Sborník příspěvků XVII. mezinárodní kolokvium o regionálních vědách Hustopeče 18.–20. 6. 2014
45
DOI: 10.5817/CZ.MUNI.P210-6840-2014-4
NUTS 2 REGIONS CLASSIFICATION: COMPARISON OF
CLUSTER ANALYSIS AND DEA METHOD
KLASIFIKACE REGIONŮ NUTS 2: POROVNÁNÍ SHLUKOVÉ
ANALÝZY A METODY DEA
ING. LUKÁŠ MELECKÝ
ING. MICHAELA STANÍČKOVÁ
Katedra evropské integrace
Ekonomická fakulta
Vysoká škola báňská - Technická univerzita Ostrava
Department of European Integration
Faculty of Economics
VŠB – Technical University of Ostrava
* Sokolská třída 33, 701 21 Ostrava, Czech Republic
E-mail: lukas.melecky@vsb.cz, michaela.stanickova@vsb.cz
Annotation
The paper deals with the problem of searching the efficient frontier and group of similar units using
two methods – multivariate Cluster Analysis and multicriteria Data Envelopment Analysis. The main
aim of the paper is to propose the way how to choose an optimal number of groups coming to
empirical analysis. Firstly it´s necessary to find the closest characteristics for a given unit according
to a previously specified criterion of similarity. After fulfilment this criterion and creating optimal
groups of similar units, it´s possible to evaluate level of efficiency of homogenous NUTS 2 regions
within “new” EU Member States based on Regional Competitiveness Index approach; efficiency is
thus seen as a mirror of competitiveness. The idea laying behind this approach is that inefficient
regions can learn more easily from those, that are more similar; and closer similarity may show for
the inefficient regions how to achieve the improvement with less effort.
Key words
classification, cluster, cluster analysis, DEA method, efficient frontier, NUTS 2 region, RCI
Anotace
Příspěvek se zabývá problematikou hledání efektivní hranice a skupiny podobných jednotek za pomocí
dvou metod – vícerozměrné shlukové analýzy a vícekriteriální metody analýzy obalu dat. Hlavním
cílem tohoto příspěvku je navrhnout způsob, jakým vybrat optimální počet skupin vstupujících do
empirické analýzy. Za prvé je nutné najít nejbližší charakteristiky pro danou jednotku podle předem
zadaného kritéria podobnosti. Po splnění tohoto kritéria a vytvoření optimálních skupin podobných
jednotek, je možné vyhodnotit míru efektivity homogenních regionů NUTS 2 v "nových" členských
státech EU na základě přístupu Indexu regionální konkurenceschopnosti; efektivita je tak tedy
chápána jako zrcadlo konkurenceschopnosti. Myšlenka, která stojí za tímto přístupem, spočívá ve
faktu, že neefektivní regiony se mohou mnohem snadněji učit od těch, které jsou jim podobné; a blízká
podobnost může ukázat neefektivním regionům jak dosáhnout zlepšení s menším úsilím.
Klíčová slova
efektivní hranice, klasifikace, metoda DEA, NUTS 2 region, RCI, shluk, shluková analýza
JEL classification: C10, C67, R11, R13
Sborník příspěvků XVII. mezinárodní kolokvium o regionálních vědách Hustopeče 18.–20. 6. 2014
46
Introduction
Although the European Union (EU) is one of the most developed parts of the world with high living
standards, there exist huge economic, social and territorial disparities between its Member States, and
especially regions. These disparities have a negative impact on the balanced development across the
EU and weaken its competitiveness in the global context. Globalization, rapid technological change,
an ageing population and new knowledge economies are external factors which are becoming a
growing threat. But there are also some internal factors which are a big challenge for the EU, e.g. the
process of enlargement, and the EU thus needs to transform its economy and society. The EU
enlargement in years 2004, 2007 and 2013 is associated with increased regional disparities what a
threat for European competitiveness and internal cohesion is. Heterogeneity of the EU Member States
and especially regions is so high in many areas. The European integration process is thus guided by
striving for two different objectives: to foster economic competitiveness and to reduce disparities
(growing after EU enlargement history) (Molle, 2007). Europe’s economic challenge is to secure its
position in global markets facing intense challenges from its competitors, but firstly the EU has to
solve its internal problems in many areas and both in “old” and “new” EU Member States. The main
aim of the paper is to recognize optimal groups of similar units for subsequent relevant efficiency
analysis of homogenous NUTS 2 regions within “new” EU Member States based on Regional
Competitiveness Index 2013 (RCI) approach.
