You are on page 1of 13

Journal of the Operational Research Society (2004) 55, 1090–1101 r 2004 Operational Research Society Ltd.

d. All rights reserved. 0160-5682/04 $30.00

www.palgrave-journals.com/jors

A multi-objective approach to determine


alternative targets in data envelopment analysis
MP Estellita Lins*, L Angulo-Meza and AC Moreira da Silva
Operational Research, Production Engineering, COPPE/UFRJ, Brazil
The choice for radial projections of classic data envelopment analysis (DEA) models, resulting in a number of
projections onto the Pareto-inefficient portion of the frontier, has been seen lately as a disadvantage in DEA. The search
for a non-radial projection method resulted in developments such as preference structure models. These models consider
a priori preference incorporation, using weights in the search for the most preferred efficient target, although presenting
some implementation difficulties. In this paper, we propose a multi-objective approach that determines the bases for a
posteriori preference incorporation, through individual projections of each variable (input or output) as an objective
function, thus allowing one to obtain a target at every extreme-efficient point on the frontier. This multi-objective
approach is shown to be equivalent to the preference structure models, yet presenting some advantages, such as the
mapping of the possible weights, assigned to partial efficiencies of an observed unit, in order to reach a specific target.
Journal of the Operational Research Society (2004) 55, 1090–1101. doi:10.1057/palgrave.jors.2601788

Keywords: data envelopment analysis; multi-objective linear programming

Introduction Kornbluth6 noticed that the DEA model could be


expressed as a multi-objective linear fractional programming
From a managerial point of view, there could be other
problem. The objective function of the model has the same
projections that are preferable to the radial projections of
expression as in the CCR multipliers model, but applied to
classic data envelopment analysis (DEA) models. Moreover,
maximize efficiency of every DMU, instead of one at a time,
radial projections often lead us to Pareto-inefficient portions
the restrictions remaining unchanged.
of the frontier. These two issues have a common source: the
Also, Joro et al,7 observing the problem of characterizing
equi-proportional variation of input or output levels,
efficient facets, made a structural comparison of the CCR8
proposed by Farrell1 as a means of measuring efficiency.
and BCC9 models with the reference point approach10 to
Alternatives to radial projections exist in DEA literature.
solving multi-objective problems.
One of the main approaches is through models that
Another methodology based upon a multi-objective
incorporate preferences using preference structures, as
model was proposed by Halme et al,11 known as Value
proposed by Thanassoulis and Dyson2 and Zhu3 or by the Efficiency Analysis. In this methodology, a multi-objective
specification of an ideal point.2 model is used to determine an existing or virtual (combina-
The appropriateness of using multi-objective linear tion of existing) DMU preferred by the decision-maker,
programming (MOLP) in a DEA context is supported by which is called the Most Preferred Solution (MPS). Once the
the concept of Pareto efficiency, which DEA and MOLP MPS is identified, an efficient frontier is determined in a
share. Both of them search for a set of different units that are DEA-like manner. Joro12 made an extension of the Value
better in at least one aspect (or objective function) than the Efficiency Analysis method to determine targets for ineffi-
others, thus characterizing an efficient facet of the feasible cient DMUs.
solution space of a problem.4 Li and Reeves13 presented a multi-objective model that
Credit for the very first work integrating DEA and MOLP considers two additional efficiency measures: the minimiza-
is due to Golany,5 who proposed an interactive multi- tion of the sum of the DMU distances to the frontier
objective procedure (IMOLP—interactive MOLP) to deter- (minisum) and the minimization of the largest distance
mine efficient output levels. The algorithm consists of (minimax), besides the maximization of the classical
sequential solutions to a set of related linear programming efficiency in DEA.
problems in which the objective function is to maximize a The Multi-Objective Simplex approach proposed in this
weighted sum of the former objectives. paper, not only allows calculation of the coordinates of all
alternative extreme efficient DEA targets, but also enables
*Correspondence: MP Estellita Lins, Rua Belisário Távora 80, ap 506
Laranjeiras, Rio de Janeiro RJ, Brazil. putting forward a new mapping of the weight range given to
E-mail: estellit@iis.com.br each input/output improvement onto those targets. This
MP Estellita Lins et al—Data envelopment analysis 1091

promotes insights into the DM preferences, regarding  0 and I 0. The model is:
variables: R
X X
variations in input or output intensity. We propose an Maximize wrþ fr  w
i ji
additional reference point approach that allows a better r2R0 i2I0
analyst-user interaction environment, benefiting from a 0 1
better visualization of the DEA targets. X X
þ e@ s
i þ srþ A ð1Þ
The content of this paper is organized in the following
i2I 0 r2R0
way: the next section comprises a revision of two models to
determine targets in DEA using the preference structure, Subject to
their description, and the disadvantages of this approach. In X
n

the following section, the first of DEA-MOLP models is fr yrj0  lj yrj ¼ 0; r 2 R0 ; ð2Þ
j¼1
introduced: MORO (Multiple Objective Ratio Optimiza-
tion), which optimizes the ratios between observed and X
n
target inputs (or outputs) of a DMU. We include a concise ji xij0  lj xij ¼ 0; i 2 I0 ; ð3Þ
value judgement-driven classification of MOLP methods, j¼1

which helps understand the relationship among the models X


n
discussed here. Also, a straight comparison between DEA- lj yrj  srþ ¼ yrj0 ; r 2 R0 ð4Þ
MOLP and preference structure models is shown, revealing j¼1

