USING ANALYTIC HIERARCHY PROCESS AS A TOOL FOR RANKING OF

EFFICIENT UNITS IN DEA MODELS
1


Josef Jablonsky-
Department of Econometrics
University of Economics Prague
Prague, 13067 Czech Republic
E-mail: jablon@vse.cz


ABSTRACT

Traditional data envelopment analysis (DEA) models cannot discriminate among efficient decision
making units (DMUs) because all of them have maximum efficiency score 100%. The aim of the paper is
to present an original approach for ranking of efficient DMUs based on the analytic hierarchy process
(AHP) developed by T. Saaty. The approach runs in two basic steps. The first one is traditional DEA
analysis and specification of efficient DMUs. In the second step the AHP model is created with second
hierarchical level containing all ratios outputs/inputs. Their priorities are derived as average weights from
DEA analysis. Finally the DMUs are evaluated with respect to all criteria and their global priorities are
derived. The priorities generate complete ranking of DMUs. The proposed approach is illustrated on a
numerical example with real-world background. The results of the DEA/AHP model are compared with
other DEA ranking approaches.

Keywords: analytic hierarchy process, data envelopment analysis, efficiency, super-efficiency


1. Introduction
Data envelopment analysis (DEA) is a tool for evaluation and measuring the efficiency of a set of
decision making units (DMUs) that consume multiple inputs and produce multiple outputs. Efficiency
score which is one of the main information given by DEA models reflects efficiency of transformation of
multiple inputs into multiple outputs. In typical case, higher inputs influence the efficiency score in
a negative way and in the contrary higher outputs in a positive way. DEA models split the DMUs into two
groups – efficient and inefficient. The efficient units are those lying on the efficient frontier which is
estimated by the DEA model. Each DMU receives its efficiency score – efficient units 100% and
inefficient units (depending on the model used) lower that 100%. The inefficient units can be easily
ranked according to their efficiency score. The efficient units cannot be ranked in standard DEA models
as their efficiency score is equal to 100%. That is why many models have been formulated in order to
allow ranking of efficient units in DEA models. This stream in DEA research is widely developed since
1993 when (Andersen and Petersen, 1993) published their super-efficiency model. This group of models
assigns to efficient units the efficiency score higher than 100% which allows their ranking.

Super-efficiency models are based on measuring the distance of the evaluated unit DMU
q
from a new
efficient frontier given by the removal of this unit from the set of units. Except Andersen and Petersen
model several other super-efficiency models have been formulated up to the present. Tone’s model (Tone,
2002) is one of the most popular. Other researchers have used different mathematical modelling

1
The research is supported by the Grant Agency of the Czech Republic - project no. 402/ 09/0231, and by IGA
project F4/ 18/2011.
-
Corresponding author
Proceedings of the International Symposium on the Analytic Hierarchy Process 2011

2
principles for ranking of efficient units. One of the possible principles that can be used for this purpose is
the analytic hierarchy process (AHP). The AHP was used for complete ranking of DMUs by (Sinuany-
Stern et al., 2000) and (Jablonsky, 2007). An extensive review of ranking models in DEA is given in
(Adler et al., 2002).

The aim of the paper is to present an original procedure for ranking of DMUs in DEA models based on
combination of AHP and DEA principles. It is organized as follows. The next section contains basic
definitions and formulations of standard DEA models and presents several basic super-efficiency models.
Section 3 presents an original AHP model for ranking efficient units as mentioned above. Section 4
contains a numerical illustration of the presented model and its comparison with other ranking models.
The last section summarizes the results and identifies main directions for future research.