1. Importance of understanding the efficient frontier
To find the way how to choose an optimal number of groups coming to further empirical analysis,
firstly, it´s necessary to find the closest characteristics for a given unit according to a previously
specified criterion of similarity. Similarity can be interpreted as closeness between the inputs and
outputs of the assessed unit and the proposed targets, and this closeness can be measured by using
either different distance functions or different efficiency measures. Depending on how closeness is
measured, the paper solves the problem of searching the efficient frontier and group of similar units by
using two methods – multivariate Cluster Analysis (CA) and multicriteria Data Envelopment Analysis
(DEA). This approach should guarantee to reach the closest projection point on the Pareto-efficient
frontier. Thus, proposed way leads to the closest targets by means of a single-stage procedure, which
is easier to handle than those based on algorithms aimed at identifying all the facets of the efficient
frontier. Why are we applying the methods of CA and DEA in the paper? Because for fulfilment the
criterion of similarity, it´s necessary to create optimal groups of units, and after that, it´ll be possible
to evaluate level of efficiency of homogenous NUTS 2 regions of “new” EU Member States based on
RCI 2013 approach. For meaningful and relevant measuring efficiency by DEA method, it´s necessary
to note that DEA is a methodology for the assessment of relative efficiency of a set of homogenous
decision-making units (DMUs) that use several inputs to produce several outputs. DEA models
generally provide both efficiency measures for each of the assessed DMUs and information on the
peers and target that have been used in the efficiency assessment in the case of inefficient DMUs. The
most important mentioned point is that DMUs must be homogenous, i.e. mutually comparable in the
field of inputs and outputs. How is possible to recognize that group of DMUs are in fact homogenous?
Dividing the whole group of initial DMUs to smaller group of homogeneous units could be ensured
just by using of CA or DEA.
Cluster analysis represents a group of multivariate methods whose primary purpose is to cluster
objects based on the characteristics they possess. CA classifies objects that are very similar to others
in the cluster based on a set of selected characteristics. The resulting cluster of objects should exhibit
high internal (within-cluster) homogeneity and high external (between-cluster) heterogeneity (Hair,
Black, et al., 2009). Objects in a specific cluster share many characteristics, but are very dissimilar to
objects not belonging to the cluster. The aim of CA is to minimize variability within clusters and
maximize variability between clusters and from this point of view CA seems to be convenient as one
of the methods used for creating groups of units which are relevant to efficiency analysis. There is
several clustering procedure how to form the groups of objects; in the paper we used the hierarchical
Sborník příspěvků XVII. mezinárodní kolokvium o regionálních vědách Hustopeče 18.–20. 6. 2014
47
CA using the dissimilarities such as distances between objects when forming the clusters. The distance
is defined as Squared Euclidean distance suitable for categorical variable. After the determination of
the distance measure, the clustering algorithm has to be selected. The most frequently used method is
Ward’s method used in the paper. The last step of CA is interpretation of the results. The most
important is to select the cluster solution representing the best data sample.
DEA is multicriteria decision making method for evaluating efficiency of a homogenous group
(DMUs). The aim of DEA method is to examine DMU if they are efficient of inefficient by the size
and quantity of consumed resources and by the produced outputs. DEA can successfully separate
DMUs into two categories which called efficient DMUs and inefficient DMUs (Cook, Seiford, 2009).
Efficient DMUs have equivalent efficiency score. However, they don’t have necessarily the same
performance. DMU is efficient if the observed data correspond to testing whether the DMU is on the
imaginary ‘efficient frontier’. All other DMU are simply inefficient. For every inefficient DMU, DEA
identifies a set of corresponding efficient units that can be utilized as benchmarks for improvement.