their formal equivalence. In the ensuing section, we propose


X
n
the use of indifference regions to map the range of weights lj xij þ s
i ¼ xij0 ; i 2 I0 ð5Þ
onto the target set. Then, we present a second DEA model: j¼1
MOTO (Multiple Objective Target Optimization), which
directly optimizes the target values, thus promoting the use fr X1 8r 2 R0 ; ð6Þ
of one interactive MOLP approach: the Pareto Race. Final
jip1 8i 2 I0 ð7Þ
sections are devoted to a case study using the MORO model
and conclusion. lj X0 8j ð8Þ

s þ
i ; sr X0 8i 2 I 0 e 8r 2 R0 ; e40 ð9Þ
Non-radial models to determine targets in DEA The decision-maker’s preference structure is expressed
Classic DEA models, such as constant returns to scale (CRS) through the weights wrþ and w i for the factors fr and ji,
or variable returns to scale (VRS) yield targets obtained as where fr reflects the increment in the value of output r, and
the equi-proportional reduction of input values, when input- ji reflects the decrease in the value of input i, in order to
oriented, or the equi-proportional increase in output values, reach the target.
in the output-oriented case (without including slacks), thus Thus, the objective function consists of maximizing the
providing radial efficiency measures, according to the weighted sum of the ratios (considering output increments
Farrell1 measure of efficiency. and input reductions) between observed and projected data.
From a managerial point of view, there could be several Projections are considered for sets of inputs and outputs that
interesting alternative targets to be explored (characterized can be improved, R0 and I0, through constraints (2) and (3).
by input and output levels), besides the radial approach. For the non-discretionary variables R  0 and I 0, we have
Several researchers are concerned with the determination of constraints (4) and (5).
targets through different approaches, the most popular one Bounds imposed on f (6) and j (7) prevents the levels of
being based on preference structures. These models require a outputs from decreasing and those of inputs from increasing.
priori information from the user or decision-maker about If fr* ¼ ji* ¼ 1, 8rAR0 , 8iAI0, and srþ * ¼ s 
i * ¼ 0 8rAR0,

the relative importance of variables, expressed in the form of 


8iAI 0. then the DMU j0 is Pareto efficient. Otherwise, the
weights. DMU j0 is inefficient and the target is given by
Two alternative preference structure models are as yrj0 ¼ fr yrj0 8r 2 R0 ð10Þ
follows.
yrj0 ¼ yrj0 þ srþ  8r 2 R0 ð11Þ

Thanassoulis and Dyson model xij0 ¼ ji xij0 8i 2 I0 ð12Þ


Thanassoulis and Dyson2 proposed assigning weights to xij0 ¼ xij0  s 8i 2 I0 ð13Þ
i
changes in the variables in order to implement a preference
structure. For a DMU j0, outputs and inputs are classified In real-life applications, this model may be used
into two categories: variables that we want to change, set R0 interactively, by testing different sets of weights (wrþ and
for the outputs and I0 for the inputs; and unchangeable wi ) and allowing the decision-maker to search for appro-
1092 Journal of the Operational Research Society Vol. 55, No. 10

priate DMU targets. It is important to remark that this will allow knowledge of weight mixes that will lead to
model does not indicate the efficiency score for the DMU. interesting solutions from the decision-maker’s perspective.

Zhu model Multi-objective model for ratio optimization (MORO)


Based on Russell’s efficiency measure, Zhu presented a 3 and with dominance (MORO-D)
model to determine targets that also incorporate a preference Considering an efficient DMU, we propose here that each
structure: output to be increased (or input to be reduced) can be
X s Xm
treated as a separate objective function within a Multiple
Maximize wrþ fr  w
i ji
r¼1 i¼1
ð14Þ Objective Linear Programming context. Thus, the following
Subject to model considers as objective functions the independent
optimization of the increment/decrease of each input/output,
X
n
in order to obtain an efficient target for the DMU j0:
lj yrj  srþ ¼ fr yrj0 ; r ¼ 1; . . . ; s ð15Þ
j¼1 max f1 ð20Þ

X
n ...
lj xij þ s
i ¼ ji xij0 ; i ¼ 1; . . . ; m ð16Þ
j¼1 max fs

ji ; fr free 8i; r; s þ
i ; sr X0 8i; r ð17Þ min j1 ð21Þ

X
s X
m ...
where wrþ  w
i ¼1
r¼1 i¼1 min jm
The sum of weights condition prevents the user from Subject to
inadvertently testing a redundant weight set that is equal to a X
n
previous set multiplied by a constant amount. Although the fr yrj0 ¼ yrj lj 8r ¼ 1; . . . ; s ð22Þ
j¼1
Zhu model does not specify restrictions for the non-
discretionary variables, this can be considered by assigning X
n
zero weights to respective selected inputs/outputs in the ji xij0 ¼ xij lj 8i ¼ 1; . . . ; m ð23Þ
objective function. The major difference in the Zhu model is j¼1

the unboundedness of ji and fr, allowing output reduction


fr ; ji ; lj X0 8r; i; j
and/or input increment, that is, the observed DMU is not
necessarily dominated by the target DMU, like in the This DEA-MOLP model will be referred to as MORO
Thanassoulis and Dyson model. CRS – Multiple Objective Ratio Optimization, as it
In real-life cases, from a managerial point of view, an optimizes ratios between inputs (outputs) of the target
input increment (or output reduction) to reach the efficient DMU and the observed DMU, under constant returns to
frontier could be feasible and preferable. For example, in the scale (CRS). To consider variable returns to scale (VRS), we
labour field where the managers would prefer a solution that add the convexity restriction:
implies hiring instead of dismissing workers, which could X n
produce a negative impact on a firm’s public image. lj ¼ 1 ð24Þ
If fr* ¼ ji* ¼ 1 and srþ * ¼ s
i * ¼ 0 then the DMU j0 is
j¼1