2. Data envelopment analysis models
Let us suppose that the set of DMUs contains n elements. The DMUs are evaluated by m inputs and r
outputs with input and output values x
ij
, i = 1, 2, …, m, j = 1, 2, …, n and y
kj
, k = 1, 2, …, r, j = 1, 2, …, n,
respectively. The efficiency of the DMU
q
can be expressed as the weighted sum of outputs divided by the
weighted sum of outputs with weights reflecting the importance of single inputs and outputs v
i
, i =
1, 2, …, m and u
k
, k = 1, 2, …, r as follows:
.
1
1
¿
¿
=
=
=
m
i
iq i
r
k
kq k
q
x v
y u
u (1)
Standard CCR input oriented DEA model formulated by Charnes, Cooper and Rhodes in 1978 consists in
maximization of efficiency score (1) of the DMU
q
subject to constraints that efficiency scores of all other
DMUs are lower or equal than 1. The linearized form of this model is as follows:
minimize
¿
=
=
m
i
iq i q
x v
1
u
subject to , 1
1
=
¿
=
r
k
kq k
y u (2)

. ,..., 2 , 1 , ,..., 2 , 1 , ,
, ,..., 2 , 1 , 0
1 1
m i r k v u
n j x v y u
i k
m
i
ij i
r
k
kj k
= = >
= s ÷
¿ ¿
= =
c

If the optimal value of the model (2) u
*
q
= 1 then the DMU
q
is CCR efficient and it is lying on the CCR
efficient frontier. u
*
q
> 1 shows that the DMU
q
is not CCR efficient – higher value indicates lower
efficiency in this case. This measure is often presented as its reciprocal value, i.e. 1/u
*
q
which is more
understandable for decision makers - the higher value is assigned to more efficient units. The model (2)
is often referenced as primal CCR output oriented model. Its dual form is sometimes more convenient
from computational point of views and its mathematical model is as follows:
maximize
q
u
subject to ,
1
iq
n
j
i j ij
x s x = +
¿
=
÷
ì i = 1,2,...,m, (3)
J. Jablonsky
3
,
1
kq q
n
j
k j kj
y s y u ì = ÷
¿
=
+
k = 1,2,...,r,
ì
j
≥ 0, j = 1,2,…,n,
where ì
j
, j = 1, 2, …, n are weights of DMUs, s
÷
i
, i = 1, 2, …, m, and s
+
k
, k = 1, 2, …, r are slack
(surplus) variables and u
q
is the efficiency score of the DMU
q
which expresses rate of improvement of
outputs in order this unit reaches the efficient frontier. There is a problem that all efficient units identified
by the model (2) or (3) have the same efficiency score u
q
= 1. In many cases might be important to have
a tool for a diversification and ranking efficient DMUs. That is why many models for classification of
efficient units in DEA based on different methodological concepts were formulated by several researchers
in the past years. The most important category of such models is represented by super-efficiency DEA
models. This class of models supposes removal of the evaluated unit from the set of DMUs and
measuring its distance from the new efficient frontier. In super-efficiency models the efficiency scores of
inefficient units remain unchanged (lower than 1 for input oriented mode ls and higher than 1 for output
oriented) but the efficiency score of efficient units may be higher (lower) than 1. The efficient units can
be simply ranked according to their super-efficiency scores. Among super-efficiency models Andersen
and Petersen model (AP model) and Tone’s SBM model (SBMT model) are the most often used. AP
model was formulated in (Andersen and Petersen, 1993). Its output oriented formulation (4) is very close
to the standard output oriented formulation of the CCR model (3). Only difference is that the weight of
the DMU
q
, i.e. ì
q
, is set to zero in this model. It causes that the DMU
q
is removed from the set of units
and the efficient frontier changes its shape after this removal. Super-efficiency score
AP
q
u measures the
distance of the evaluated DMU
q
from the new efficient frontier.

