Computers & Industrial Engineering 101 (2016) 81–90

Contents lists available at ScienceDirect

Computers & Industrial Engineering

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

Resource allocation based on context-dependent data envelopment

analysis and a multi-objective linear programming approach
Jie Wu a, Qingyuan Zhu a, Qingxian An b,c,⇑, Junfei Chu a, Xiang Ji a
a
School of Management, University of Science and Technology of China, Hefei, Anhui Province 230026, PR China
b
School of Business, Central South University, Changsha, Hunan Province 410083, PR China
c
Industrial Systems Optimization Laboratory, Charles Delaunay Institute and UMR CNRS 6281, University of Technology of Troyes, Troyes 10004, France

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

Article history: This paper discusses a mechanism for the allocation of resources among a set of decision making units
Received 8 December 2014 (DMUs) which are managed by a centralized control unit in an organization. Data envelopment analysis
Received in revised form 23 July 2016 (DEA) and multi-objective linear programming (MOLP) are integrated to deal with this resource alloca-
Accepted 28 August 2016
tion problem. Also, context-dependent DEA is introduced to identify the changed production possibility
Available online 29 August 2016
set after resource allocation, which determines the production plans that are feasible with input increase
or decrease in general. For the centralized unit, the MOLP approach is proposed to simultaneously max-
Keywords:
imize total output and effectiveness while minimizing the total allocated variable input consumption.
Resource allocation
Data envelopment analysis
Among these objectives, the effectiveness is determined by the output growth rate for all DMUs, which
Multi-objective linear programming can reflect the effects obtained by allocating the input resources that are not used up, such as new equip-
Most productive scale size ment. In addition, we restrict the production of limited resources to the new most productive scale size
(MPSS) region where the DMUs have the best economic characteristics. Finally, an example is employed
to illustrate the approach.
Ó 2016 Published by Elsevier Ltd.

1. Introduction remain unchangeable, examples being housing, land, and work-

Resource allocation is a problem that commonly appears in
organizations with a centralized decision-making environment in
which a set of units are operating under a central unit with power
to control some decision parameters like resources. Resource allo-
cation is an important issue which can greatly affect the develop-
ment of corporations (Amirteimoori & Shafiei, 2006; Wu, An, Ali,
& Liang, 2013). Commonly, resource allocation is a crucial tool
for improving performance (Golany, 1988; Golany and Tamir,
1995). For example, for some car and refrigerator factories, the cen-
tral decision making unit (CDMU) may have purchased several
items of new, large-scale equipment which need to be allocated
to its branch factories. With the new equipment allocated, some
other more flexible input resources like car/refrigerator parts need
to be arranged to suit the new equipment. In many enterprises, the
flexible input resources are more easily available than other kinds
of resources. In contrast, some inflexible input resources may
⇑ Corresponding author at: School of Business, Central South University,
Changsha, Hunan Province 410083, PR China.
E-mail addresses: jacky012@mail.ustc.edu.cn (J. Wu), zqyustc@mail.ustc.edu.cn
(Q. Zhu), anqingxian@163.com (Q. An), cjf0731@mail.ustc.eudu.cn (J. Chu), signji
@mail.ustx.edu.cn (X. Ji).
shops. The CDMU is assumed to be interested in obtaining the most
beneficial results by allocating these available resources (large
scale equipment, spare parts, and so on) to its individual branch
factories. In real life, some allocated resources like large scale
equipment are always a limited resource, so how to allocate them
plays a significant role in the development of a company. However,
how to allocate these different kinds of resources reasonably and
effectively remains an open issue.
The problem of resource allocation has become one of the clas-
sical applications of management science and has great practical
application value (Korhonen & Syrjänen, 2004). Data envelopment
analysis (DEA) brought a new perspective on the resource alloca-
tion problem. Developed by Charnes, Cooper, and Rhodes (1978),
DEA is a non-parametric mathematical approach which is used to
evaluate the relative performance of a group of homogenous
DMUs, especially with multiple inputs and multiple outputs (Wu,
Xiong, An, Sun, & Wu, 2015; An, Yan, Wu, & Liang, 2016). It does
not require the assumption of functional form in the production
function for measuring the efficiency. For an introduction to DEA
basic theory and applications, refer to Cooper, Seiford, and Tone
(2007), or Cook and Seiford (2009). The existing DEA studies on
the resource allocation problem can be divided into different cate-
gories. One is resource allocation and target setting (Bi, Ding, Luo, &