But this improvement could be the best, we promotes the idea that inefficient DMUs can learn more
easily from those, that are more similar; and closer similarity may show for the inefficient DMUs how
to achieve the improvement with less effort. DEA also offers a possible way for creating similar
groups. Evaluated DMUs can be divided into groups-levels according to all efficient frontiers via
Context-Dependent DEA approach. By this stratification, into efficiency analysis we will enter more
homogenous groups of regions, which will be evaluated separately according to closer features.
The intent of frontier estimation is to deduce empirically the production function in the form of an
efficient frontier. That is, rather than knowing how to convert functionally inputs to outputs, these
methods take the inputs and outputs as given, map out the best performers, and produce a relative
notion of the efficiency of each. The problem with the existing methods is that they each measure
efficiency in a conceptually suspect, albeit computationally effective, way. If the DMUs are plotted in
their input/output space, then an efficient frontier that provides a tight envelope around all of the
DMUs can be determined. The main function of this envelope is to get as close as possible to each
DMU without passing by any others. A simple example of an efficient frontier is shown in Figure 1.
In utilizing projections from an inefficient DMU onto the efficient frontier, it is important to
understand the implications of where the projection lands. That is, it may well be that the projection
lands beyond any existing DMU or any convex combination of an existing DMU. If the efficient
frontier is split into two sections, one that represents either an observable or convex combination of
observed input-output combinations, and another that represents extrapolations of DMUs beyond
those defined in the first section, then we have what we determine as the observable portion and nonobservable
portion of the frontier, respectively. There are many instances in which it is not practical to
have a benchmark that extends beyond the scale observed by any existing DMUs (that is, to a nonobservable
point), and thus, independent projections onto the observable frontier are necessary. To
illustrate this, Fig. 1 duplicates the two-dimensional example from above with the observable portion
of the frontier represented by a thicker line than the non-observable portion. In addition, another
inefficient DMU has been added in order to illustrate that this DMUs shortest projection onto the
entire frontier lands at a non-observable point. That is, the projected output for D6 is 7.2 whereas the
largest observed output that of D4, is only 7. There may very well be instances in which the shortest
projection onto the observable frontier, in this case directly onto D4, is the best solution. Thus, it is
necessary to report both the overall efficiency score as well as the observable efficiency score. In the
case of inefficient DMUs is possible to define “peer-units” as identifications efficient targets and
range of improvements.
Sborník příspěvků XVII. mezinárodní kolokvium o regionálních vědách Hustopeče 18.–20. 6. 2014
48
Fig. 1: The Observable efficient frontier
Source: own elaboration, 2014
For solution of CA method we used software toll IBM SPSS Statistics 22 and for DEA method
software tool based on solving linear programming problems – Solver in MS Excel 2010, such as the
DEA Frontier 2011 is used in the paper.
2. Theoretical background of empirical analysis
Analysis starts with building a database of indicators, which are part of RCI 2013 approach. RCI 2013
approach covers a wide range of areas having impact for territorial competitiveness including
innovation, quality of institutions, infrastructure and measures of health and human capital (Annoni,
Dijkstra, 2013). RCI 2013 is thus divided into pillars according to the different input and output
dimensions of territorial competitiveness they describe. The terms ‘inputs’ and ‘outputs’ are meant to
classify pillars into those which describe driving forces of competitiveness, also in terms of long-term
potentiality, and those which are direct or indirect outcomes of a competitive society and economy
(Annoni, Kozovska, 2010). Initial variables coming into following analysis are competitiveness scores
of RCI 2013 and thus covered group of RCI 2013 input pillars and group of RCI 2013 output pillars.
RCI 2013 pillars on the side of inputs are Institutions, Infrastructure, Health, Higher Education and
Lifelong Learning, and Technological Readiness. RCI 2013 pillars on the side of outputs present
Labour Market Efficiency, Market Size, Business Sophistication, and Innovation. RCI 2013 scores for
DMUs are adjusted to positive values through Factor analysis and are shown in Tab. 3. In Tab. 1, the
main descriptive statistics of initial dataset (input and output pillars) are presented across all DMUs.
