 5771.
Mar-Molinero, C., Prior, D., Segovia, M.M., Portillo, F., 2014. On centralized resource utilization and its reallocation by using DEA. Ann. Oper. Res. 221 (1),
273283.
Mathew, T.V., Khasnabis, S., Mishra, S., 2010. Optimal resource allocation among transit agencies for fleet management. Transp. Res. Part A 44 (6), 418432.
Melachrinoudis, E., Kozanidis, G., 2002. A mixed integer knapsack model for allocating funds to highway safety improvements. Transp. Res. Part A 36 (9),
789803.
Mishra, S., 2013. A synchronized model for crash prediction and resource allocation to prioritize highway safety improvement projects. Procedis-Soc. Behav.
Sci. 104, 9921001.
Mishra, S., Golias, M.M., Sharma, S., Boyles, S.D., 2015. Optimal funding allocation strategies for safety improvements on urban intersections. Transp. Res.
Part A 75, 113133.
Mishra, S., Khasnabis, S., 2012. Optimization model for allocating resources for highway safety improvement at urban intersections. J. Transp. Eng. 138 (5),
535547.
Mishra, S., Mathew, T.V., Khasnabis, S., 2010. Single stage integer programming model for long term transit fleet resource allocation. J Transp. Eng.-ASCE 136
(4), 281290.
Mishra, S., Sharma, S., Khasnabis, S., Mathew, T.V., 2013. Preserving an aging transit fleet: an optimal resource allocation perspective based on service life
and constrained budget. Transp. Res. Part A 47, 111123.
Murillo-Hoyos, J., Athigakunagorn, N., Labi, S., 2015. Methodology for safety improvement programming using constrained network-level optimization.
Transp. Res. Part C 50, 106116.
Pastor, J.T., Ruiz, J.L., Sirvent, I., 1999. An enhanced DEA Russell graph efficiency measure. Eur. J. Oper. Res. 115, 596607.
Pita, J.P., Barnhart, C., Antunes, A., 2012. Integrated flight scheduling and fleet assignment under airport congestion. Transp. Sci. 47 (4), 477492.
Rexing, B., Barnhart, C., Kniker, T., Jarrah, A., Krishhnaumurthy, A., 2000. Airline fleet assignment with time windows. Transp. Sci. 34 (1), 120.
Scheel, H., 2001. Undesirable outputs in efficiency valuations. Eur. J. Oper. Res. 132 (2), 400410.
Seiford, L.M., Zhu, J., 2002. Modeling undesirable factors in efficiency evaluation. Eur. J. Oper. Res. 142 (1), 1620.
Shintani, K., Imai, A., Nishimura, E., Papadimitriou, S., 2007. The container shipping network design problem with empty container repositioning. Transp.
Res. Part E 43, 3959.
Tone, K., 2001. Slack based measure of efficiency in data envelopment analysis. Eur. J. Oper. Res. 130, 498509.
Tone, K., 2004. Dealing with undesirable outputs in DEA: a slacks-based measure (SBM) approach. Oper. Res. Soc. Jpn., 4445
Wang, S., Meng, Q., 2012a. Liner ship fleet deployment with container transshipment operations. Transp. Res. Part E 48 (2), 470484.
Wang, S., Meng, Q., 2012b. Sailing speed optimization for container ships in a liner shipping network. Transp. Res. Part E 48 (3), 701714.
Wang, S., Meng, Q., Sun, Z., 2013. Container routing in liner shipping. Transp. Res. Part E 49, 17.
White, W.W., Bomberault, A.M., 1969. A network algorithm for empty freight car allocation. IBM Syst. J. 8 (2), 147169.
Yang, C.S., Lu, C.S., Haider, J.J., Marlow, P.B., 2013. The effect of green supply chain management on green performance and firm competitiveness in the
context of container shipping in Taiwan. Transp. Res. Part E 55, 5573.
Yu, M.M., Chern, C.C., Hsiao, B., 2013. Human resource rightsizing using centralized data envelopment analysis: evidence from Taiwans airports. Omega 41
(1), 119130.
Zhou, P., Ang, B.W., Poh, K.L., 2006. Slacks-based efficiency measures for modeling environmental performance. Energy Econ. 60 (1), 111118.
Zhou, P., Poh, K.L., Ang, B.W., 2007. A non-radial DEA approach to measuring environmental performance. Eur. J. Oper. Res. 178 (1), 19.

Anda mungkin juga menyukai