http://dx.doi.org/10.1016/j.cie.2016.08.025
0360-8352/Ó 2016 Published by Elsevier Ltd.
82 J. Wu et al. / Computers & Industrial Engineering 101 (2016) 81–90
Liang, 2011). The second category is centralized resource allocation
(Lozano & Villa, 2004). The third category encompasses other per-
spectives (Korhonen & Syrjänen, 2004; Du, Cook, Liang, & Zhu,
2014).
There has been much research on resource allocation and target
setting during the past few years. In order to identify the trade-offs
between overall output and equity from alternative allocations of
service resources among different public service delivery sites,
Mandell (1991) proposed two related bi-criteria mathematical pro-
gramming models. To handle the problem of allocating central
grants to Greek local authorities, Athanassopoulos (1995) inte-
grated resource allocation and target setting in multilevel planning
problems based on goal programming and data envelopment anal-
ysis. In order to jointly determine input resources for each DMU,
Golany, Phillips, and Rousseau (1993) proposed a DEA-based linear
model to allocate the available resources. Bi et al. (2011) proposed
a DEA-based methodology for resource allocation and target set-
ting considering a parallel production system. Hosseinzadeh Lotfi
et al. (2013) introduced an allocation mechanism that is based
on a common dual weights approach.
Some other researches focus on centralized resource allocation.
Lozano and Villa (2004) made the first attempt and presented
two new BCC-DEA models for centralized resource allocation
(CRA-BCC). One type of model seeks radial reductions of the total
consumption of every input while the other type seeks separate
reductions for each input according to a preference structure
(Cecilio & Diego, 2006; Lozano, Villa, & Canca, 2011). Asmild,
Paradi, and Pastor (2009) suggested modifying one of the central-
ized models to only consider adjustment of previously inefficient
units. Fang (2013) introduced a new, generalized, centralized
resource allocation model which extends Lozano and Villa’s
(2004) and Asmild et al.’s models (2009) to a more general case.
Fang and Li (2015) proposed a centralized approach for reallocating
resources based on an extended revenue efficiency model. Hatami-
Marbini, Tavana, Agrell, Lotfi, and Beigi (2015) proposed an alterna-
tive common-weights DEA model to determine the amount of input
and output reduction needed for each DMU to increase score of all
the DMUs. An, Wen, Xiong, Min, and Chen (2016b) built a DEA
model for allocating the carbon dioxide emission permits so as to
control the amount of carbon dioxide emission.
In addition, there are some other perspectives on the resource
allocation problem. Beasley (2003) established a non-linear alter-
native DEA-based resource allocation model aiming to maximize
the average unit efficiency. His model is set up on the basis of
non-linear formulation, which may lead to local optima instead
of a global one. Korhonen and Syrjänen (2004) developed an inter-
active formal approach based on DEA and multiple-objective linear
programming (MOLP) to find the most preferred allocation plan
that maximizes the values of multiple output variables simultane-
ously. Their approach is based on the assumption that units are
able to modify their production in the current production possibil-
ity set formed by some efficient DMUs. They also assume that the
units can modify their production plans using constant return to
scale, and that the efficiency remains unchanged. Recently, Wu
et al. (2013) extended the previous studies by considering both
the economic and environmental factors for the resource allocation
problem. Du et al. (2014) used the cross-efficiency concept in DEA
to approach resource allocation problems.
From the published literature, we know that the existing DEA
models for resource allocation are mostly based on one of two
assumptions. One is that the efficiency for each DMU may be dif-
ferent after resource allocation. The other is that the efficiency
for each DMU is constant regardless of resource allocation. The
main limitations of these two assumptions about resource alloca-
tion are that: (I) Some models based on the assumption of constant
efficiency have changed the production possibility set (PPS) formed
by all the DMUs. Although there are DEA-based models that do not
change the PPS based on the assumption of constant efficiency,
such models may be unreasonable, since for most production sys-
tems the efficiency always changes with a changed production
scale. Therefore, considering efficiency changes after resource allo-
cation is necessary. (II) Most of the research based on the assump-
tion of changeable efficiency after resource allocation assumes that
each unit can have its production on the efficient frontier formed
by some efficient DMUs. They do not consider the units’ actual pro-
duction ability, and they give output targets that may not be
achieved easily for some units.
In addition, few studies have considered the characteristics of
input resources. In real life, there are different kinds of input
resources. For example, some resources may be inflexible and have
fixed characteristics, while some other resources may be flexible
and have variable characteristics. Some resources may be scarce
and some may be relatively abundant. Besides this, there may be
some new additional resources which need to be allocated when
some resources are used but are superfluous for other resources.
In the above example of the car and refrigerator factories, the large
scale equipment and spare parts belong to the category of addi-
tional allocated resource, and the housing, land, and workshops
can be seen as input resources whose allocation cannot change.
The large-scale equipment, housing, land, and workshops may be
classified as having invariable characteristics and the spare parts
may have variable characteristics. This shows that classifying the
input resources is very important.
The focus of this paper is to design a reasonable and effective
resource allocation mechanism, which can bring the greatest ben-
efits for the central organization. We apply and extend DEA in
order to design the resource allocation mechanism. In addition,
the multi-objective linear programming (MOLP) approach is pro-
posed to comprehensively optimize the CDMU’s multiple
objectives.
In our resource allocation problem, the central unit has addi-
tional resources and needs to allocate these resources to the DMUs
based on their inputs and outputs in the current period. Compared
with many other methods for doing so, our technique has the fol-
lowing advantages.
(I) We first classify the input resources into three groups. One
group is the non-allocated constant inputs, another is allo-
cated constant inputs, and the last group is allocated vari-
able inputs. The characteristics of each kind of input
resource are as follows. The non-allocated constant inputs
are those that are unchangeable for each DMU, so they are
neither used up nor allocated; they remain the same in the
next production period. Land is an example. An allocated
constant input for a DMU is one that is not used up by pro-
duction so it never decreases, but it may increase in the next
production period if the central control unit allocates more
to the DMU from a fixed available amount (e.g. recently
obtained and not yet assigned to a production unit). Large-
scale equipment is an example of an allocated constant
input. The third group, allocated variable inputs, consists
of inputs that are used up in production in the current period
but for which there is a ready supply, such as the car and
refrigerator parts in the above example. For each DMU, the
amount of allocated variable inputs can always be increased
or decreased. Put simply, non-allocated inputs never change,
allocated constant inputs never decrease but may increase,
and allocated variable inputs may increase or decrease. This
classification conforms to the situation in many enterprises.
(II) A second advantage of our method concerns the objectives
for resource allocation. The CDMU may want to consider
not only the total output and input consumption in next
 J. Wu et al. / Computers & Industrial Engineering 101 (2016) 81–90
period, but also the effectiveness, which is defined as the
output growth rate for all DMUs and which can reflect the
effects brought by allocating the additional input resources.
To take this into account, we construct a DEA-based MOLP
model which maximizes total output and effectiveness,
while simultaneously minimizing total allocated variable
input consumption in the entire organization.
(III) To take into account changes in production after resource
allocation, context-dependent DEA is introduced in our
model to identify each DMU’s original technology. Whether
the input increases or decreases, the DMUs can only carry on
production using their original technology. The context-
dependent DEA technique has been successfully applied by
Wu, Zhu, and Liang (2016) to allocate the CO2 emissions
and energy intensity reduction over provincial industrial
sectors in China.
(IV) Lastly, considering that the allocated constant inputs are
limited, we apply the new most productive scale size (MPSS)
according to the definition of the new MPSS concept in the
work of Zhu and Shen (1995). The production resulting from
allocation of this kind of resource is restricted to the new
MPSS region.
The structure of this paper is as follows. The preliminaries are
given in the Section 2. In Section 3, the proposed model integrating
context-dependent DEA and MOLP approach is formulated. In Sec-
tion 4, a real world example is applied to illustrate our proposed
approach. Finally conclusions are shown in Section 5.
2. Preliminaries
2.1. Production possibility set and MPSS in DEA
Consider n DMUs, where each DMU consumes varying amounts
of m different inputs to produce s different outputs. Specifically,
DMUj consumes amount xij of input i and produces amount yrj of
output r. Define xj ¼ ðx1j ; x1j ; . . . ; xmj ÞT and yj ¼ ðy1j ; y1j ; . . . ; ysj ÞT as
the input and output vectors of DMUj. The production possibility
set (PPS) T can be described as
T ¼ fðx; yÞjy can be produced from xg
Some regularity assumptions are usually imposed on T. Charnes
et al. (1978) deduced the following PPS, using the constant returns
to scale assumption.
( )
X
n X
n are those which are considered invariable. That is, these input
T CCR ¼ ðx; yÞj kj xij 6 x; kj yrj P y; kj P 0
j¼1 j¼1
Later, Banker, Charnes, and Cooper (1984) deleted the ray
unboundedness postulate from the CCR postulates and deduced
the following production possibility set (BCC). This set is defined
on that basis of the variable returns to scale (VRS) assumption
about the production technology. The VRS assumption has been
widely used in forming the PPS.
( )
X
n X
n X
n
T BCC ¼ ðx; yÞj kj xij 6 x; kj yrj P y; kj ¼ 1; kj P 0
j¼1 j¼1 j¼1
Based on VRS, non-increasing returns to scale (NIRS) and non-
decreasing returns to scale (NDRS) models were presented by
Färe, Grosskopf, and Lovell (1985) and Seiford and Thrall (1990).
These are known as the FGL and ST approaches respectively.
The efficiency of a specific DMU0 can be evaluated by different
DEA models. Under different assumptions about the production
possibility set T we have different models. BCC models have been
83
widely used in past years. Hence, the BCC model as an example
is given below:
!
X
m X
s
Max h þ e si þ sþr
i¼1 r¼1
X
n
s:t: kj xij þ si ¼ xi0 8i
j¼1
X
n ð3Þ
kj yrj  sþr ¼ hyr0 8r
j¼1
X
n
kj ¼ 1
j¼1
kj P 0 8j
where the s þ
i and sr are slack variables for inputs and outputs. Here,
e > 0 is a so-called non-Archimedean element defined to be smaller
than any positive real number (Amin & Toloo, 2004; Cooper et al.,
2007). The DMU under evaluation is efficient if and only if (1)
h ¼ 1 and (2) all slacks s þ
i ¼ sr ¼ 0.
In economics, it is very important to analyze the economies of
scale, especially the most productive scale size (MPSS). MPSS in
DEA was first introduced by Banker (1984) and denotes DMUs in
the production possibility set whose outputs are the biggest mul-
tiples of inputs. The DMUs in this area are all at their optimal size.
Therefore, for some scarce and costly input resources such as large-
scale equipment, changes should keep DMUs within the MPSS set
so as to bring the optimal marginal outputs. It should be noted that
Zhu and Shen (1995) have indicated that the traditional MPSS con-
cept can only be used when a set of efficient DMUs exhibits linear
dependency, and they provided a simple remedy for the MPSS con-
cept in situations with nonlinear dependency. The new MPSS con-
cept was proposed on basis of BCC and FGL models. In this paper,
we will apply this new MPSS concept to set the DMUs’ targets.
2.2. Resource allocation problem
Assume an organization contains n homogeneous units. Each
unit consumes multiple input resources to produce multiple out-
puts. In the next period, the organization may have some addi-
tional input resources which need to be allocated to these units.
Firstly, for generalization, we discriminate the total input resources
for production into two classes: one is the non-allocated inputs G,
and the other is the allocated inputs X. The non-allocated inputs G
ð1Þ resources are necessary for production but not used up, so they
remain unchanged in next period. Land and workshops are exam-
ples of non-allocated inputs. Here, we note that the degradation of
this type of resource through use in two adjacent production peri-
ods is negligible. Now consider the allocated inputs X. We further
classify them into two parts: one is constant inputs D, and the
other is variable inputs F. The constant inputs D are those resources
which remain unchanged by production, such as equipment and
employees. The allocated input resources in D are always limited
in next period. A DMU’s allocated constant inputs will never
ð2Þ
decrease but may increase if central control allocates to the DMU
some extra from a limited amount that was newly obtained during
the current period. The variable inputs F, such as raw material,
parts, water, and energy, will be used up in the current period of
production, thus the amount of the total resources of variable
inputs in the next period will be only those allocated for the next
period. These variable input resources F are always more available
than other kinds of input resource in each period; we assume an
unlimited supply can be obtained by the central control unit.
84 J. Wu et al. / Computers & Industrial Engineering 101 (2016) 81–90