Pareto efficient. Otherwise, the DMU is inefficient and the The model including this restriction (24) will be referred to
target is given by as MORO VRS version. For both versions, the decision
yrj0 ¼ fr yrj0 þ srþ  8r ð18Þ variables are fs, lj, jm.
When the problem results in optimal solutions
xij0 ¼ ji xij0  s
i 8i ð19Þ fr* ¼ ji* ¼ 1 8r, i, then DMU j0 lies on the Pareto efficient
(non-dominated) frontier. The targets are:
As in the previous model, the decision-maker can search
yrj0 ¼ fr yrj0 8r ð25Þ
interactively for more appropriate targets for the DMU,
testing different weight sets. xij0 ¼ ji xij0 8i ð26Þ
There is an important drawback when determining DEA
targets in both the Thanassoulis and Dyson and the Zhu The final value of fr* and ji* depends on the decision-
models: the arbitrary choice of weights to be given to factors maker’s choice among the non-dominated solutions of the
fr and ji. We have no guidance for establishing weights problem (in multi-objective linear programming we get
before testing the models; just the trial and error experience efficient solutions, also known as non-dominated solutions).
MP Estellita Lins et al—Data envelopment analysis 1093

When we require dominance of the target over the Malivert,20 among others, undertook the problem of
observed DMU, we will add the following restrictions: finding the efficient set in MOLP.
fr X1 8r ¼ 1; . . . ; s ð27Þ
In the following three sections, we will develop MORO
jip1 8i ¼ 1; . . . ; m ð28Þ
model equivalence to preference structures and indifference
The resulting DEA-MOLP model will be called MORO-D regions, using the multi-objective simplex method to solve
(Multi-objective Ratio Optimization with Dominance), both DEA-MOLP models, a method that belongs to the third
CRS—considering expressions (20)–(23) and VRS—adding class and gives all the extreme points of the efficient set. We
the convexity restriction (24). While desirable, dominance have run both the ADBASE software21 and the DEA-
can be dismissed in some real-world cases as Zhu3 pointed MOLP simplex implemented by Frota Neto.22 After that, we
out. Indeed, besides the labour field example given in the will implement one interactive reference point method: the
previous section about Zhu model, there are cases where Pareto Race,18 to implement the MOTO model. It is worth
trade offs over the frontier allow a large increase of outputs, remarking that, although the authors did not mention it,
compensating for a small increase in one input. preference structure models could be considered as weighted
In solving the MORO and MORO-D models, we can sum a priori preference aggregation MOLP models.
provide the decision-maker with a complete set of extreme
efficient targets for the observed DMU, supporting the
Comparison of the multi-objective and preference
implementation of value judgements.
structure models
Let us consider the following example in order to make
Classification of methods for solving MOLP problems comparisons among the MORO and MORO-D models with
the Zhu, and the Thanassoulis and Dyson model:
In a MOLP problem an optimal solution, that is, one that
In Table 1, we consider seven DMUs that use input X to
optimizes all objectives simultaneously, is generally impos-
produce output Y. We applied Linear Programming soft-
sible to find. So, the process of solving MOLP problems
ware (LINDO) to run the preference structure models,
consists of finding non-dominated solutions, that is, solu-
testing different weights for j and f until the complete range
tions in the feasible set of the decision variable space, which
cannot be altered to improve one objective function value
without deteriorating at least one other. This concept is
Table 1 Original data set
analogous with the Pareto–Koopmans efficiency.
According to Clı́maco et al,14 there are several possible Variable
classification criteria for MOLP methods. We will adopt
DMU X Y
here the following classification, which is based on the stage
at which value judgements are incorporated. Three classes A 2 12
resulted: B 4 15
C 6 6
(i) Methods with a priori preference aggregation. These D 2 9
methods usually transform the MOLP problem into a E 3 15
single objective one (LPP) through an equivalent F 1 7
G 2 4
objective function, whose optimal solution is a non-
dominated solution for the original MOLP problem. This
function, known as the scalarizing function, expresses the X vs Y
DM’s preferences. There are three kinds of scalarizing 16
functions, which consist of: (a) the optimisation of one 14 E B
objective function restricting the others; (b) the weighted
12 A
sum of all objective functions; and (c) the use of a
distance function or reference point approach. 10
D
(ii) Methods characterized by a progressive preference
Y

8
articulation, where the decision-maker maintains a 6 F C
continuous interaction with the problem. For an
4 G
extended discussion of the several classification alter-
2
natives for interactive methods, readers are referred to
Steuer.15 0

(iii) Methods that generate efficient solutions and incorpo- 0 1 2 3 4 5 6 7

rate preferences a posteriori. Geoffrion (1968),16 Evans X


and Steuer,17 Zeleny,18 Ecker et al19 and Armand and Figure 1 CRS frontier.
1094 Journal of the Operational Research Society Vol. 55, No. 10