AP model was criticized with respect to its properties many times. That is why several other models were
formulated with motivation to improve stability and interpretation of given results. One of them is Tone’s
super-efficiency model presented in (Tone, 2002) which is a modification of his DEA SBM model. This
model removes the evaluated unit DMU
q
from the set of units and looks for a DMU* with inputs x
i
*, i =
1, 2, ..., m, and outputs y
k
*, k = 1, 2, ..., r, being SBM (and CCR) efficient after this removal. It is clear
that all inputs of the unit DMU* have to be greater or equal than inputs of the unit DMU
q
and all outputs
will be lower or equal comparing to outputs of DMU
q
. The super-efficiency measure
SBM
q
u is the distance
of units DMU
q
and DMU* in their input and output space. Mathematical formulation of the SBMT model
can be found e.g. in (Tone, 2002) or (Jablonsky, 2007). Tone’s model returns optimal objective value
greater or equal 1. The optimal efficient score is greater than 1 for efficient DMUs – higher value is
assigned to more efficient units. All the SBM inefficient units reach in the super SBM model optimal
score 1. That is why this model cannot be used for classification of inefficient units. The model has to be
used in two steps. The first step is applied to the entire set of units in order to identify efficient units and
classify inefficient units. The second step is the computation of the super-efficiency scores by means of
the super SBMT model.

3. Using AHP model for ranking of efficient units
AHP is a powerful tool for analysis of complex decision problems. AHP models organize decision
problems as a hierarchical structure with several levels. The first (topmost) level defines the main goal of
the decision problem and the last (lowest) level usually describes the decision alternatives (DMUs in our
case). The levels in between can contain secondary goals, criteria and sub-criteria of the decision
problem. Our aim is to use AHP model for evaluation and discrimination among efficient DMUs. That is
why the first level is the goal – evaluation of efficient units, the last level of the hierarchy contains DMUs
identified as efficient by appropriate DEA model. The evaluation criteria are inputs and outputs used in
DEA analysis but they usually cannot be used directly due to possible high differences among input
Proceedings of the International Symposium on the Analytic Hierarchy Process 2011

4
and/or output values. This problem can be solved by using all possible ratios output/input as decision
criteria in AHP model instead of using single inputs and outputs. The ratios can be explained as particular
efficiency characteristics. They are easily comparable and that is why AHP can be an ideal tool for
evaluation of DMUs using such characteristics. Figure 1 presents a simple hierarchy for evaluation of
efficient DMUs in accordance with the presented idea. The second level contains criteria of the
evaluation, i.e. particular efficiency characteristics. The last level of the hierarchy contains the DMUs
determined as efficient by a DEA model.






. . . . .



. . . . .
O
1
/ I
1

v
1

GOAL
Evaluation of efficient DMUs
DMU
1

w
1j

O
1
/ I
2

v
2

O
r
/ I
m

v
k

DMU
2

w
2j

DMU
n

w
nj


Figure 1 AHP model - evaluation of efficient DMUs.

The weights of the criteria (particular efficiency characteristics) v
k
, k = 1, 2, …, m.r, are derived by
pairwise comparisons in AHP models. This standard approach can be used too but we offer using an
alternative way which connects the AHP model with DEA model. We suggest using geometric mean of
appropriate input and output weights. They can be taken as:

- M1 - average weights of inputs and outputs of all DMUs given by a DEA model,
- M2 - average weights of inputs and outputs of all efficient DMUs identified by a DEA model,
- M3 – optimal weights of single efficient DMUs.

Using pairwise comparisons of elements on the last level of the hierarchy (efficient DMUs) preference
indices of DMUs with respect to the particular efficiency measures w
ij
, i = 1, 2, …, n, j = 1, 2, …, m.r,
are derived. Depending on the number of efficient DMUs either AHP with absolute or relative
measurement can be used. Global preference indices of all DMUs are given by a simple sum of
preference indices w
ij
:
. ,..., 2 , 1 , ) (
.
1
n i w DMU u
r m
j
ij i
= =
¿
=
(4)
Global preference indices (4) allow final ranking of efficient DMUs.

4. Numerical illustration
The models presented in the last two sections of the paper are illustrated on the set of 194 DMUs – bank
branches of one of the Czech commercial banks. The following three inputs and two outputs are used in
this study: I
1
– total operational costs in thousands of CZK per year, I
2
– the number of inhabitants within
the region of the branch, I
3
– the number of employees, O
1
– value of credits in millions of CZK, and
O
2
– the total number of accounts.