3. DEA-based resource allocation models 3.2. DEA-based MOLP model for resource allocation

3.1. Identification for production technology Suppose there are n DMUs controlled by a centralized unit in an
organization. For each j, DMUj uses w non-allocated inputs
Most of the existing models for the resource allocation problem Gj ¼ ðg 1j ; g 2j ; . . . ; g wj ÞT , m allocated constant inputs
are based on the assumption that each DMU may have its produc- Dj ¼ ðd1j ; d2j ; . . . ; dmj Þ ,
T
and t allocated variable inputs
tion on the efficient frontier of production possibility set formed by T
Fj ¼ ðf 1j ; f 2j ; . . . ; f tj Þ to produce s outputs Yj ¼ ðy1j ; y2j ; . . . ; ysj ÞT .
all the DMUs after having allocated the additional input resources
(Fang, 2013; Korhonen & Syrjänen, 2004; Lozano & Villa, 2004; Wu Suppose that for the next period of production the organization
et al., 2013), which potentially will bring the optimal output rev- has total additional inputs Ri ; i ¼ 1; 2; . . . ; m for each allocated con-
enue. This assumption may be reasonable in the long term because stant input resource and total additional inputs Ei ; i ¼ 1; 2; . . . ; t,
the inefficient DMUs may learn from or imitate efficient DMUs to for each allocated variable input resource. (Here ‘‘additional”
improve their technology. In short-term real-world contexts, how- means acquired for use in the next allocation.) The CDMU desires
ever, a mere increase or decrease in input resources may not be to allocate a proper proportion of these resources to each of a num-
enough to make their production technology efficient. Even after ber of DMUs in a favorable way. In this paper, we assume that the
resource allocation, the DMUs continue to produce using their production possibility set will not change after the resource alloca-
original technology and how to evaluate their original technology tion, an assumption which has been commonly used in previous
is a pivotal issue. Here, context-dependent DEA is introduced to studies such as Lozano and Villa (2004), Korhonen and Syrjänen
identify the actual production technology. Accordingly we define (2004), and Wu et al. (2013).
a new production possibility set which characterizes the possible In this section, in order to allocate the additional resources effec-
short-term production changes for each unit. tively, we start with the following DEA-based bi-objective LP model
The context-dependent DEA technique (Seiford & Zhu, 2003; to simultaneously maximize total changed output production and
Zhu, 2003) was introduced to measure the relative attractiveness minimize total allocated variable input resource consumption.
of a particular DMU when compared to others. In fact, the set of
X
n X
s
DMUs can be divided into different levels of efficient frontier, Max Dyrq ðIÞ
which can be obtained using the above model (3). VRS has been q¼1 r¼1

assumed in this model because each level of efficient frontier in X

n X
t

VRS depicts each kind of production better. If we remove the orig- Min Df iq ðIIÞ
q¼1 i¼1
inal efficient frontier, then the remaining inefficient DMUs can be X
used to determine a new second-level efficient frontier. If we s:t: kjq g ij 6 g iq i ¼ 1;.. .;w k ¼ 1;.. .;p q 2 LðkÞ ð5:1Þ
j2LðkÞ
remove this second-level efficient frontier, a third-level efficient X
frontier can be formed, and so on, until no DMU is left. Each such kjq dij 6 diq þ Ddiq i ¼ 1;... ;m k ¼ 1; ...; p q 2 LðkÞ ð5:2Þ
j2LðkÞ
efficient frontier can provide an evaluation context to compare X
the original technologies of the DMUs. Those DMUs in the same kjq f ij 6 Df iq i ¼ 1;. ..;t k ¼ 1; ...;p q 2 LðkÞ ð5:3Þ
j2LðkÞ
efficient frontier are called same layer DMUs. It is reasonable to X
expect the same layer DMUs to have some common properties. kjq yrj P yrq þ Dyrq r ¼ 1;.. .;s k ¼ 1;. ..;p q 2 LðkÞ ð5:4Þ
The following Algorithm I (Seiford & Zhu, 2003) can identify those j2LðkÞ
mpss
DMUs in the same layer. Ddiq ¼ 0 when diq P diq i ¼ 1;. ..;m q ¼ 1;... ;n ð5:5Þ
mpss mpss
diq þ Ddiq 6 diq when diq 6 diq i ¼ 1;.. .;m q ¼ 1; ; n ð5:6Þ
(1) Step 1: Set Layer 1, k ¼ 1. Using the above model (3) to eval- X
n
uate the entire set of DMUs, E1 , to obtain the first-level effi- Ddiq ¼ Ri i ¼ 1;. ..;m ð5:7Þ
cient DMUs, set L1 (the first-layer efficient frontier). q¼1