Table 2 MORO model (VRS)


Actual value Objective function value Target value

DMU Y X Max f Min j Y X

A 12 2 0.58333 0.5 7 1
1 1 12 2
1.25 1.5 15 3
B 15 4 0.46667 0.25 7 1
0.8 0.5 12 2
1 0.75 15 3
C 6 6 2.5 0.5 15 3
2 0.33333 12 2
1.16667 0.16667 7 1
D 9 2 0.77778 0.5 7 1
1.33333 1 12 2
1.66667 1.5 15 3
E 15 3 0.46667 0.33333 7 1
0.8 0.66667 12 2
1 1 15 3
F 7 1 1 1 7 1
1.71429 2 12 2
2.14286 3 15 3
G 4 2 1.75 0.5 7 1
3 1 12 2
3.75 1.5 15 3

X vs Y X vs Y
16 16

14 E B 4 E
14 B

12 A 12 A
10 10
D D
8
Y

6 F C F
6 C

4 G 4 3 G

2 2

0 0
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
X X

Figure 2 VRS frontier for DMU D. Figure 3 CRS frontier, projections for DMU G.

of basic efficient solutions was obtained. For the multi- efficient DMUs (F, A and E), corresponding to basic non-
objective models, we used ADBASE,21 which gives all dominated solutions in the MOLP problem.
efficient, basic solutions, although we can also find non-basic In Figure 2, we present the alternative targets for DMU
solutions, as will be done when applying MOTO model. D. For example, if the DM plans to increase the input X of
The results concerning CRS, for the Zhu model as for the DMU D by 50%, then he should consider a required output
MORO model, show that any point on the efficient CRS increase of 66.7%, taking DMU E as a possible efficient
frontier (Figure 1) serves as a target for every efficient or target, according to Table 2.
inefficient DMU, given that f and j are unbounded. In this Non-extreme efficient solutions were not attained, the
sense, there is equivalence between solutions for both Zhu reason for which will become clear through the concept of
and MORO models. indifference regions, in the respective section.
The same equivalence is true under VRS, given that Zhu The MORO-D model solved in ADBASE gave efficient
and MORO models yield the same set of efficient solutions. basic solutions that were compared to those obtained by the
Results from ADBASE for the MORO VRS model show Thanassoulis and Dyson model, considering CRS and VRS.
(Table 2) that the target set is unique, no matter which For the CRS frontier (Figure 3), we obtained different sets
DMU is being projected. This set is composed of all extreme of targets for each projected DMU, shown in Table 3.
MP Estellita Lins et al—Data envelopment analysis 1095

Table 3 MORO-D model (CRS)


Actual value Objective function value Target value

DMU Y X Max f Min j Y X

A 12 2 1.166667 1 14 2
1 0.857143 12 1.71
B 15 4 1.866667 1 28 4
1 0.535714 15 2.14
C 6 6 7 1 42 6
1 6
D 9 2 1.555556 1 14 2
1 0.642857 9 1.29
E 15 3 1 0.741935 15 2.23
1.4 1 21 3
F 7 1 1 1 7 1
G 4 2 1 0.285714 4 0.57
3.5 1 14 2

The set of possible targets (basic feasible solutions) for X vs Y


16
DMU G (Figure 3) are the efficient points 3 and 4. It is
worth noting that the targets for every DMU are propor- 14 E B

tional to input and output values for the DMU F, which 12 A


defines the efficient frontier under CRS. 10
Adding the convexity condition to the MORO-D D
8
Y

formulation, we get the VRS frontier. Thanassoulis and F


6 C
Dyson program was solved for DMU G, and resulted in the
targets corresponding to DMUs F and A, shown in Figure 4. 4 4 G

We obtained an identical set of basic efficient solutions using 2


MORO-D. 0
The Non-Pareto efficient frontier cannot be attained at all, 0 1 2 3 4 5 6 7
unlike the classic models that would yield point 4 in Figure 4 X
as the input oriented projection of DMU G.
Figure 4 VRS frontier, projections for DMU G.
Table 4 shows results for the MORO-D (VRS) program:
the values for f and j, the targets and the strictly positive l’s
corresponding to the reference set of DMUs.
Results from DEA-MOLP models, compared to those MOLP models yielded all the non-dominated solutions in
from Zhu and Thanassoulis and Dyson models, pointed out one run, while preference structure models required the
the equivalence between target sets for the DMUs. DEA- specification of new weighting sets in every new run,

Table 4 MORO-D model (VRS)


Actual value Objective function value Target value Lambdas

DMU Y X Max f Min j Y X lA lE lF

A 12 2 1 1 12 2 1
B 15 4 1 0.75 15 3 1
C 6 6 2 0.333333 12 2 1
2.5 0.5 15 3 1
1.166667 0.166667 7 1 1
D 9 2 1.333333 1 12 2 1
1 0.7 9 1.4 0.4 0.6
E 15 3 1 1 15 3 1
F 7 1 1 1 7 1 1
G 4 2 3 1 12 2 1
1.75 0.5 7 1 1
1096 Journal of the Operational Research Society Vol. 55, No. 10

sometimes failing to obtain a new target, because various X


s X
m

weighting sets can lead to the same target solution. This can Maximize wrþ fr  w
i ji ð31Þ
r¼1 i¼1
be seen with the aid of the indifference regions in the
corresponding section. Subject to
X
n
fr yrj0 ¼ yrj lj 8r ¼ 1; . . . ; s
Formalizing equivalence j¼1