J. Jablonsky
5
By this set of characteristics a business activity of branches is measured. The standard envelopment DEA
model with constant returns to scale and output orientation was applied and totally 12 DMUs were
identified as efficient. Due to a limited space of the paper the data set is not given here and particular
efficiency characteristics for all efficient DMUs are presented in Table 1 only.

DMU O
1
/I
1
O
1
/I
2
O
1
/I
3
O
2
/I
1
O
2
/I
2
O
2
/I
3

26 0.902 2.006 18.380 0.158 0.352 3227.429
28 0.813 1.743 29.720 0.143 0.307 5242.800
37 0.942 0.424 33.325 0.134 0.060 4739.500
71 0.577 3.561 16.884 0.133 0.822 3896.286
79 0.520 5.722 22.008 0.095 1.046 4021.846
82 0.461 9.472 14.406 0.101 2.085 3170.800
83 0.546 5.715 14.490 0.124 1.295 3282.800
105 0.811 2.324 18.465 0.149 0.427 3390.000
133 0.679 6.514 18.979 0.116 1.113 3241.333
147 0.377 40.369 10.779 0.074 7.893 2107.333
182 0.800 4.624 28.764 0.117 0.674 4189.714
184 0.410 7.353 10.714 0.118 2.121 3091.000
Table 1 Particular efficiency characteristics

The DEA/AHP model can be applied in several steps:

1. Application of a standard DEA model and identification of efficient units (CCR output oriented
model in our case).
2. Modification of the original data set for efficient units – criterion values are particular efficiency
measures. The efficient DMUs are evaluated by 2.3 = 6 criteria in our example (Table 1).
3. Using AHP model with absolute or relative measurement. The absolute measurement consists in
assigning the evaluated DMUs into elements of the evaluation scale. The relative measurement is a
standard pairwise comparison approach. In our example we use absolute measurement with five
elements evaluation scale: excellent, very good, good, poor and very poor. By pairwise comparisons
of evaluation scale’s elements their relative strength p
i
, i = 1, 2, …, 5 is derived.

AP model Super SBMT DEA/AHP M1 DEA/AHP M2
DMU θ
q
AP
Rank θ
q
SBM
Rank u() Rank u() Rank
26 1.103 6 1.035 9 0.213 4 0.176 7
28 1.227 2 1.100 3 0.243 3 0.219 3
37 1.153 3 1.069 4 0.197 5 0.176 8
71 1.067 9 1.033 10 0.164 10 0.154 11
79 1.093 7 1.040 7 0.173 9 0.190 6
82 1.088 8 1.054 6 0.243 2 0.276 2
83 1.020 11 1.008 11 0.156 12 0.164 10
105 1.005 12 1.003 12 0.178 7 0.147 12
133 1.066 10 1.037 8 0.161 11 0.170 9
147 4.274 1 2.043 1 0.311 1 0.350 1
182 1.134 4 1.112 2 0.177 8 0.193 5
184 1.129 5 1.058 5 0.194 6 0.205 4
Table 2 Ranking of efficient DMUs by different models
Proceedings of the International Symposium on the Analytic Hierarchy Process 2011

6
4. The DMUs are evaluated according to the criteria (particular efficiency characteristics) by using the
elements of the evaluation scale.
5. Using the weights of the criteria global preference indices of DMUs are computed. The following
two sets of weights are used in our illustration:
M1 - average weights of inputs and outputs of all DMUs; the vector of weights in our example is as
follows: v = (0.1023, 0.2321, 0.0679, 0.1519, 0.3448, 0.1009);
M2 – average weights of inputs and outputs of efficient DMUs only;
v = (0.0702, 0.3013, 0.0863, 0.0831, 0.3568, 0.1022).
The results, i.e. the global preference indices of all efficient DMUs including their ranking, are
presented in Table 2 together with results of two mentioned super-efficiency DEA models.

It is clear that both super-efficiency models (AP, SBMT) lead to very close rankings. All models identify
the unit 147 as the most efficient. Rankings on other places more or less vary but a more detailed analysis
is not subject of this paper. Correlation coefficients of rankings are presented in Table 3.