(2) Step 2: Exclude the first-layer efficient DMUs from future X

DEA runs. Eðkþ1Þ ¼ Ek  Lk . If Eðkþ1Þ ¼ u, the algorithm stops;
else go to step 3.
(3) Step 3: Evaluate the new subset of ‘‘inefficient” DMUs, Eðkþ1Þ ,
to obtain a new set of efficient DMUs Lðkþ1Þ (the new-layer
efficient frontier).
(4) Step 4: Let k ¼ k þ 1. Go to step 2.
Because those DMUs on the same efficient frontier have similar
performance in production, here we suppose that those DMUs in
the same layer have the same technology. With an increase or
decrease in inputs, the DMUs can perform their productions simi-
lar to other DMUs in the same layer. Therefore, those DMUs can
produce their products on their own changed production possibil-
ity set after the additional resources are allocated to them. The
changed production possibility set based on layer k is shown here.
8 9
< X X X
T BCC ¼ ðx; yÞ 2 LðkÞj kj xij 6 x; kj yrj P y; kj ¼ 1; kj P 0
: j2LðkÞ j2LðkÞ j2LðkÞ
ð4Þ
The changed production possibility set T CPPS defines each DMU’s
feasible production region after the additional input resources are
allocated to it.
n
Df iq 6 Ei i ¼ 1;. ..;t ð5:8Þ
q¼1
X
kjq ¼ 1 k ¼ 1; ...;pD fq 2 LðkÞ ð5:9Þ
j2LðkÞ
Ddiq 6 bi diq i ¼ 1; ...;m q ¼ 1; ;n ð5:10Þ
kjq P 0 Df iq P 0 8j 2 LðkÞ k ¼ 1;.. .;p i ¼ 1;. ..;m
ð5Þ
where L(k) denotes the set of observed DMUs that belong to layer k.
Here Ddiq , Df iq , and Dyrq denote the change amounts of constant
input i, variable input i, and output r respectively.
In model (5), maximizing the sum of all output changes among
all DMUs in next period models the fact that bringing the highest
amount of production is the primary objective of resource alloca-
tion for a centralized organization’s CDMU. Minimizing the sum
of all allocated variable inputs allocated to all DMUs in next period
=
as the CDMU’s secondary objective means that the organization
; also considers saving resources if that is possible while maintain-
ing maximum organizational output. The constraints from (5.1)–
(5.4) guarantee the new production for each DMU is in its own
changed production possibility set. Constraints (5.7) and (5.8) in
model (5) indicate that the total allocated resources among all
DMUs cannot exceed the total additional input resources. Con-
 J. Wu et al. / Computers & Industrial Engineering 101 (2016) 81–90 85

straint (5.9) indicates that VRS is assumed in this model. To guar-

antee that the proportional scaling is managerially feasible, follow-
ing Korhonen and Syrjänen (2004), we limit the change in constant D
9
inputs to be Ddiq 6 bi diq . These constraints eliminate solutions in C
8 G
which all resources are allocated to just a few DMUs with advanced
technology, and thereby reflect the idea that the CDMU should be 7
fair to all DMUs. 6
The allocated constant input resources D are assumed to be very F
5 B
valuable for production but scarce because of cost and/or availabil-
ity. The CDMU has expended great effort and/or funds to obtain 4
such resources, so decision makers would naturally want a return 3
on that investment as soon as possible. Therefore we assume that E
2
this type of resource should be allocated completely; none should
be kept in reserve. Recognizing the great value of allocated constant 1 A
input resources makes us want to guarantee they are allocated to
1 2 3 4 5 6 7 8 9
those DMUs who really need them, so we apply the concept of
new MPSS (Zhu & Shen, 1995), which is represented by constraints Fig. 1. Illustration of a simple one-input one-output example.
mpss
(5.5) and (5.6). diq means the maximum value of inputs i; i ¼
1; . . . ; m by DMUq in the area of new MPSS. For each layer, there is organization while maintaining maximum output. This leads us
only one such maximum value, that is, the DMUs in same layer have to define this growth rate of output as the effectiveness of a DMU
the same maximum value. The maximum value in each layer can be and the effectiveness of an organization is expressed as follows,
calculated by the following Algorithm II (Zhu & Shen, 1995). which offers a comprehensive method for resource allocation.
Step 1: Calculate the efficiency under the DMUs in each layer
using the FGL model. Definition 1. The effectiveness for an organization is defined as
Step 2: Find the efficient DMUs in each layer using the FGL the output growth rate for its all DMUs. It is a benefit index which
model. can be calculated as:
Step 3: Calculate the biggest value for each input i; i ¼ 1; . . . ; m X
n X
s
Dyrq
among the efficient DMUs in each layer. w¼ ð6Þ
q¼1 r¼1
yrq

The biggest value for each input i; i ¼ 1; . . . ; m in each layer is the The organization should not only consider the change of total
mpss
maximum value, which is denoted by diq . output and consumption of input resources, but also take into
Next, we will consider a numerical example. The raw dataset for account the effectiveness of the resource allocation. Therefore,
this example is shown in Table 1. the above model (5) can be replaced by model (7) which integrates
Fig. 1 shows this simple input and simple output dataset. The effectiveness.
DMUs are classified into two layers using the BCC model. Layer 1 X
n X
s
contains 4 DMUs: Lð1Þ ¼ fA; B; C; Dg. Layer 2 contains 3 DMUs: Max Dyrq ðIÞ
Lð2Þ ¼ fE; F; Gg. q¼1 r¼1

Suppose that these DMUs are all controlled by an organizational X

n X
t

central control. Also suppose that there is only one input in the Min Df iq ðIIÞ
q¼1 i¼1
class of allocated constant resources D but now the organization
has one additional unit of constant input R ¼ 1 that needs to be
n X
X s
Dy
Max rq
yrq
ðIIIÞ
allocated to one of those seven DMUs. Obviously, DMUA and DMUE q¼1 r¼1
both have the biggest marginal output for each efficient frontier in X
Fig. 1. Using model (5), DMUA or DMUE will be allocated the addi-
s:t: kjq g ij 6 g iq i ¼ 1;. .. ;w k ¼ 1; .. .; p q 2 LðkÞ
j2LðkÞ
tional one unit of input resource. That is to say, there are two allo- X
cation schemes which bring same maximum total output. kjq dij 6 diq þ Ddiq i ¼ 1;. .. ;m k ¼ 1; .. .; p q 2 LðkÞ
j2LðkÞ
Furthermore, the allocated unit of input can bring four units of X
additional output for DMUA or DMUE. However, the additional out- kjq f ij 6 Df iq i ¼ 1; .. .; t k ¼ 1; .. .; p q 2 LðkÞ
put shows different change proportion for the different DMUs. For j2LðkÞ
X
example, the output growth rate for DMUA would be 400% while for kjq yrj P yrq þ Dyrq r ¼ 1; .. .; s k ¼ 1;. .. ;p q 2 LðkÞ
the growth rate for DMUE would be 200%. It is reasonable to believe j2LðkÞ
that the additional output brings more satisfaction to the DMU with Ddiq ¼ 0 when diq P diq
mpss
i ¼ 1; .. . ;m q ¼ 1; .. .; n
smaller scale of current outputs. In order to maximize satisfaction mpss mpss
within the organization, the additional one unit of input will be diq þ Ddiq 6 diq when diq 6 diq i ¼ 1;. .. ;m q ¼ 1; .. .; n
allocated to DMUA where it will bring a bigger satisfaction for the X
n
Ddiq ¼ Ri i ¼ 1; . .. ;m
Table 1 q¼1
Input and output data of observed DMUs. X
n
Df iq 6 Ei i ¼ 1;. .. ;t
DMU Input Output Layer
q¼1
A 1 1 1 X
B 2 5 1 kjq ¼ 1 k ¼ 1; .. .; p q 2 LðkÞ
C 4 8 1 j2LðkÞ
D 6 9 1 Ddiq 6 bi diq i ¼ 1;. .. ;m q ¼ 1; . .. ;n
E 3 2 2
F 4 6 2
kjq P 0 Df iq P 0 8j 2 LðkÞ k ¼ 1;. .. ;p i ¼ 1; .. .; m
G 6 8 2 ð7Þ
86 J. Wu et al. / Computers & Industrial Engineering 101 (2016) 81–90