In the following paragraphs we are going to formalize the X


n
ji xij0 ¼ xij lj 8i ¼ 1; . . . ; m
equivalence between the models analysed. Conceptually, this j¼1
arises from a result obtained by Geoffrion,16 which states the
equivalence between an LP model with multiple objective fr X1 8r ¼ 1; :::; s
functions and an LP model whose sole objective function
ji p1 8i ¼ 1; . . . ; m
consists of a convex linear combination of the former objective
functions. He showed that, given a multi-objective model: fr ; ji ; lj X0 8r; i; j
Max CX ð29Þ Ifwrþ , w
are strictly greater than zero, then there will be no
i

S.t. slacks, and we can assume:


yrj0 ¼ fr yrj0 8r 2 R0 ð32Þ
AXpb
xij0 ¼ ji xij0 8i 2 I0
XX0
If we admit any component of wrþ or w i equal to zero
X* is an efficient solution for this model if and only if ( y*, (the unchanged variables case in the Thanassoulis and
y*40, Syi* ¼ 1, such that X* is optimal for Dyson model), then, we should use
Maxy CX ð30Þ Xn
lj yrj Xfr yrj0 ; r ¼ 1; . . . ; s
S.t. j¼1

AXpb X
n
lj xij pji xij0 ; i ¼ 1; . . . ; m
XX0 j¼1

Kornbluth23 stated that, given a MOLP problem, for and add such slacks (multiplied by e) to the objective
every weighting set there is a non-dominated basic solution. function. As we can see, the MORO model considering
Cohon24 added that weights should be strictly positive in dominance is equivalent to the Thanassoulis and Dyson
order not to generate weakly efficient solutions, but model, in which all variables are considered to improve. The
Romero25 showed that, while possible from a theoretical Zhu model includes slacks in constraints, in order to let ji,
point of view, this is not likely to happen in practice. In fact, fr be free, but not within the objective function, because wrþ
this happens in the presence of multiple optimal solutions or w
i is not allowed to be equal to zero.

and can be avoided if we substitute a small infinitesimal


amount for a zero coefficient in the objective function.
Obtaining indifference regions
As we will apply this theorem to a known vector y in order
to obtain an optimal solution for model (30), it is easy to see Let us consider that a given projection on the frontier can be
P
that imposing restriction yi* ¼ 1 will not be necessary. achieved through various different sets of weights. Thus,
The equivalence between models (29) and (30) implies each weight can be allowed to vary inside a limited
that, for any given efficient solution X*, we can use the (indifference) region, while projecting a given observed unit
associated y* to form a weighted function y*CX*. This onto the same point on the frontier. As we will show, the
linear function is an example of a utility function for the indifference regions can be used to map the range of weights
decision-maker, where y* represents the trade-offs between (j and f), given to each objective function, onto all the
objectives, with X* as an optimal solution. In the case of the possible targets. These weights indicate the ‘importance’
Zhu model, and Thanassoulis and Dyson model, this y* given to a variable improvement, therefore revealing the
vector is equivalent to the weighting vector composed of wrþ preference structure that leads to a given target.
and wi , given to variables j and f. According to Kornbluth,23 a solution x* is efficient in
We can easily show that the MORO model, presented in relation to any set of weights y that satisfy the Kuhn–Tucker
expressions (20)–(23), including dominance (27) and (28) conditions. Each set of weights y characterizes one
may be put in the form of a preference structure model using indifference region: the interval where the weights y can
(30) and substituting wrþ and w i for y: vary and maintain the efficient solution x*.
MP Estellita Lins et al—Data envelopment analysis 1097

Table 5 MORO model (VRS)


Actual value Objective function value Target value

DMU Y X Max f Min j Y X

C 6 6 2.5 0.5 15 3
2 0.333333 12 2
1.166667 0.166667 7 1

The indifference regions can be determined by imposing θ1=0 0.16667 0.25 θ1=1
the optimality conditions on the coefficient of DMU j in the
objective function y*Cx, under the canonical form, that is,
Sol.3 Sol.2 Sol.1
zjcjX0. Once y*Cx is the sum of q objective functions, the A
F E
coefficient for each DMU j is
      Figure 5 Indifference regions.
y1 z1j  c1j þ y2 z2j  c2j þ þ yq zqj  cqj

where the optimality condition for a given non-dominated


solution is solutions. For the second solution, corresponding to DMU
      A, we obtained: 0.166667py1p0.25, and for the third
y1 z1j  c1j þ y2 z2j  c2j þ þ yq zqj  cqj X0 solution (DMU F): 0py1p0.16667. Then, imposing the
P optimality conditions, we can represent the indifference
and yiX0, i ¼ 1,y,q and qi ¼ 1yi ¼ 1.
regions of weights y1 (and y2 ¼ 1–y1) for the three DEA
We will implement these conditions to find the indiffer-
extreme-efficient solutions 1 (DMU E), 2 (DMU A) and 3
ence regions of the following MORO VRS model, applied to
(DMU F), as shown in Figure 5.
DMU C, with data from Table 1:
It is worth noting that an efficient, but non-extreme-
Max f efficient, solution can be expressed as a convex linear
Min j combination of some extreme-efficient solutions (for exam-
Subject to ple: A and E in Figure 5), corresponding to contiguous
indifference regions, and can be attainable with a single
6f  12l1  15l2  6l3  9l4  15l5  7l6  4l7 ¼ 0
value of y1 ¼ 0.25 (and y2 ¼ 0.75). In this case, a preference
6j þ 2l1 þ 4l2 þ 6l3 þ 2l4 þ 3l5 þ l6 þ 2l7 ¼ 0 structure model would exhibit multiple optimal solutions,
which are not shown by usual LP software.
l1 þ l2 þ l3 þ l4 þ l5 þ l6 þ l7 ¼ 1 Establishing the following equivalence: w1þ
y1 e w 1
y2,