Tab. 1: Descriptive statistics of initial dataset
RCI 2013 pillars
N Range MIN MAX Sum Mean Std. Deviation Variance
Statistic Statistic Statistic Statistic Statistic Statistic Std. Error Statistic Statistic
I_Institutions 57 3.120 1.210 4.330 173.400 3.04211 .088500 .668158 .446
I_Infrastructure 57 1.230 2.750 3.980 177.710 3.11772 .040042 .302313 .091
I_Health 57 3.15 1.36 4.51 155.38 2.7260 .08933 .67446 .455
I_HigherEducation
LifelongLearning
57 2.55 2.14 4.69 187.55 3.2904 .07921 .59802 .358
I_Technological
Readiness
57 2.58 1.85 4.43 180.82 3.1723 .08723 .65853 .434
O_LaborMarket
Efficiency
57 1.55 2.96 4.51 206.70 3.6263 .05461 .41229 .170
O_MarketSize 57 1.66 2.43 4.09 171.92 3.0161 .05322 .40184 .161
O_Business
Sophistication
57 3.07 2.16 5.23 172.48 3.0260 .09567 .72233 .522
O_Innovation 57 2.52 2.51 5.03 187.60 3.2912 .07236 .54629 .298
Valid N (listwise) 57 / / / / / / / /
Source: own calculation and elaboration, 2014
Empirical analysis is applied to regional (NUTS 2) level within “new” EU Member States, i.e. 13
countries entered to the EU in years 2004, 2007 and 2013. EU13 countries cover 57 NUTS 2 regions –
Bulgaria 6 (BG), Cyprus 1 (CY), Czech Republic 7 (CZ)6
, Estonia 1 (EE), Croatia 2 (CR), Hungary 7
(HU), Lithuania 1 (LT), Latvia 1 (LV), Malta 1 (MT), Poland 16 (PL), Romania 8 (RO), Slovenia 2
6 The initial number of Czech NUTS 2 regions is 8; but in RCI 2013 approach, capital region CZ01 Prague is
merged with its neighbouring region CZ02 Central Bohemia.
Sborník příspěvků XVII. mezinárodní kolokvium o regionálních vědách Hustopeče 18.–20. 6. 2014
49
(SI) and Slovakia 4 (SK). The group of EU13 countries was chosen for empirical analysis because
after accession to the EU, these countries has focused attention with regard to their functioning in “EU
club of developed countries” in comparison with “old” EU Member States, i.e. EU15 countries as
previous research (Melecký, 2013) mentioned. “New” EU Member States have to align their
regulatory systems (legislation and related enforcement) with the internal Market acquis in order to be
able to fully participate in the Single internal market and thus to be a full and competitive member this
club. They must also have administrative capacity to implement the acquis. But accession to the EU is
neither a necessary nor a sufficient condition for economic growth. The combined effects of market
access and economic liberalization, not EU membership, optimize economic growth. From this point
of view, the group of EU13 countries is a homogenous unit of competitive countries, resp. regions
able to compete in comparison with EU15 countries and their regions? Does all NUTS 2 regions
within EU13 countries have the same conditions for success of the Single internal market?
3. Application of DEA for efficiency evaluation of EU13 NUTS 2 regions
In the initial phase of empirical analysis, CA was applied for finding group of similar units, based on
the value of competitiveness scores of RCI 2013 for each NUTS 2 region. The object is sorted into
clusters, so that the degree of association is strong between members of the same cluster and weak
between members of different clusters. To determine the optimum solution, in the paper is used the
most common approach – method of hierarchical cluster analysis and the clustering algorithm is
Ward’s method applying Squared Euclidean Distance as the distance or similarity measure. It helps to
obtain the optimum number of clusters we should work with. On the basis of the Ward Linkage –
Agglomeration schedule, the part “Coefficients” helped us to decide how many clusters are optimal
for representation of the data. The cluster formation should be stop when the increase in Coefficients
is large. In this case, the best interpretation of data ensures six-cluster solution in the case of RCI
2013. In following Tab. 2, it is possible to seen all obtain six clusters and membership of all evaluated
NUTS 2 regions within EU13 countries in relevant cluster; the total number of NUTS 2 region
belonging to each cluster is also marked in Tab. 2. According to membership of NUTS 2 regions, with
respect to national jurisdiction of each region, it is possible to say that the clusters seem to be created
by homogenous regions. Cluster I presents predominantly Czech NUTS 2 regions, but also by one
small region from Slovenia, Estonia and Malta. Cluster II is created by some NUTS 2 regions within
Visegrad Four (V4) countries, thus CZ, HU, PL, SK, and then one SI region. Croatian NUTS 2
regions together with one small region from Bulgaria and Romania belong to Cluster III. Cluster IV is
represented predominantly by Hungarian NUTS 2 regions, and by one small region from Latvia and
Lithuania. Cluster V is the largest cluster relative to the number of regions – 21 of all 57 regions
belong to this group. Cluster V is created by V4 regions, i.e. Polish regions are the most contained,
then Slovak, Hungarian and one Czech region. Finally, the second largest group of 12 regions, is
Cluster VI; this cluster is presented only by Romanian and Bulgarian NUTS 2 region. Graphical
representation of CA is Dendogram showing the boundaries of each cluster groups (see Appendix 1).