where maximizing the total outputs is the primary objective, min- Now, suppose that the paper company’s headquarters have
imizing the variable input resources is the secondary objective, and recruited an additional 500 employees (R ¼ 500) and should allo-
maximizing the effectiveness of the organization is considered as cate this additional staff to the eight branches. The staff resource
the tertiary objective. Model (7) is a multiple-objective program- is an allocated constant input, so the total staff input resources
ming problem in DEA (Amirteimoori & Emrouznejad, 2012; for a given branch in next period are the sum of its newly allocated
Amirteimoori & Kordrostami, 2012; Keshavarz & Toloo, 2014, staff and the original staff in the current period. Coal and water are
2015). Following the Multiple-objective programming approach the allocated variable inputs F, so the total amount of these two
(Amirteimoori & Kordrostami, 2012), we can transform the multiple input resources in next period will be exactly what is allocated
programming model (7) into the following single objective for the next period, no more. Because these allocated variable input
programming model (8): resources F are always abundantly available in each period, for our
example we will suppose each allocated variable resource
X
n X
s X
n X
t X
n X
s
Dyrq P
Max Dyrq  e1 Df iq þ e2 yrq
increases by 10% for the next period, (E1 ¼ 1:1 8q¼1 f 1q ; E2 ¼
q¼1 r¼1 q¼1 i¼1 q¼1 r¼1
P8
X 1:1 q¼1 f 2q ). Fixed investment belongs to set of non-allocated
s:t: kjq g ij 6 g iq i ¼ 1; .. .; w k ¼ 1;. .. ;p q 2 LðkÞ inputs G, which has invariable characteristics, so the amount of
j2LðkÞ fixed investment in the next period is the same as in the current
X
kjq dij 6 diq þ Ddiq i ¼ 1; .. .; m k ¼ 1;. .. ;p q 2 LðkÞ period. Therefore, the problem for the headquarters is how to
j2LðkÞ effectively and fairly allocate the additional staff, coal, and water
X
kjq f ij 6 Df iq i ¼ 1; .. . ;t k ¼ 1;. .. ;p q 2 LðkÞ to the eight branches, so as to bring optimal results for the
j2LðkÞ company.
X Our proposed model (8) in Section 4 is used to help the com-
kjq yrj P yrq þ Dyrq r ¼ 1;. .. ;s k ¼ 1; .. .; p q 2 LðkÞ
j2LðkÞ
pany to allocate these additional input resources. We previously
mpss
mpss determined that R ¼ 500, E1 ¼ 81; 965, E2 ¼ 2296, dq ¼
Ddiq ¼ 0 when diq P diq i ¼ 1;. .. ;m q ¼ 1;. .. ;n mpss
mpss mpss 1952ðq ¼ 1; 2; 3; 6Þ, and dq ¼ 944ðq ¼ 4; 5; 7; 8Þ. In addition, we
diq þ Ddiq 6 diq when diq 6 diq i ¼ 1; .. .; m q ¼ 1;. .. ;n
set Ddq 6 0:3  dq ðq ¼ 1; . . . ; 8Þto guarantee managerial feasibility.
X
n
Also, we set e1 ¼ 0:05 and e2 ¼ 0:01 in this application. Note that
Ddiq ¼ Ri i ¼ 1;. .. ; m
q¼1
one can set different values for e1 and e2 according to one’s prefer-
X
n ences. Through model (8), we can obtain a solution for allocating
Df iq 6 Ei i ¼ 1;. . .; t these input resources, as shown in Table 4.
q¼1 Table 4 lists the relative changes for each input and output. The
X
kjq ¼ 1 k ¼ 1; .. . ;p q 2 LðkÞ last row of Table 4 shows the total variation in each input/output
j2LðkÞ measure. Firstly, since fixed investment is a non-allocated invari-
Ddiq 6 bi diq i ¼ 1; .. .; m q ¼ 1;. .. ; n ant resource, the change for each DMU is zero. Secondly, the sum
of the staff increases in the DMUs is 500, which indicates that all
kjq P 0 Df iq P 0 8j 2 LðkÞ k ¼ 1; .. .; p i ¼ 1;. .. ;m
of the recruited staff are allocated. We see that DMU2, DMU6,
ð8Þ DMU7, and DMU8 are allocated 118, 62, 131, and 189 staff respec-
tively. Thirdly, the total allocation for coal and water are 78638.8
where e1 ð0 < e1 6 1Þ and e2 ð0 < e2 6 1Þ are the weights of input
and 2290.6 respectively, which are both smaller than the available
resources and effectiveness. The weight choice has a big influence
allocated variable resource (E1 ¼ 81; 965, E2 ¼ 2296). The results
on the optimal solutions. Different studies have used different
indicate that the scheme for resource allocation can save allocated
approaches to choose the weights (Amirteimoori & Emrouznejad,
variable input resources while maximizing output. Thirdly, the
2012; Amirteimoori & Kordrostami, 2012). This paper sets the
remaining four DMUs which have not been allocated new staff will
weights small enough (like 0.01 and 0.001) and requires e1 > e2 since
maintain their original production. Take DMU1 for example. Its
we put the total output in the most important position, followed
original input/output production was (19,569, 504, 1952, 86;
by input resource, with effectiveness as the least important. We
1740, 254). In the resource allocation scheme, no staff is allocated
should note that different CDMUs can set different weights when
to DMU1. Therefore, along with the unchanged staff and fixed
they have different preferences on input resource and effectiveness.
investment, DMU1’s coal and water allocations are also unchanged
The constraints are all the same as in model (7). Using model (8), a
to maintain its current production. Lastly, from the total change in
new comprehensive scheme for resource allocation is established.
output, we know the paper company can obtain 363.03 ten-
thousand Yuan more after those additional resources are allocated.
4. Application to a paper company The new production plans and BCC efficiencies for all branch
companies after resource allocation are presented in Table 5.
In this section, we apply our model to analyze a real example of Table 5 lists the resulting new input/output targets in columns
a paper company. The company consists of 8 branches (DMUs), 3–8 for each DMU for the next period. New BCC efficiency scores
each of which uses four inputs and two outputs, as listed in Table 2. after resource allocation and original BCC efficiency are calculated
The four inputs can be classified into three sets. Fixed invest-
ment G1 belongs to the set G of non-allocated inputs. Staff D1
Table 2
belongs to set D of allocated constant inputs. Coal F1 and water Input and output variables.
F2 belong to the set F of allocated variable inputs. The raw input
and output dataset and the output-oriented BCC efficiency of each Input/output Variable Units

branch using the VRS assumption are given in Table 3. Input Coal Tons
It can be seen from Table 3 that in this example, 4 out of the 8 Water 10-thousand tons
Staff Persons
paper branches are technically efficient. Through the algorithm in
Investment 10-thousand Yuan
Section 4, the 8 paper branches are divided into two layers. Layer
Output Value 10-thousand Yuan
1 contains 4 DMUs: Lð1Þ ¼ f1; 2; 3; 6g. Layer 2 contains 4 DMUs: Profit and Tax 10-thousand Yuan
Lð2Þ ¼ f4; 5; 7; 8g.
 J. Wu et al. / Computers & Industrial Engineering 101 (2016) 81–90 87

Table 3
The original dataset and technical efficiency.