f; j; lj X0 we have a preference structure where w1þ X0.25 and


w1þ þ w 1 ¼ 1 will lead us to the extreme-efficient solution
This program yielded three non-dominated solutions, E. A preference structure where 0.16667pw1þ p0.25 and
shown in Table 5, corresponding to the DMUs E, A and F, w1þ þ w 1 ¼ 1 will lead us to the solution A, and, ultimately,
in Figure 4. to a preference structure where w1þ p0.16667 and
Imposing the optimality conditions, zjcjX0, we obtained w1þ þ w 1 ¼ 1 will lead us to the solution F.
the following inequalities system for the first non-dominated
solution (DMU E in Figure 4):
0:5y1  0:1667y2 X0 Interactive method using the MOLP model for target
optimization (MOTO)
0:1667y2 X0
Instead of the ratio, we can consider each objective function
1:5y1 þ 0:5y2 X0 as the projected value (target), that is, the ratio multiplied by
y1  0:16667y2 X0 the observed output or input. With this simple change in the
objective functions in MORO (20 and 21), we obtain the
1:3333y1  0:3333y2 X0 MOTO model:
max y1j0 f1 ð33Þ
1:8333y1  0:16667y2 X0

y1 þ y2 ¼ 1 ...

After solving this system, we obtained the region max ysj0 fs


0.25py1p1. Using the same procedure, we determined the
indifference regions for the other two non-dominated min x1j0 j1 ð34Þ
1098 Journal of the Operational Research Society Vol. 55, No. 10

... to notice that interactive methods not only show extreme


efficient points, but also non-extreme efficient points, that is,
min xmj0 jm interactive approaches can explore the whole non-dominated
S.t. set.
X
n The number of possible efficient solutions to be assessed
fr yrj0 ¼ yrj lj 8r ¼ 1; . . . ; s ð35Þ is, however, unlimited. This method should, therefore, be
j¼1 used in a complementary way to the simplex multi-objective
X
n approach.
ji xij0 ¼ xij lj 8r ¼ 1; . . . ; m ð36Þ Let us look at an example using the MOTO–VRS model
j¼1 to analyse the efficiency of DMU C (previous example). In
MOTO model presents: Figure 3, we observed three possible targets for C, the
extreme efficient points (DMUs) A, E and F. The points
The same variables and restrictions as MORO model. between them, inside the segments A–E and A–F are also
The possibility of imposing restrictions on f and j, as efficient and can be explored using this interactive approach.
presented in (27) and (28) for MORO-D, resulting in the Figure 6 shows the interface of VIG, with an efficient
MOTO-D model, under CRS or VRS. solution chosen arbitrarily as a target for DMU C.
If f*r ¼ j*i ¼ 1 8r, i, then DMU j0 is a Pareto-efficient This solution corresponds to a virtual DMU, obtained by
DMU. Otherwise, the targets are: convex linear combination of DMU A (lA ¼ 0.9177545) and
DMU E (lE ¼ 0.0822455), which is the target for DMU C:
ynrj0 ¼ fr yrj0 ; r ¼ 1 . . . s
Y ¼ 12.2467 and X ¼ 2.08225 as seen in Figure 6 (in the
MORO model the solutions visualized for this same target
xnij0 ¼ ji xij0 ; i ¼ 1; . . . ; m
would be f ¼ 2.04112 and j ¼ 0.347042).

Visualization of the target for each variable, instead of the


value of the increment or reduction (f or j) given by the
The case study
MORO model, becomes more informative and understand-
able, improving the application of interactive solution Dental, as well as other branches of medical assistance, is a
methods. One of them was applied in this paper: the Pareto right guaranteed by the Brazilian Constitution. Thus, the
Race,26 whose metric is based on a distance to a reference or Public Health System is supposed to provide free preventive
ideal point, implemented in the VIG software, where the and conclusive medical care for every Brazilian citizen.
visualization of the objective function values is made Although the Public Health System performs as a unified
through bar charts on the computer screen. system, the way it is managed varies according to the region.
A change proposed by the expert/decision-maker in an In Rio de Janeiro city, the Federal Government supports
objective function, and its consequences for other objective public hospitals and clinics, which are managed by the local
functions are easily visualized. With this approach, one can administration. Both Federal Government and local man-
know directly from the screen which values (input/output) agers are interested in the assessment of the efficiency of the
can be obtained, given the changes proposed. It is important local units—geographical districts that constitute the muni-

Figure 6 Interactive method for MOTO-VRS to DMU C.