SBMT DEA M1 DEA M2
AP model 0.916 0.671 0.748
SBMT 0.587 0.825
DEA M1 0.741
Table 3 Correlation coefficients of rankings given by different models

5. Conclusions
The aim of the paper was to present the AHP model for ranking of efficient units in DEA models. The
results given by this model were compared with other two standard super-efficiency DEA models. The
paper does not contain a more detailed analysis of differences in rankings given by presented models. It is
an interesting task which can be taken as a starting point for a future research. It can be concentrated on
comparison of ranking models with randomly generated data sets of different size and under assumption
of different returns to scales.

REFERENCES

Adler, N., Friedman, L.,& Sinuany-Stern, Z. (2002). Review of ranking methods in data envelopment
analysis. European Journal of Operational Research, 140(2), 249-265.
Andersen, P.,& Petersen, N.C. (1993). A procedure for ranking efficient units in data envelopment
analysis. Management Science, 39(10), 1261-1264.
Cooper, W.W., Seiford, L.M.,& Tone, K. (2000). Data Envelopment Analysis. Kluwer Publ., Boston.
Jablonsky, J. (2004). Models for evaluation of efficiency of production units. Politicka ekonomie, 2, 206-
220.
Jablonsky, J. (2007). Measuring the efficiency of production units by AHP models. Mathematical and
Computer Modelling, 46 (7-8), 1091–1098.
Saaty, T.L. (1990). The Analytic Hierarchy Process. RWS Publications, Pittsburgh.
Sinuany-Stern, Z., Mehrez, A.,& Hadad, Y. (2000). An AHP/DEA methodology for ranking decision
making units. International Transactions in Operational Research, 7(2), 109-124.
Tone, K. (2002). A slack-based measure of super-efficiency in data envelopment analysis. European
Journal of Operational Research, 143(1), 32-41.

r as follows: q   u k y kq k 1 m r  vi xiq i 1 . The AHP was used for complete ranking of DMUs by (SinuanyStern et al. 2. i = 1. This measure is often presented as its reciprocal value. m and u k. Its dual form is sometimes more convenient from computational point of views and its mathematical model is as follows: maximize subject to q  xij  j  si  xiq . (3) 2 . An extensive review of ranking models in DEA is given in (Adler et al. It is organized as follows. 2. …. The last section summarizes the results and identifies main directions for future research. n and ykj . n.the higher value is assigned to more efficient units.. 2. m. * q > 1 shows that the DMUq is not CCR efficient – higher value indicates lower efficiency in this case.2.. One of the possible principles that can be used for this purpose is the analytic hierarchy process (AHP). …. 2002). ….. The next section contains basic definitions and formulations of standard DEA models and presents several basic super-efficiency models. 2.. 2. r (2) j  1. i 1 m u k . 2000) and (Jablonsky. …. The aim of the paper is to present an original procedure for ranking of DMUs in DEA models based on combination of AHP and DEA principles..2. The model (2) is often referenced as primal CCR output oriented model. i.. 1/* q which is more understandable for decision makers ..2. j = 1. respectively..2. i  1.e.. The linearized form of this model is as follows: minimize subject to  q   vi xiq i 1 m  u k ykq  1. Section 3 presents an original AHP model for ranking efficient units as mentioned above. Data envelopment analysis models Let us suppose that the set of DMUs contains n elements...Proceedings of the International Symposium on the Analytic Hierarchy Pro cess 2011 principles for ranking of efficient units. The DMUs are evaluated by m inputs and r outputs with input and output values xij . …... Cooper and Rhodes in 1978 consists in maximization of efficiency score (1) of the DMU q subject to constraints that efficiency scores of all other DMUs are lower or equal than 1. n. ….. 2.  k 1 k 1 r u k y kj   vi xij  0. If the optimal value of the model (2) * q = 1 then the DMUq is CCR efficient and it is lying on the CCR efficient frontier.m. 2. k = 1. 2007).. j = 1. The efficiency of the DMU q can be expressed as the weighted sum of outputs divided by the weighted sum of outputs with weights reflecting the importance of single inputs and outputs vi . r . k  1. r. vi   .. Section 4 contains a numerical illustration of the presented model and its comparison with other ranking models. (1) Standard CCR input oriented DEA model formulated by Charnes. i = 1. m.. k = 1.. j 1 n i = 1.

