Vol. 6, No.

1, Feb 2003

WEB JOURNAL OF

Simulating Weights Restrictions in Data Envelopment Analysis by

the Subjective and Objective Integrated Approach

Chun Chu LIU

Printed by

Committee on China Research and Development

Faculty of Business Administration
The Chinese University of Hong Kong

http://www.baf.cuhk.edu.hk/ocrd/cmr.htm

Vol. 6, No.1, Feb 2003,

68

Simulating Weights Restrictions in Data Envelopment Analysis by

the Subjective and Objective Integrated Approach
Chun-Chu LIU
Department of International Business
Chang Jung Christian University
E-maillcc@mail.cju.edu.tw

Abstract
Data Envelopment Analysis (DEA) is a mathematical programming approach to accessing
relative efficiencies within a group of Decision Making Units (DMUs). An important outcome of
such an analysis is a set of virtual multipliers or weights accorded to each (input or output) factor
taken into account. Due to the factor weight flexibility of the DEA model, a DMU may assign
very low factor weight values to some of its unfavorable inputs or outputs and appear as efficient,
thus producing nonsensical results in some cases. This study proposes a subjective and objective
integrated approach to restricting weight flexibility in data envelopment analysis. The approach
incorporates subjective information provided by the Analytic Hierarchy Process (AHP) and
objective information to form a Data Envelopment Analysis. As far as we are aware, this
method is not available in the literature. An example is used to describe the applicability of the
proposed approach.
Key WordsData Envelopment Analysis (DEA), relative efficiency, weight restriction

69

Simulating Weights Restrictions in Data Envelopment Analysis by the Subjective and Objective Integrated
Approach

1. Introduction
Date envelopment analysis (DEA) was developed originally as a set of techniques for
measuring the relative efficiency of a set of decision-making units (DMUs). The essential theory
of the DEA model was originally formulated by Charnes, Copper and Rhodes (CCR, 1978).
CCR applied the optimization method of mathematical programming to generalize the Farrell
(1957) single-output singe-input technical-efficiency measure to the multiple-output
multiple-input. The past few years have seen very many publications that apply DEA, which has
thus become as new and popular management science tool for technical efficiency analysis of
DMUs in the pubic and private sectors. Complete flexibility of weights in DEA has long been
recognized frequently to lead to inappropriate estimates of technical efficiency. DMUs can
attribute low weights to certain inputs and outputs and thus effectively ignore them. A humbler
approach has been proposed to overcome the problems created by complete flexibility of weights
in DEA. Pedraja-Chaparro et al. (1997) discussed this issue in detail. See also Allen et al. (1997),
Pedraja-Chaparro et al. (1996), Thanassoulis (1995), Roll and Golany (1993), Roll et al. (1991),
Ali et al. (1991), Thompson et al. (1990), Wong et al. (1990) and Dyson (1988). Such
approaches have generally been based on estimates of sensible ranges of permissible values for
output and input weights. Thus, the means by which value judgments on the inputs and outputs
can be incorporated in DEA assessment is important issue. This study, proposes a decision
weight framework, which integrates objective and subjective information, yielding a more
accurate analysis.
This paper is structured as follows. Section 1 is the Introduction. Section 2 separately
describes the subjective and objective approaches. Section 3 proposes an integrated approach to
determine weights. Section 4 describes the application of the proposed approach. Section 5
concludes the paper.

2. Approaches to determine weights

Several approaches have been proposed to determine weights (Hwang, 1987, Saaty, 1980).
Most Majorities of them can be classified as subjective and objective approaches depending on
the information provided. The subjective approaches include the Analytic Hierarchy Process
(Saaty, 1980), Delphi method (Hwang, 1987), and weighted least square method (Chu, 1979) etc.
The objective approaches include Date Envelopment Analysis (Charnes, 1978), principal
component analysis (Fan, 1996), the entropy method (Hwang, 1981) and the multiple objective
programming model (Choo, 1985, and Fan, 1996) etc. Subjective approaches determine weights
that reflect subjective judgment, but those weights can be influenced by the DMUs. Objective
approaches determine weights by making use of mathematical models, but they neglect
subjective judgment.
This study combines both the subjective weight restriction method and the objective weight
restriction method to evaluate the twelve district garbage disposal teams for the City of
Kaohsiung in Taiwan.