DMU Input Output EVRS Class

Coal (tons) Water (10,000 tons) Staff (persons) Investment (10,000) Value (10,000) Profit and tax (10,000)
1 19,569 504 1952 86 1740 254 1.00 1
2 4867 105 631 10 539.1 152 1.00 1
3 14,570 478 1334 9 938.3 185.3 1.00 1
4 10,211 218 904 100 559.2 104 1.43 2
5 10,408 354 944 34 553.5 120.3 1.46 2
6 1584 29 205 10 187.8 45.2 1.00 1
7 4388 187 438 47 228.4 47.9 1.71 2
8 8917 213 631 36 301.4 72 1.82 2

Table 4
The results of resource allocation based on our proposed method.

Layer DMU Df 1q Df 2q Dd1q Dg 1q Dy1q Dy2q

1 1 19,569 504 0 0 0 0
2 6485.3 166.9 118 0 67.9 5.7
3 14,570 478 0 0 0 0
6 2058 40 62 0 50.7 15.4
2 4 10,211 218 0 0 0 0
5 10,408 354 0 0 0 0
7 5965.8 224 131 0 86.1 18.3
8 9371.7 305.8 189 0 162.7 29.9
Total 78638.8 2290.6 500.0 0.0 367.3 69.2

for all DMUs in column 9 and 10 respectively. Combining Tables 3
and 4, we can give a detailed explanation of the allocation and its
results. For example, DMU8 has: (a) total inputs 9371.7, 305.8, 820,
and 36 in next period, input changes 454.7, 92.8, 189, and 0; (b)
output targets 464.1 and 101.9, output changes 162.7 and 29.9;
and (c) new efficiency with resource allocation solution 1.48, an
increase of 0.34 from the original efficiency.
Note that compared with the original efficiency scores, all new
efficiencies are improved to some extent or remain efficient. Thus,
from an overall perspective of the paper company, efficiencies for
all branches are maintained or improved by this allocation of
resources.
The above results provide important and valuable information
to decision makers who supervise all branches from an overall per-
spective. By allocating the additional recruited staff, coal, and
water and resetting the production targets according to the results
listed in Table 5, decision makers are able to optimize the produc-
tion for each branch in order to achieve the maximum output and
effectiveness.
Next, in order to reflect the effects of e1 and e2 on the three
objectives, the following Tables 6–8 show the corresponding
results based on different e1 and e2 . From Table 6–8, we have the
following conclusions. (I) e1 has a big influence on the first objec-
tive of optimal total outputs, while e2 has little effect on the opti-
mal total outputs. When we care little about variable input
resources, we may get more total outputs. This fact is reflected
as the smaller of e1 , the bigger of the optimal total outputs. e2 is
used to improve effectiveness based on the optimal total outputs,
hence in Table 6 we know e2 has little influence on the optimal
total outputs. (II) e1 also has a big influence on the second objective
of variable input resources, while e2 has little effect. Obviously,
when we treat the variable input resources more important, that
is e1 increases, we may use less variable input resources. It is
reflected in Table 7 that as the smaller of e1 , the smaller of the total
variable input resources. (III) From Table 8, we know that both e1
and e2 has big influence on the third objective of effectiveness.
The effectiveness increases as e1 and e2 increase.
Last, in order to show the advantages of our proposed model,
we compare it with previous methods based on unchanged effi-
ciency after resource allocation. As a representative of unchanged
efficiency methods, Korhonen and Syrjänen’s (2004) method is
chosen to compare with our proposed model. The corresponding
results of resource allocation based on Korhonen and Syrjänen’s
unchanged efficiency method are shown in Table 9.
Combining Tables 4 and 9, we compare our proposed method
with Korhonen and Syrjänen’s model, denoted K&S. (I) The K&S
model does not consider the characteristics of the input resources.
For the total inputs of coal and water in next period after resource
allocation, they simply add the allocated resource to the original
resource, which may be unreasonable for these two inputs. For

Table 5
Resulting new input/output targets and efficiency.

Layer DMU f 1q f 2q d1q g 1q y1q y2q New EVRS Original EVRS

1 1 19,569 504 1952 86 1740 254 1.00 1.00

2 6485.3 166.9 749 10 607 157.7 1.00 1.00
3 14,570 478 1334 9 938.3 185.3 1.00 1.00
6 2058 40 267 10 238.5 60.6 1.00 1.00
2 4 10,211 218 904 100 559.2 104 1.40 1.43
5 10,408 354 944 34 553.5 120.3 1.43 1.46
7 5965.8 224 569 47 314.5 66.2 1.57 1.71
8 9371.7 305.8 820 36 464.1 101.9 1.48 1.82
88 J. Wu et al. / Computers & Industrial Engineering 101 (2016) 81–90

Table 6
The results of optimal total outputs based on different e1 and e2 .
e2
e1 0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009
0.01 436.5624 436.5624 436.5624 436.5624 436.5624 436.5624 436.5624 436.5624 436.5624 436.5624
e2
e1 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045
0.05 363.0347 363.0347 363.0347 363.0347 363.0347 363.0347 363.0347 363.0347 363.0347 363.0347
e2
e1 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
0.1 258.7285 258.7285 258.7285 258.7285 258.7285 258.7285 258.7285 258.7285 258.7285 258.7285
e2
e1 0 0.015 0.03 0.045 0.06 0.075 0.09 0.105 0.12 0.135
0.15 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943
e2
e1 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
0.2 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943
e2
e1 0 0.025 0.05 0.075 0.1 0.125 0.15 0.175 0.2 0.225
0.25 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943
e2
e1 0 0.03 0.06 0.09 0.12 0.15 0.18 0.21 0.24 0.27
0.3 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943 192.5943
e2
e1 0 0.035 0.07 0.105 0.14 0.175 0.21 0.245 0.28 0.315
0.35 192.5926 192.5943 192.5943 192.5943 192.5943 192.5943 192.5942 192.5942 192.5942 192.5942