n. 2002) or (Jablonsky. The efficient units can be simply ranked according to their super-efficiency scores. m. the last level of the hierarchy contains DMUs identified as efficient by appropriate DEA model. Only difference is that the weight of the DMUq . k = 1.. Its output oriented formulation (4) is very close to the standard output oriented formulation of the CCR model (3). AP model was formulated in (Andersen and Petersen. ….. i = 1. k = 1.….g. n are weights of DMUs. being SBM (and CCR) efficient after this removal.. 1993). 2. The first (topmost) level defines the main goal of the decision problem and the last (lowest) level usually describes the decision alternatives (DMUs in our case). The most important category of such models is represented by super-efficiency DEA models.J. and s+k . i = 1. j 1 n k = 1.. That is why the first level is the goal – evaluation of efficient units. q . 2002) which is a modification of his DEA SBM model. r are slack (surplus) variables and q is the efficiency score of the DMU q which expresses rate of improvement of outputs in order this unit reaches the efficient frontier. The second step is the computation of the super-efficiency scores by means of the super SBMT model. in (Tone. Jablonsky  ykj j  sk   q ykq .. 3. One of them is Tone’s super-efficiency model presented in (Tone... AP model was criticized with respect to its properties many times. .. 2. It is clear that all inputs of the unit DMU* have to be greater or equal than inputs of the unit DMU q and all outputs SBM will be lower or equal comparing to outputs of DMU q . criteria and sub-criteria of the decision problem. si . All the SBM inefficient units reach in the super SBM model optimal score 1.. Using AHP model for ranking of efficient units AHP is a powerful tool for analysis of complex decision problems.r. The super-efficiency measure  q is the distance of units DMUq and DMU* in their input and output space. 2. where j .. The first step is applied to the entire set of units in order to identify efficient units and classify inefficient units. This model removes the evaluated unit DMU q from the set of units and looks for a DMU* with inputs xi *. …. m. Mathematical formulation of the SBMT model can be found e. j ≥ 0. r. 2. Super-efficiency score  qAP measures the distance of the evaluated DMU q from the new efficient frontier. The optimal efficient score is greater than 1 for efficient DMUs – higher value is assigned to more efficient units. j = 1. i. Among super-efficiency models Andersen and Petersen model (AP model) and Tone’s SBM model (SBMT model) are the most often used. That is why many models for classification of efficient units in DEA based on different methodological concepts were formulated by several researchers in the past years. …. Tone’s model returns optimal objective value greater or equal 1. The model has to be used in two steps. The evaluation criteria are inputs and outputs used in DEA analysis but they usually cannot be used directly due to possible high differences among input 3 . The levels in between can contain secondary goals. 2007).2. It causes that the DMU q is removed from the set of units and the efficient frontier changes its shape after this removal. That is why several other models were formulated with motivation to improve stability and interpretation of given results. .2. In many cases might be important to have a tool for a diversification and ranking efficient DMUs. j = 1. This class of models supposes removal of the evaluated unit from the set of DMUs and measuring its distance from the new efficient frontier. That is why this model cannot be used for classification of inefficient units.e. AHP models organize decision problems as a hierarchical structure with several levels. and outputs yk*. In super-efficiency models the efficiency scores of inefficient units remain unchanged (lower than 1 for input oriented mode ls and higher than 1 for output oriented) but the efficiency score of efficient units may be higher (lower) than 1. There is a problem that all efficient units identified by the model (2) or (3) have the same efficiency score q = 1. 2. Our aim is to use AHP model for evaluation and discrimination among efficient DMUs. is set to zero in this model.