Vol. 6, No.1, Feb 2003,

70

2.1 Subjective weight restriction method

Several types of subjective weight restrict methods (such as Analytic Hierarchy Process,
Delphi, and multiple criteria decision making) are currently used. These methods are
characterized by the subjective setting of weights in the evaluation index, by experts, based on
their own experience. Different scholars may give different weights and thus, subjectivity is
the major drawback. Remedial measures such as increasing the numbers of experts, properly
selecting experts, and so on, can diminish this drawback; however, subjectivity remains. The
advantage of the subjective weight restrict method is that experts can reasonably identify the
weight index that corresponds to the actual problems. Thus, despite the different placement of
weights on the index, the method can still determine the order of priority and avoid conflicts
between the reality and the index weights, as can occur in the objective weight restrict method.
This study uses AHP, which process is described as follows.
Thomas L. Saaty first proposed the Analytic Hierarchy Process in 1971, and over the past
few decades, due to research efforts of Saaty et al, an AHP can now be categorized as one of 31
types (Smith, 1989). Now, AHP is considered to be an efficient management tool for modern
enterprises.
The strongest function of AHP is to simplify a complicated system into a hierarchy of
processes, each including simple but essential elements. In short, the procedure affects the
incentives of each decision making point and the pairwise comparisons between the nominal
scales. After the process of quantification, a comparison matrix is established to obtain the
Eigenvector, representing the weight of each hierarchy, and the eigenvalue. From the above, the
corresponding strength and weakness of the individual pairwise comparison used as information
for decision-making. In addition, if factors of AHP are interrelated in many hierarchies, the
priority and then the connection are determined to obtain the combined weight of factors in the
lowest hierarchy. Combining the consistency indices in all the comparison matrices, provides
each consistency index and ratio to evaluate on the common recognition of the entire hierarchy.
2.2 Objective weight restriction method
Researchers have been working on objective weight restriction method (DEA, Gray
prediction, Composition analysis) to avoid the shortcomings of the subjective weight restriction
method. The primary data of the objective weight restrict method are the actual figures used in
the matrix for evaluation to avoid subjective sources and ensure the weights are objectively
given. Yet, sometimes, inevitably the subjective weight may correspond to fact. The least
important index could theoretically have the largest weighting and the most important index may
not be the case. Examples can be seen in many DEA analyses.
Accordingly, the subjective weight restriction method has its advantages, and the objective
method also has some advantages if the practical situation is neglected. In the real situation,
where weights are obtained through either the subjective or the objective method, the difference
between the methods tends to be ignored and, therefore, their reliability becomes doubtful.
This study concentrates on the advantage of the integration and objectification of the weight
restriction method to offer more reliable information for decision-making.

71

Simulating Weights Restrictions in Data Envelopment Analysis by the Subjective and Objective Integrated
Approach

2.3 Integrated subjective and objective weight restriction analysis

2.3.1 Logical thinking
Although the subjective weight restriction method tends to be subjective, it is more reliable
than the objective method in setting priorities. As a result, the subjective weight restriction
method to the logic for setting priorities when determining the weight index. The weight of
each index of should be identified through the analysis of relationship analysis by subjective and
objective restriction methods. If the priority of the weight is the same, then weight for
evaluation should be obtained by the objective weight restrict method; if the priority is different,
then different handling measures should be taken according to the importance of the index
criteria.

3. Integrating subjective and objective weight restriction methods

3.1 Basic assumption
Assume that ai is the subjective weight of I, and bi is the objective weight.
weight is given by the following equation.

The final

0 1
Wi = *ai+(1-)bi

The value of is the core of the analysis.

In multiple criteria decision-making, the relevant importance of each index is different.
According to the basic concepts of multiple criteria decision-making, decision makers can
categorize an index according to important, determined by various criteria, and may or may not
show their preference. If the evaluation indices are V1,V2,..Vn and the index criteria are R1,
R2,, Rk; then importance is in the order, R1>R2>Rk that R1 is more important than R2 and
R2 is more important than R3 and so on.
3.2 Analysis model
1. If there is no difference between the evaluation indexes, i.e. the indexes are located at the
level of the same significance, and the weight priority obtained from the objective weight
restriction and the priority obtained by the subjective weight restriction is identical, then
= 0. The weight calculated by the objective restriction method is used as the weight
of the index.
2. If there is no difference between the evaluation indexes, i.e. the indexes are located at the
level of the same significance, and the weight priority obtained from the objective weight
restriction and the priority obtained by the subjective weight restriction is not identical,
then=0.5. The weighted average of the weight obtained by the subjective and
objective weight restriction methods is used as the weight of the index, or the optimal
weight is obtained by trial and error until it is accepted by the decision maker.