Tab. 2: Classification by CA – Obtain clusters of EU 13 NUTS 2 regions
Category of clusters
No. 1 No. 2 No. 3 No. 4 No. 5 No. 6
CZ03 HU10 HR03 HU32 PL31 PL62 RO21
CZ05 PL12 HR04 HU33 PL34 HU21 RO42
CZ06 CZ00 BG41 HU23 PL32 HU22 RO11
CZ07 SI02 RO32 HU31 PL33 PL11 BG31
CZ08 SK01
4 NUTS 2
regions
LT00 PL42 PL51 RO22
CY00
5 NUTS 2
regions
LV00 PL43 PL21 BG34
SI01
6 NUTS 2
regions
PL41 PL63 BG32
EE00 PL61 CZ04 BG42
MT00 PL52 SK02 RO12
9 NUTS 2
regions
SK03 PL22 RO41
SK04 RO31
21 NUTS 2 regions
BG33
12 NUTS 2
regions
Source: own calculation and elaboration, 2014
Sborník příspěvků XVII. mezinárodní kolokvium o regionálních vědách Hustopeče 18.–20. 6. 2014
50
In Context-Dependent DEA method, we used a basic approach in the form of constant returns to scale
(CRS) function for obtain levels generating all the efficient frontiers (see Tab. 3). With respect to
obtained results, it´s possible to say that optimal solution is presented by three groups/levels of units.
The object is sorted into homogenous groups of DMUs (regions) based on the size and quantity of
consumed resources and by the produced outputs. Via Obtain Levels function, a continuous
calculation in Context-Dependent DEA method, initial number of 57 NUTS 2 regions was divided in
three groups – 1st
group is created by 20 regions (predominantly by Bulgarian, Romanian and
Hungarian regions; to a lesser content by one region from Cyprus, Estonia, Croatia, Malta, Poland and
Slovenia); 2nd
group is presented by 25 regions (predominantly by NUTS 2 regions within V4; then by
one region from Bulgaria, Croatia, Latvia, Lithuania, Romania and Slovenia); 3rd
group is represented
by 12 regions (predominantly by Czech and Polish regions; then by one region from Bulgaria,
Hungary and Slovakia).