MP Estellita Lins et al—Data envelopment analysis 1099

Figure 7 Geographical regions of Rio de Janeiro municipality.

cipality, shown on the map (Figure 7). These efficiencies 2). Target 3 implies reducing the target population, which is
were assessed through the MORO-D model. not so straightforward, but could mean arranging for local
The final formulation considered the number of dentists population to be assigned to clinics in another district. The
and potential target demand (which comprises both the three targets can also be expressed as the following convex
number of children and pregnant women belonging to low- linear combination of the inputs/ outputs from the DMUs
income families) as inputs and the total number of (IV) Botafogo, (V) Copacabana and (XXI) Paquetá.
preventive plus conclusive dental procedures as the output,
25 7
according to data shown in Table 6.

The following DMUs revealed a 100% efficiency using the 5667  0; 2377 þ 2612  0; 2284

BCC Output oriented model: (III) Rio Comprido, (IV) 83629 22477
Botafogo, (V) Copacabana, (IX) Vila Isabel, (XIX) Santa
5 10; 21

Cruz, (XX) Ilha do Governador, (XXI) Paquetá, (XXIII)
þ 205  0; 5339 ¼ 2053
Santa Tereza, (XXV) Pavuna and (XXVI) Guaratiba.
3507 26884
The number of possible targets for each inefficient DMU
according to MORO-D/VRS formulation varies, depending
25 5 11; 767
on the enveloping frontier and on the DMU input–output
5667  0; 3383 þ 205  0; 6617 ¼ 2053
vector itself. The Multi-objective Simplex algorithm based
83629 3507 30615
on Guerreiro27 was implemented, resulting in a number of

targets for the inefficient units, ranging from three (Portuária 25 5 10; 83

and Centro) to 19 (Penha). These targets, apart from any 5667  0; 2918 þ 205  0; 7082 ¼ 1798

further analysis, are already a useful tool for the Decision- 83629 3507 26884
Makers. Table 6 also shows the number of targets for each
unit. Notice that efficient DMUs still project only on It is possible, therefore, to explore non-basic feasible
themselves. solutions, as any convex linear combination of the targets
Take, for instance, the targets for DMU (II) Centro 1–3.
shown in Table 7: We can note that any other target aiming to increase the
These three targets belong to the efficient frontier. The number of procedures (as an output oriented model would
DM could choose to dismiss/relocate 11 dentists and keep try to do) will also decrease the number of dentists (once the
production constant (target 1), or relocate nearly nine new target should be expressed itself as a convex linear
dentists and increase production to 30 615 procedures (target combination of the three targets presented) in order to be
1100 Journal of the Operational Research Society Vol. 55, No. 10

Table 6 Original data and number of alternative targets


DMUs Inputs Output MORO-D Results

Districts Number of dentists Target population Preventive plus conclusive procedures No. of targets

I Portuária 6 5720 22 105 3


II Centro 21 2053 26 884 3
III Rio Comprido 17 7529 66 836 1
IV Botafogo 25 5667 83 629 1
V Copacabana 7 2612 22 477 1
VI Lagoa 31 8892 44 985 14
VII São Cristóvão 12 8604 42 101 9
VIII Tijuca 13 5705 35 818 9
IX Vila Isabel 38 7023 91 465 1
X Ramos 22 50 726 34 340 18
XI Penha 19 34 172 58 424 19
XII Inhaúma 21 13 320 63 294 14
XIII Méier 50 25 721 103 245 13
XIV Irajá 20 16 652 71 818 9
XV Madureira 25 35 332 92 136 10
XVI Jacarepaguá 46 42 185 101 144 13
XVII Bangu 44 81 238 170 336 12
XVIII Campo Grande 39 56 396 146 758 12
XIX Santa Cruz 46 49 542 253 913 1
XX Ilha do Governador 32 13 973 122 375 1
XXI Paquetá 5 205 3507 1
XXII Anchieta 10 19 521 47 231 4
XXIII Santa Tereza 6 3493 23 416 1
XXIV Barra da Tijuca 21 6293 30 488 14
XXV Pavuna 8 30 258 42 655 1
XXVI Guaratiba 11 15 181 59 033 1

Table 7 Alternative targets for DMU Centro


Variables No. of dentists Target population No. of procedures

Observed DMU II Centro 21 2053 26 884

Targets 1 10.21 2053 26 884


2 11.76 2053 30 615
3 10.83 1799 26 884

efficient. We recall, indeed, that oriented models in practice of procedures. All of them behave strictly as Pareto-efficient
frequently yield non-oriented projections. points on the frontier. So, it is up to the DM to decide which
Under the MORO model without dominance, the (basic one would be more adequate.
feasible solution) targets for DMU (II) Centro would
correspond to each one of the efficient DMUs. Even so,
this DMU could hardly avoid dentists being relocated,
Conclusions
unless a huge increase in the production and decrease in
target population were attained. This paper proposes the use of a multi-objective linear
Next, we will focus on the DMU (VI) Lagoa, which has 14 programming approach to determine non-radial projections
targets, according to MORO-D. We will select three from for inefficient DMUs onto the Pareto efficient frontier. We
among a group of six that present an unchanged target established the equivalence between the efficient solutions set
population. The first one implies reduction in the number of obtained by the multi-objective (MORO and MORO-D)
dentists from 31 to roughly nine, while also keeping the models and the optimal solutions set obtained in models
number of procedures constant. A second target is with preference structure (Thanassoulis and Dyson, and
characterized by 18 dentists, but with an increase of 62% Zhu). We pointed out the advantages presented by the
in the number of procedures. Yet a third target implies no proposed multi-objective models: they allow a posteriori
reduction in dentists, but an increase of 120% in the number preference incorporation and do not require a priori
MP Estellita Lins et al—Data envelopment analysis 1101