The ratios can be explained as particular efficiency characteristics... This standard approach can be used too but we offer using an alternative way which connects the AHP model with DEA model.evaluation of efficient DMUs. i  1. i. Or / Im vk DMU1 w1j DMU2 w2j . Numerical illustration The models presented in the last two sections of the paper are illustrated on the set of 194 DMUs – bank branches of one of the Czech commercial banks..2. We suggest using geometric mean of appropriate input and output weights. The last level of the hierarchy contains the DMUs determined as efficient by a DEA model.. …. are derived by pairwise comparisons in AHP models. 2.average weights of inputs and outputs of all efficient DMUs identified by a DEA model. 2. The weights of the criteria (particular efficiency characteristics) vk. k = 1.. This problem can be solved by using all possible ratios output/input as decision criteria in AHP model instead of using single inputs and outputs. Depending on the number of efficient DMUs either AHP with absolute or relative measurement can be used. Global preference indices of all DMUs are given by a simple sum of preference indices wij : u ( DMU i )   wij ....r.Proceedings of the International Symposium on the Analytic Hierarchy Pro cess 2011 and/or output values. M3 – optimal weights of single efficient DMUs... The second level contains criteria of the evaluation. I3 – the number of employees. …. They are easily comparable and that is why AHP can be an ideal tool for evaluation of DMUs using such characteristics. i = 1. j = 1. j 1 m. M2 . are derived. They can be taken as:    M1 . ….e. 4 . m.. DMUn wnj Figure 1 AHP model . Using pairwise comparisons of elements on the last level of the hierarchy (efficient DMUs) preference indices of DMUs with respect to the particular efficiency measures wij . Figure 1 presents a simple hierarchy for evaluation of efficient DMUs in accordance with the presented idea. m. particular efficiency characteristics. The following three inputs and two outputs are used in this study: I1 – total operational costs in thousands of CZK per year. 4.. GOAL Evaluation of efficient DMUs O1 / I1 v1 O1 / I2 v2 . O1 – value of credits in millions of CZK.r (4) Global preference indices (4) allow final ranking of efficient DMUs. and O2 – the total number of accounts. I2 – the number of inhabitants within the region of the branch. 2.n.average weights of inputs and outputs of all DMUs given by a DEA model.r. n.

243 2 82 1.093 7 1.779 147 0.154 11 0.176 7 0.353 10.046 2.276 2 0.060 0. i = 1.333 4189.069 4 0. DEA/AHP M2 u() Rank 0.3 = 6 criteria in our example (Table 1).846 3170.369 10.147 12 0.500 3896.103 6 1. 3.227 2 1.054 6 0.008 79 0.561 16. 5 is derived.324 18.067 9 1.020 11 1.377 40.190 6 0.490 83 0.000 AP model Super SBMT DEA/AHP M1 θq AP Rank θq SBM Rank u() Rank DMU 1.714 3091.514 18. The standard envelopment DEA model with constant returns to scale and output orientation was applied and totally 12 DMUs were identified as efficient.472 14. In our example we use absolute measurement with five elements evaluation scale: excellent. 2.112 2 0.194 6 184 Table 2 Ranking of efficient DMUs by different models 5 .546 5.134 4 1. The efficient DMUs are evaluated by 2. poor and very poor.465 105 0.822 1.134 0.813 1.352 0.424 33.902 2.085 1.101 0.893 0.520 5.153 3 1.286 4021.003 12 0.040 7 0. Application of a standard DEA model and identification of efficient units (CCR output oriented model in our case).164 10 71 1.884 71 0.008 11 0.800 3390.074 0.158 0. good.429 5242.219 3 0.577 3.800 3282.066 10 1.406 82 0.100 3 0.427 1.149 0. Jablonsky By this set of characteristics a business activity of branches is measured.121 O2 /I3 3227.800 4739. Modification of the original data set for efficient units – criterion values are particular efficiency measures.720 28 0.274 1 2.178 7 105 1.088 8 1.979 133 0. The absolute measurement consists in assigning the evaluated DMUs into elements of the evaluation scale.161 11 133 4.380 26 0.095 0.197 5 37 1.295 0.811 2.722 22.058 5 0.410 7.156 12 83 1.764 182 0.006 18. By pairwise comparisons of evaluation scale’s elements their relative strength p i . The relative measurement is a standard pairwise comparison approach.461 9.000 3241.243 3 28 1.113 7.679 6. very good. Due to a limited space of the paper the data set is not given here and particular efficiency characteristics for all efficient DMUs are presented in Table 1 only.325 37 0.129 5 1. Using AHP model with absolute or relative measurement.193 5 0.205 4 O2 /I1 0.307 0.743 29. DMU O1 /I1 O1 /I2 O1 /I3 0.037 8 0.035 9 0.173 9 79 1.116 0.143 0.164 10 0.005 12 1.942 0.674 2.311 1 147 1.117 0.624 28. 2.124 0.800 4.J.176 8 0. ….043 1 0.170 9 0.350 1 0.714 184 Table 1 Particular efficiency characteristics The DEA/AHP model can be applied in several steps: 1.177 8 182 1.133 0.213 4 26 1.033 10 0.333 2107.118 O2 /I2 0.715 14.