Vol. 6, No.1, Feb 2003,

72

3. If each evaluation index is located at different level of significance, but the weight
priority obtained from the objective weight restriction and the weight priority obtained by
the subjective weight restriction is identical, then it illustrates that the more important
index has a larger weight. To eliminate the human subjective factor, only the weight
obtained by the objective weight restriction method is used as the weight of the index,
and then=0.
4. If each evaluation index is located at a different level of significance, the weight priority
obtained from the objective weight restriction and the weight priority obtained by the
subjective weight restriction is not identical, but the priorities of indexes according to the
significance level are identical, then=0.5.
The weighted average of the weight
obtained by the subjective and objective weight restriction methods is used as the weight
of the index, or the optimal weight is obtained by trial and error until it is accepted by the
decision maker.
5. If each evaluation index is located at a different level of significance, the weight priority
obtained from the objective weight restriction and the weight priority obtained by the
subjective weight restriction is not identical, but the priorities of indexes according to the
significance level are not identical, then it shows that the weight obtained from the
subjective weight restriction method is no longer calculated from the significance level of
the index itself. Therefore, the weight obtained by the objective weight restriction
method gives no reference value to the determination of the index weight, and only the
weight obtained by the subjective weight restriction method is used as the weight of the
index, and then=1.
6. Here is a special situation. If the value of one or K of the weight(s) obtained from the
objective weight restriction method is zero (commonly happens in the DEA analysis), it
shows that the attributive of each solution for such indexes is the same, and such indexes
will not affect the decision of the priority. Then, such kind of indexes should be
eliminated from the evaluation indexes. The indexes remaining should be processed
according to each of the aforementioned situations.
From the above analysis, if the weight priorities obtained from the subjective and objective
weight restriction methods are identical, then the weight obtained by the objective weight
restriction method is used as the final weight of each weight, effectively eliminating the
subjectivity of the index weight. If the priorities of the weights obtained from these two
methods according to the significance of the weight are not consistent, the weight obtained from
the subjective weight restriction method is used as the final weight for each index. It can
eliminate the mistake of the conflict between the weight determined by the objective weight
restriction method and the actual significance of the weight of the index.
A compromising
method is adopted when the situation is mediocre. Such analysis method integrates the
advantages of subjective and objective weight restriction methods, so that the weight determined
by the DEA method becomes more reasonable.

73

Simulating Weights Restrictions in Data Envelopment Analysis by the Subjective and Objective Integrated
Approach

4. Empirical Results
4.1 Steps for assessment
This research used the Kaohsiung Garbage Disposal Team as an example (Wu, & Liu, 1998)
to obtain the input and output DMU weights without restrictions from the CCR model. Table 1
indicates that the input and output for the weight value is zero. The zero weight value made no
contribution to the relevant efficiency. This also indicates problems from the above discussion,
whether the given weight of relevant efficiency is reasonable and acceptable? This researcher
therefore proposed a modified model (Liu, 1998) for considering the facts. The research set up
a comparison matrix based on weights reached by experts, scholars, environmental officials, and
the general public for calculating the weight priority for the input and output shown in Table 2.
Through CR (Consistency Ratio) and CI (Consistency Index) inspection, a satisfied conclusion
was obtained. (CI and CR 0.1). To probe further and make the result more reasonable, weights
evaluated from the objective weight restriction method are shown in Table 3. The evaluation
steps are as follows:
1. The input and output index priority decided by decision makers is the same as the input
index value (in the same criteria). The output is classified into two criteria. Criteria 1 is the
amount of cleaning and number of trucks and 2 is the service person and cleaning district.
2. According to the assumption in Item 6, the input and output indices are assumed to be zero
and then set up the priority for the rest.
3. The objective weight restrictive method (DEA) is adopted to obtain the input index weight,
because the input index satisfies the same criteria
4. The output index has been categorized into two groups and the priorities determined through
the subjective weight restrict method are R1 (cleaning amount and car time) and R2
(cleaning people time and district), which differ from the priorities determined through the
objective weight restrict method. Therefore, the weight is obtained according to the
aforementioned method.
5. The most appropriate weight is selected and calculated in the DEA model to obtain the
relevant efficient value.