Table 7
The results of variable input resources based on different e1 and e2 .
e2
e1 0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009
0.01 80929.3914 80929.3914 80929.3914 80929.3914 80929.3914 80929.3914 80929.3914 80929.3914 80929.3914 80929.3914
e2
e1 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045
0.05 79249.2279 79249.2279 79249.2279 79249.2279 79249.2279 79249.2279 79249.2279 79249.2279 79249.2279 79249.2279
e2
e1 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
0.1 77634.4532 77634.4532 77634.4532 77634.4532 77634.4532 77634.4532 77634.4532 77634.4532 77634.4532 77634.4532
e2
e1 0 0.015 0.03 0.045 0.06 0.075 0.09 0.105 0.12 0.135
0.15 77149.5273 77149.5273 77149.5273 77149.5273 77149.5274 77149.5273 77149.5274 77149.5273 77149.5273 77149.5273
e2
e1 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
0.2 77149.5272 77149.5272 77149.5272 77149.5272 77149.5272 77149.5272 77149.5272 77149.5272 77149.5272 77149.5272
e2
e1 0 0.025 0.05 0.075 0.1 0.125 0.15 0.175 0.2 0.225
0.25 77149.5272 77149.5272 77149.5272 77149.5272 77149.5272 77149.5272 77149.5272 77149.5272 77149.5272 77149.5272
e2
e1 0 0.03 0.06 0.09 0.12 0.15 0.18 0.21 0.24 0.27
0.3 77149.5272 77149.5272 77149.5272 77149.5272 77149.5272 77149.5272 77149.5272 77149.5272 77149.5272 77149.5272
e2
e1 0 0.035 0.07 0.105 0.14 0.175 0.21 0.245 0.28 0.315
0.35 77149.5223 77149.5271 77149.5272 77149.5272 77149.5271 77149.5270 77149.5270 77149.5269 77149.5268 77149.5268

example, the inputs of coal and water are those resources which
may be used up as the production is in progress. That is, the total
inputs of coal and water in next period are just the newly-allocated
resources; none of the previous period’s allocation remains. There-
fore, the needed allocated amounts of input resource for produc-
tion in next period are f iq þ Df iq , i ¼ 1; 2; q ¼ 1; . . . ; 8, not Df iq ,
i ¼ 1; 2; q ¼ 1; . . . ; 8. (II) Using the K&S solution, the total 500 staff
is allocated to six DMUs. What is more, we know that DMU2,
DMU6, DMU4, DMU5, DMU7, and DMU8 are allocated 20, 16, 173,
104, 131, and 55 new staff respectively. The remaining two DMUs
 J. Wu et al. / Computers & Industrial Engineering 101 (2016) 81–90 89

Table 8
The results of effectiveness based on different e1 and e2 .
e2
e1 0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009
0.01 0.4170 0.8340 1.2510 1.6680 2.0850 2.5020 2.9190 3.3360 3.7530 4.1700
e2
e1 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045
0.05 2.0747 4.1495 6.2242 8.2989 10.3737 12.4484 14.5231 16.5979 18.6726 20.7473
e2
e1 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
0.1 3.9947 7.9894 11.9840 15.9787 19.9734 23.9681 27.9628 31.9575 35.9521 39.9468
e2
e1 0 0.015 0.03 0.045 0.06 0.075 0.09 0.105 0.12 0.135
0.15 5.9500 11.9000 17.8500 23.8001 29.7501 35.7001 41.6501 47.6001 53.5501 59.5002
e2
e1 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
0.2 7.9334 15.8667 23.8001 31.7334 39.6668 47.6001 55.5335 63.4668 71.4002 79.3335
e2
e1 0 0.025 0.05 0.075 0.1 0.125 0.15 0.175 0.2 0.225
0.25 9.9167 19.8334 29.7501 39.6668 49.5835 59.5002 69.4169 79.3335 89.2502 99.1669
e2
e1 0 0.03 0.06 0.09 0.12 0.15 0.18 0.21 0.24 0.27
0.3 11.9000 23.8001 35.7001 47.6001 59.5002 71.4002 83.3002 95.2003 107.1003 119.0003
e2
e1 0 0.035 0.07 0.105 0.14 0.175 0.21 0.245 0.28 0.315
0.35 13.8834 27.7667 41.6501 55.5335 69.4169 83.3002 97.1836 111.0670 124.9503 138.8337

Table 9
The results of resource allocation based on Korhonen and Syrjänen’s (2004) method.

Layer DMU Df 1q Df 2q Dd1q Dg 1q Dy1q Dy2q

1 1 978 25 0 0 0 0
2 106 0 20 0 11.5 1.0
3 729 24 0 0 0 0
6 16.6 0 16 0 13.4 4.1
2 4 699.1 0 173 0 101.9 26.5
5 969.5 79.2 104 0 73.6 5.5
7 217.3 55.8 131 0 57.2 32.0
8 2165.4 56.4 55 0 21.8 13.7
Total 5880.7 240.5 500 0 279.5 82.7