specifications of the weights given to each objective (input or 9 Banker RD, Charnes A and Cooper WW (1984). Some models
output improvement) and yield indifference regions that for estimating technical and scale inefficiencies in data envelop-
ment analysis. Mngt Sci 30: 1078–1092.
map weight ranges onto the target alternatives. 10 Wierzbicki A (1980). The use of reference objectives in multi-
For each set of weights, this method finds an efficient objective optimization. In: Fandel G and Gal T (eds). Multiple
point or tangential point of a family of segments y1 f1 Objective Decision Making, Theory and Application. Springer-
(x) þ y2 f2 (x) ¼ l to a polygon, in the case of two objectives. Verlag, New York.
In the generalized case, parallel hyperplanes will be 11 Halme M et al. (1998). A value efficiency approach to
incorporating preference information in data envelopment
tangential to a convex polyhedron set at an extreme-point.19 analysis. Mngt Sci 45: 103–115.
This is why the Thanassoulis and Dyson and Zhu preference 12 Joro T (1998). Models for identifying target units in data
structure models will always give at least one extreme- envelopment analysis: comparison and extension. IIASA, Interim
efficient point as a target. Report IR-98-055.
13 Li X-B and Reeves GR (1999). A multiple criteria approach to
We also proposed the use of an interactive approach
data envelopment analysis. Eur J Opl Res 115: 507–517.
(Pareto race) applied to solve the MOTO model, exploring 14 Clı́maco J, Antunes CH and Alves MJG (1996). Programac¸ão
the non-extreme efficient solutions. Linear Multiobjetivo: Métodos Interactivos, ‘Software’ e Apli-
We presented a case study where the MORO-D model cac¸ões. Faculdade de Economia da Universidade de Coimbra e
was implemented, illustrating the diversity of Pareto-efficient INESC.
15 Steuer RE (1986). Multiple Criteria Optimization: Theory,
solutions that could possibly be obtained. Computation, and Application. Krieger Publishing: Malabar, FL.
Finally, we consider that both approaches could rather be 16 Geoffrion AM (1968). Proper efficiency and the theory of vector
integrated, as the mapping of the weights onto the efficiency maximisation. J Math Anal Appl 22: 618.
regions can guide the DM on his/her interactive search for 17 Evans JP and Steuer RE (1973). A revised simplex method for
multiple objective programs. Math Programming 5: 54–72.
the ultimate target. After obtaining extreme points and
18 Zeleny M (1973). Compromise programming. In: Cochrane JL
indifference regions, the DM could explore non-extreme and Zeleny M (eds). Multiple Criteria Decision Making.
efficient solutions through interactive methods. University of South Carolina Press, Columbia, pp 262–301.
19 Ecker JG, Hegner NS and Kouada IA (1980). Generating all
maximal efficient faces for a multiple objective linear program.
J Optim Theory Appl 30: 353–381.
References 20 Armand P and Malivert C (1991). Determination of the efficient
set in multiobjective linear programming. J Optim Theory Appl
1 Farrell MJ (1957). The measurement of productive efficiency. 70: 467–489.
J Roy Stat Soc Ser A 120: 253–281. 21 Steuer RE (1992). ADBASE, Department of Management
2 Thanassoulis E and Dyson RG (1992). Estimating preferred Science and Information Technology, University of Georgia,
target input–output levels using data envelopment analysis. Eur Athens, GA.
J Opl Res 56: 80–97. 22 Frota Neto (2002). Implementac¸ão e Análise de Programac¸ão
3 Zhu J (1996). Data envelopment analysis with preference Multiobjetivo na Soluc¸ão de Problemas de Análise Envoltória de
structure. J Opl Res Soc 47: 136–150. Dados. M.Sc. dissertation, COPPE/UFRJ, Brazil.
4 Stewart TJ (1996). Relationships between data envelopment 23 Kornbluth JSH (1974). Duality, indifference and sensitivity
analysis and multicriteria decision analysis. J Opl Res Soc 47: analysis in MOLP. Opl Res Quart 25: 599–614.
654–665. 24 Cohon JL (1978). Multiobjective Programming and Planning.
5 Golany B (1988). An interactive MOLP procedure for the Academic Press: New York.
extension of DEA to effectiveness analysis. J Opl Re Soc 39: 25 Romero C (1993). Teorı́a de la Decisión Multicriterio:
725–734. Conceptos, técnicas y aplicaciones. Alianza Editorial: Madrid.
6 Kornbluth JSH (1991). Analysing policy effectiveness using 26 Korhonen P and Wallenius J (1988). A Pareto Race. Naval Res
cone restricted data envelopment analysis. J Opl Res Soc 42: Logist 35: 615–623.
1097–1104. 27 Guerreiro J et al. (1985). Programac¸ão Linear. McGraw-Hill:
7 Joro T, Korhonen P and Wallenius J (1998). Structural Portugal.
comparison of data envelopment analysis and multiple objective
linear programming. Mngt Sci 44: 962–970.
8 Charnes A, Cooper WW and Rhodes E (1978). Measuring Received September 2002;
efficiency of decision making unit. Eur J Opl Res 2: 429–444. accepted April 2004 after three revisions
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

You might also like