Vol. 6, No.1, Feb 2003,

74

Table 1 Weights and Priority from DEA Analysis Model (CCR Model)
Car Time
Yencheng
Priority
Gushan
Priority
Tzuoying
Priority
Nanzi
Priority
East Sanmin
Priority
West Sanmin
Priority
Hsinhsing
Priority
Chienjin
Priority
Lingya
Priority
Chienchen
Priority
Chijin
Priority
Hsiaokang
Priority

.00096
(1)
.00520
(1)
.00035
(1)
0
.00034
(1)
.00038
(1)
.00052
(1)
.00078
(1)
.00025
(1)
.00025
(1)
0
0

Output
Cleaning
Service
Amount
People
0
.00001
(2)
0
0

Cleaning
Area

Input
Car

Human
Resource

.04044

Fuel
Consumption
.05150

.05

.00009
(2)
.00008
(3)
0

.05353

.00001
(2)
0

.00571

.02415

.00461

.00239

.00031
(1)
.00001
(2)
0

.03704

.00559

.01027

.03288

.04187

.02381

.02439

.00069
(1)
.00005
(2)

.00001
(2)
0

.15974

.00022
(1)

.00405

.01710

SourceAuthors' calculations

Table 2 Priorities of Inputs and Outputs

Cleaning
Amount
1

Output
Car Time Scale of
Service
Area
2
3

SourceAuthors' calculations

Service
People
Time
4

Car

Input
Fuel
Consumption

Human
Resource

75

Simulating Weights Restrictions in Data Envelopment Analysis by the Subjective and Objective Integrated
Approach

Table 3 Weight Obtained from Subjective and Objective Weight Restrict Method
Output
Cleaning Amount Service People
0.0001

Car Time
Yencheng
0.00096
Gusan
0.00520
Tzuoying
0.00035
0.00009
Nantzi
0.4905
East Sanmin 0.00034
West Sanmin 0.00038
Hsinhsing
0.00052
Chienjin
0.00078
Lingya
0.00025
Chienchen
0.00025
Chijin
0.00069
Hsiaokang
0.4905
Blank means the index has been removed.
Source: Authors' calculations

0.1015

Cleaning Area

0.2028
0.00001

0.00001
0.2028

4.2 Evaluation results

According to the above evaluation steps, a weight value of zero is obtained from the
original DEA analysis model, so the value must be removed before the priorities shown in Table
1 are established. Step 4 is used to identify the proper weights, shown in Table 3, and except
for Nantzi and the Hsiaokang District Team, index belongs to different criteria and the priority
differs from. Notably, the weight is not obtained from the index, itself. As a result, the weight
obtained from the objective weight restrict method does not serve as the reference value. Only
the subjective weight restrict method is used to obtained the weight of index, it will take=1.
The other district team, but the priorities obtained from both methods are the same, showing the
significance of the importance index. =0 is assumed to prevent influence by human factors.
Based on the final index for evaluation, Table 3 presents the output index, Table 1 presents the
input index and presents the final efficiency calculation in Table 4.

Vol. 6, No.1, Feb 2003,

76

Table 4 Priority of Efficiency for Subjective and Objective Weight Restrict Methods

Teams

Priority of Efficiency

Yencheng

10

Gushan

Tzaoying

Nanzi

East Sanmin

West Sanmin

Hsinhsing

Chienchin

Lingya

Chienchen

11

Chijing

12

Hsiaokang

Source: Authors' calculations

5. Conclusion
This study is based on the DEA model. Since Charnes, Cooper, and Rhodes proposed the
mathematical programming model in 1978, such analysis model has been improved and applied
by numerous researchers and becomes practical. The weight setting still has its shortcoming, but
basically the study on weight setting either adopts the absolute subjective weight restriction or
the objective weight restriction method to find the value of the weight of the evaluation index.
These two methods have their merits and demerits. Therefore, this study proposes a more
reasonable weight determination model (If the weight priorities obtained by the subjective and
objective weight restriction methods are identical, then the weight obtained from the objective
weight restriction method is used as the final weight of each index, which can effectively
eliminate the subjectivity of the weight of the index. If the priorities of the indexes according
to the level of significance of the weights obtained by these two methods are not consistent, the