Mathematical and Computer Modelling. Jablonsky.. 0. K.3448. Cooper. the global preference indices of all efficient DMUs including their ra nking.& Sinuany-Stern. L. W. K. Data Envelopment Analysis. 0.e.748 SBMT 0. Measuring the efficiency of production units by AHP models. The DMUs are evaluated according to the criteria (particular efficiency characteristics) by using the elements of the evaluation scale. Rankings on other places more or less vary but a more detailed analysis is not subject of this paper. 143(1). The Analytic Hierarchy Process. Saaty. Friedman. REFERENCES Adler. Boston. It is clear that both super-efficiency models (AP. (2000).3013. (1993). It is an interesting task which can be taken as a starting point for a future research. SBMT DEA M1 DEA M2 AP model 0. 6 . (2004)..Proceedings of the International Symposium on the Analytic Hierarchy Pro cess 2011 4.M. Management Science. 5. (2007). 206220.& Tone. 0. Z. (2002).1022).587 0. All models identify the unit 147 as the most efficient. Sinuany-Stern. 249-265. 46 (7-8). A slack-based measure of super-efficiency in data envelopment analysis. Kluwer Publ. 2. the vector of weights in our example is as follows: v = (0. 0.671 0. Andersen.825 DEA M1 0. SBMT) lead to very close rankings.. Pittsburgh.1009). P. J.W. (2002).0863. i. T.. Tone. Seiford. Politicka ekonomie. (2000). 0.741 Table 3 Correlation coefficients of rankings given by different models 5. v = (0. M2 – average weights of inputs and outputs of efficient DMUs only.& Hadad. A.L. Mehrez. N. L.1519. An AHP/DEA methodology for ranking decision making units.3568. J.. 0. 0. International Transactions in Operational Research. 7(2). Models for evaluation of efficiency of production units. The results. 140(2). European Journal of Operational Research.. The paper does not contain a more detailed analysis of differences in rankings given by presented models. 0. 32-41.0702. Using the weights of the criteria global preference indices of DMUs are computed. Conclusions The aim of the paper was to present the AHP model for ranking of efficient units in DEA models.. European Journal of Operational Research. A procedure for ranking efficient units in data envelopment analysis. Y. The results given by this model were compared with other two standard super-efficiency DEA models. It can be concentrated on comparison of ranking models with randomly generated data sets of different size and under assumption of different returns to scales. 1091–1098. 1261-1264.916 0. Review of ranking methods in data envelopment analysis. N. Jablonsky. 0.& Petersen.0831. Correlation coefficients of rankings are presented in Table 3.0679. Z. 109-124. 0.1023. (1990). The following two sets of weights are used in our illustration: M1 . 39(10).C.2321..average weights of inputs and outputs of all DMUs. RWS Publications. are presented in Table 2 together with results of two mentioned super-efficiency DEA models.