(DMU1 and DMU3) are allocated zero new staff, but do get some
increase in amounts of coal and water. However, the allocated coal
and water bring no increase in output. For example, DMU1 is
allocated an extra 978 units of coal and 25 units of water, but its
outputs do not increase with the additional coal and water. (III)
With the same staff and fixed investment, our proposed model
needs just 78638.8 and 2290.6 units of coal and water for the pro-
duction in next period, whereas the K&S method needs 80394.7
(=5880.7 + 74,514) and 2328.5 (=240.5 + 2088) units of coal and
water. With less input, our proposed model can still bring 436.5
ten-thousand Yuan after those additional resources are allocated
and used in production, whereas the K&S method results in just
362.2 ten-thousand Yuan. (IV) The K&S method is based on
unchanged efficiency after resource allocation, and we know from
Table 5 that our proposed method can lead to better efficiency for
all DMUs.
5. Conclusions
The resource allocation problem is a hot topic in both theoreti-
cal and practical areas because of its importance in management.
In this paper, we have addressed a common resource allocation
problem that occurs in decision making environments where a
central unit controls all the resources of a set of DMUs. For gener-
alization, the input resources are classified into three groups: non-
allocated inputs, allocated constant inputs, and allocated variable
inputs. We combine context-dependent DEA and MOLP to help
the decision makers allocate the additional resources reasonably.
An illustrative application to a paper company is used to show
the applicability of the model to the resource allocation problem.
This paper has several contributions to the study of resource
allocation. Firstly, context-dependent DEA is used to characterize
the changed production possibility set, which determines which
production plans are feasible with input increase or decrease in
general. The set is based on the observed input/output of the DMUs
under consideration. Each DMU has its own changed production
possibility set. The changed production possibility set is a concept
introduced in this paper.
Secondly, the resource allocation problem is solved by using the
MOLP method which simultaneously maximizes total change out-
put and effectiveness, yet minimizes the total allocated variable
input consumption in the entire organization. The effectiveness is
defined as the output growth rate for all DMUs, which can reflect
90 J. Wu et al. / Computers & Industrial Engineering 101 (2016) 81–90
the effects brought by allocating the additional resources.
Compared with some other allocation methods, the proposed
model has considered the outputs, effectiveness, and inputs simul-
taneously while other resource allocation methods consider only
one of these objectives.
In the application illustration, we find that the weight choice of
e1 and e2 for input resources and effectiveness has a big influence
on the optimal value. This reveals that choosing proper weights
for the CDMU should be further studied. Future studies could also
consider real-time resource allocation and how to accurately fore-
cast each unit’s further production. Another future study may turn
to the problem where there are interdependencies between units.
How to allocate the input resources and decide production in a
stochastic environment based on a stochastic production frontier
approach is also an interesting research direction.
Acknowledgments
The research is supported by National Natural Science Founda-
tion of China (Nos. 71222106, 71501189, 71571173 and
71631008), Research Fund for the Doctoral Program of Higher Edu-
cation of China (No. 20133402110028), Foundation for the Author
of National Excellent Doctoral Dissertation of the People’s Republic
of China (No. 201279), and Top-Notch Young Talents Program of
China. Qingxian An thanks the support of the International Post-
doctoral Exchange Fellowship Program by the Office of China Post-
doctoral Council.
References
Amin, G. R., & Toloo, M. (2004). A polynomial-time algorithm for finding e in DEA
models. Computers & Operations Research, 31(5), 803–805.
Amirteimoori, A., & Emrouznejad, A. (2012). Optimal input/output reduction in
production processes. Decision Support Systems, 52(3), 742–747.
Amirteimoori, A., & Kordrostami, S. (2012). Production planning in data
envelopment analysis. International Journal of Production Economics, 140(1),
212–218.
Amirteimoori, A., & Shafiei, M. (2006). Characterizing an equitable omission of
shared resources: A DEA-based approach. Applied Mathematics and Computation,
177(1), 18–23.
An, Q., Wen, Y., Xiong, B., Min, Y., & Chen, X. (2016b). Allocation of carbon dioxide
emission permits with the minimum cost for Chinese provinces in big data
environment. Journal of Cleaner Production. http://dx.doi.org/10.1016/j.
jclepro.2016.07.072.
An, Q., Yan, H., Wu, J., & Liang, L. (2016). Internal resource waste and centralization
degree in two-stage systems: An efficiency analysis. Omega, 61, 89–99.
Asmild, M., Paradi, J. C., & Pastor, J. T. (2009). Centralized resource allocation BCC
models. Omega, 37(1), 40–49.
Athanassopoulos, A. D. (1995). Goal programming & data envelopment analysis
(GoDEA) for target-based multi-level planning: Allocating central grants to the
Greek local authorities. European Journal of Operational Research, 87(3),
535–550.
Banker, R. D. (1984). Estimating most productive scale size using data envelopment
analysis. European Journal of Operational Research, 17(1), 35–44.
Banker, R. D., Charnes, A., & Cooper, W. W. (1984). Some models for estimating
technical and scale inefficiencies in data envelopment analysis. Management
Science, 30(9), 1078–1092.
Beasley, J. E. (2003). Allocating fixed costs and resources via data envelopment
analysis. European Journal of Operational Research, 147(1), 198–216.
Bi, G., Ding, J., Luo, Y., & Liang, L. (2011). Resource allocation and target setting for
parallel production system based on DEA. Applied Mathematical Modelling, 35(9),
4270–4280.
Cecilio, M. M., & Diego, P. (2006). On centralized resource allocation using DEA.
Working Paper, Kent Business School, University of Kent, UK.
Charnes, A., Cooper, W. W., & Rhodes, E. (1978). Measuring the efficiency of decision
making units. European Journal of Operational Research, 2(6), 429–444.
Cook, W. D., & Seiford, L. M. (2009). Data envelopment analysis (DEA)–Thirty years
on. European Journal of Operational Research, 192(1), 1–17.
Cooper, W. W., Seiford, L. M., & Tone, K. (2007). Data envelopment analysis: A
comprehensive text with models, applications, references and DEA-solver software
(2nd ed.). Springer. ISBN, 387452818, 490.
Du, J., Cook, W. D., Liang, L., & Zhu, J. (2014). Fixed cost and resource allocation
based on DEA cross-efficiency. European Journal of Operational Research, 235(1),
206–214.
Fang, L. (2013). A generalized DEA model for centralized resource allocation.
European Journal of Operational Research, 228(2), 405–412.
Fang, L., & Li, H. (2015). Centralized resource allocation based on the cost–revenue
analysis. Computers & Industrial Engineering, 85, 395–401.
Färe, R., Grosskopf, S., & Lovell, C. K. (1985). The measurement of efficiency of
production. Springer Science & Business Media.
Golany, B. (1988). An interactive MOLP procedure for the extension of DEA to
effectiveness analysis. Journal of the Operational Research Society, 725–734.
Golany, B., Phillips, F. Y., & Rousseau, J. J. (1993). Models for improved effectiveness
based on DEA efficiency results. IIE Transactions, 25(6), 2–10.
Golany, B., & Tamir, E. (1995). Evaluating efficiency-effectiveness-equality trade-
offs: A data envelopment analysis approach. Management Science, 41(7),
1172–1184.
Hatami-Marbini, A., Tavana, M., Agrell, P. J., Lotfi, F. H., & Beigi, Z. G. (2015). A
common-weights DEA model for centralized resource reduction and target
setting. Computers & Industrial Engineering, 79, 195–203.
Hosseinzadeh Lotfi, F., Hatami-Marbini, A., Agrell, P. J., Aghayi, N., & Gholami, K.
(2013). Allocating fixed resources and setting targets using a common-weights
DEA approach. Computers & Industrial Engineering, 64(2), 631–640.
Keshavarz, E., & Toloo, M. (2014). Finding efficient assignments: An innovative DEA
approach. Measurement, 58, 448–458.
Keshavarz, E., & Toloo, M. (2015). Efficiency status of a feasible solution in the
Multi-Objective Integer Linear Programming problems: A DEA methodology.
Applied Mathematical Modelling, 39(12), 3236–3247.
Korhonen, P., & Syrjänen, M. (2004). Resource allocation based on efficiency
analysis. Management Science, 50(8), 1134–1144.
Lozano, S., & Villa, G. (2004). Centralized resource allocation using data
envelopment analysis. Journal of Productivity Analysis, 22(1–2), 143–161.
Lozano, S., Villa, G., & Canca, D. (2011). Application of centralised DEA approach to
capital budgeting in Spanish ports. Computers & Industrial Engineering, 60(3),
455–465.
Mandell, M. B. (1991). Modelling effectiveness-equity trade-offs in public service
delivery systems. Management Science, 37(4), 467–482.
Seiford, L. M., & Thrall, R. M. (1990). Recent developments in DEA: The
mathematical programming approach to frontier analysis. Journal of
Econometrics, 46(1), 7–38.
Seiford, L. M., & Zhu, J. (2003). Context-dependent data envelopment analysis—
Measuring attractiveness and progress. Omega, 31(5), 397–408.
Wu, J., An, Q., Ali, S., & Liang, L. (2013). DEA based resource allocation considering
environmental factors. Mathematical and Computer Modelling, 58(5),
1128–1137.
Wu, J., Xiong, B., An, Q., Sun, J., & Wu, J. (2015). Total-factor energy efficiency
evaluation of Chinese industry by using two-stage DEA model with shared
inputs. Annals of Operations Research. http://dx.doi.org/10.1007/s10479-015-
1938-x.
Wu, J., Zhu, Q., & Liang, L. (2016). CO2 emissions and energy intensity reduction
allocation over provincial industrial sectors in China. Applied Energy. http://dx.
doi.org/10.1016/j.apenergy.2016.01.008.
Zhu, J. (2003). Quantitative models for performance evaluation and benchmarking:
Data envelopment analysis with spreadsheets and DEA excel solver. Springer.
Zhu, J., & Shen, Z. H. (1995). A discussion of testing DMUs’ returns to scale. European
Journal of Operational Research, 81(3), 590–596.