77

Simulating Weights Restrictions in Data Envelopment Analysis by the Subjective and Objective Integrated
Approach

weight obtained by the subjective weight restriction method is used as the final weight of each
index, which can eliminate the mistake of the conflict between the weight determined by the
objective weight restriction method and the degree of actual significance of the index. If the
situation is mediocre, then a compromising method is adopted). In other words, a weight
determination model is integrated according to the advantages of the subjective and objective
methods (avoiding the subjectivity of the subjective weight restriction method and the bias
casued by the objective weight restriction method to provide a thinking direction for decision
makers.
Acknowledgment
The author would like to thank the National Science Council of the Republic of China for
financially supporting this research under Contract No. NSC90-2146-H-309-009-.

Reference
1. Allen, R., Athanassopoulos, R. G. Dyson, and E. Thanassoulis(1997), "Weights restrictions
and value judgments in Data Envelopment Analysis", Annual Operations Research, 73, pp.
13-34.
2. Ali, A.I., W.D. Cook and L.M. Seiford, Strict vs(1991), "Weak Ordinal Relations for
multipliers in Data Envelopment Analysis", Management Science, 37(6), pp. 733-738
3. Charnes, A., W.W. Cooper and E.Rhodes(1978), "Measuring the Efficiency of Decision
Making Units"European Journal of Operational Research, 12(6), pp. 429-444
4. Chu, A. T. W., R. E. Kalaba., Spingarn, K. (1979), "A comparison of two methods for
determining the weights of belonging to fuzzy sets", Journal of Optimisation Theory and
Application, 27, pp. 531-538
5. Dyson, R. G. and E. Thanssoulis(1988), "Reducing Weight Flexibility in Data Envelopment
Analysis", Journal of The Operational Research Society, 39(6), pp. 563-576
6. Fan, Z. P.(1996), "Complicated Multiple Attribute Decision Making: Theory and
Applications", Ph.D. Dissertation, North-eastern University, Shenyang, PRC,
7. Farrell, M. J. (1957), "The Measurement of productive Efficiency", Journal of the Royal
Statistical Society, Vol.120, Part 3pp. 253-281
8. Hwang, C. L., M. J. Lin.( 1987), "Group decision making under multiple criteria: method
and application", Spring Berlin
9. Smith J. P.( 1989), "Bibliographical Research on the Analytic Hierarchy Process (AHP)",
Socio-Econ. Plann. Sci., 23(3), pp. 161-167

Vol. 6, No.1, Feb 2003,

78

10. Hwang, C. L., K. Yoon(1981), Multiple Attribute Decision Making: Methods and
Applications, Springer, Berlin
11. Liu, C. C.(1998), " Public sector efficiency measurement Application of DEA and
AHP"Chinese Management Review, 1(2), pp. 1-12
12. Pedraja-Chaparro, Francisco(1996), "An assessment of the efficiency of Spanish Courts
using DEA" , Applied Economics, 21(28), pp. 1391-1403
13. Pedraja-Chaparro, F., Salinas-Jiminez, and P. Smith.(1997), "On the Role of Weight
Restrictions in Data Envelopment Analysis", Journal of Productivity Analysis, 8, pp.
215-230
14. Roll, Y. and B. Golany(1993), "Alternate Methods of Treating Factor Weights in DEA ",
OMEGA, 21(1), pp. 99-109
15. Roll, Y., W. D. Cook and B. Golany(1991), "Controlling Factor Weights in Data
Envelopment Analysis", IEEE Transactions Journal , 23, pp. 2-9
16. Saaty, T. L.(1977), "A scaling method for priorities in hierarchical structures", Journal of
Mathematical Psychology, 15, pp. 234-281
17. Thompson, R.G., L.N. Langemeier, C.T. Lee, E. Lee and R.M. Thrall(1990), "The Role of
Multiplier Bounds in Efficiency Analysis with Application to Kansas Farming", Journal of
Econometrics, 46, pp. 93-108
18. Wong, Y. H., and Beasley, J. E.(1990), "Restricting weight flexibility in Data Envelopment
Analysis", Journal of the Operational Research Society , 41, pp. 829-835
19. Wu. C. C., and Liu, C. C.(1998), "Applying DEA in analyzing the efficiency of the District
Garbage Disposal Team for the city of Kaoshiung", Sun Yat-Sen management Review, 6(3),
pp. 879-902