Tab. 3: Classification by DEA CRS model – Obtain efficiency levels of EU 13 NUTS 2 regions
No. DMU I1
I2
I3
I4
I5
O1
O2
O3
O4
Level 1 (CRS model) – No. of DMU 20
1 BG31 1.65 2.80 2.16 2.14 2.04 2.96 2.45 2.67 2.63
3 BG33 2.89 2.85 1.90 2.88 2.02 2.97 2.45 3.10 2.88
4 BG34 1.95 2.81 1.42 2.59 2.11 3.24 2.43 2.64 2.54
5 BG41 2.18 3.03 2.87 3.40 2.62 4.10 2.84 4.66 3.69
7 CY00 3.91 2.99 4.42 3.71 3.34 4.35 3.13 3.94 3.69
15 EE00 3.88 2.79 3.16 3.98 3.94 3.70 2.46 3.64 4.20
17 HR04 2.21 3.07 2.91 2.89 3.09 3.29 3.22 3.79 3.24
18 HU10 3.05 3.61 2.61 3.78 3.81 3.95 3.59 4.72 4.40
21 HU23 3.65 2.85 1.86 3.30 3.45 3.38 2.74 2.95 3.60
23 HU32 3.59 3.00 1.36 3.13 3.31 3.30 2.86 2.71 3.12
27 MT00 4.33 2.84 4.51 2.53 4.43 3.60 2.68 4.01 3.67
31 PL22 2.96 3.46 2.91 3.80 3.21 3.59 3.78 3.06 3.17
44 RO11 2.81 2.80 2.10 2.43 2.12 4.07 2.59 2.39 3.03
46 RO21 1.97 2.79 2.16 2.42 1.85 3.94 2.55 2.35 2.59
47 RO22 2.03 2.78 1.93 2.29 1.99 3.14 2.58 2.40 2.51
48 RO31 2.21 3.10 2.00 2.42 2.13 3.31 2.95 2.16 2.58
49 RO32 1.21 3.39 2.92 4.08 2.74 4.46 3.79 4.13 4.65
50 RO41 2.44 2.75 2.33 2.35 2.07 3.44 2.61 2.18 2.90
51 RO42 1.83 2.87 1.93 2.62 2.18 3.84 2.57 2.41 3.26
53 SI02 3.80 3.33 3.71 4.64 3.64 4.45 3.48 4.53 4.43
Average values 2,73 3.00 2.56 3.07 2.80 3.65 2.89 3.22 3.34
Level 2 (CRS model) – No. of DMU 25
2 BG32 1.91 2.79 2.24 2.92 2.23 3.07 2.58 2.60 2.82
13 CZ07 3.57 3.20 3.44 3.55 3.82 3.85 3.33 2.71 3.25
14 CZ08 3.67 3.26 3.38 3.85 3.97 3.56 3.55 2.72 2.92
16 HR03 1.95 2.90 2.89 2.88 3.22 3.36 2.74 3.74 3.05
19 HU21 3.65 3.45 2.07 3.54 3.61 3.72 3.24 2.74 3.38
22 HU31 3.59 3.10 1.84 3.31 3.40 3.18 3.04 2.51 3.55
24 HU33 3.59 3.07 1.78 3.02 3.42 3.67 2.84 2.75 3.24
25 LT00 3.10 2.88 2.09 3.52 3.63 3.39 2.69 2.89 3.45
26 LV00 3.17 2.94 2.19 3.34 3.11 3.24 2.45 3.53 3.22
28 PL11 3.18 3.21 2.57 3.24 3.24 3.82 3.27 2.88 3.16
29 PL12 3.02 3.46 2.84 4.01 3.24 4.23 3.49 4.16 4.00
30 PL21 3.12 3.28 3.21 3.41 3.21 3.69 3.38 2.84 3.70
32 PL31 3.16 2.85 2.86 3.42 3.09 3.72 2.80 2.46 3.14
34 PL33 3.17 2.95 2.70 3.40 3.09 3.25 3.10 2.24 2.71
35 PL34 3.06 2.76 2.99 3.22 3.09 3.77 2.66 2.45 2.91
36 PL41 2.96 3.14 2.82 2.98 3.30 3.48 3.14 2.53 3.04
38 PL43 3.05 3.22 2.55 2.94 3.30 3.77 2.93 2.62 3.01
39 PL51 2.90 3.19 2.64 3.45 3.32 3.81 3.23 3.04 3.22
42 PL62 3.28 2.84 2.70 2.75 3.20 3.32 2.69 2.62 2.89
43 PL63 3.21 3.03 2.97 3.37 3.20 3.78 2.96 3.03 3.42
45 RO12 2.51 2.85 2.20 2.57 2.07 3.18 2.62 2.45 2.67
52 SI01 3.80 3.22 3.64 4.22 3.64 4.07 3.23 3.49 3.51
54 SK01 3.44 3.98 3.52 4.69 3.79 4.51 4.09 5.23 5.03
56 SK03 3.26 2.97 2.87 3.19 3.61 3.16 3.16 2.86 3.03
57 SK04 3.26 2.92 2.66 2.93 3.52 2.97 2.94 2.97 2.93
Average values 3.14 3.10 2.71 3.35 3.29 3.58 3.05 2.96 3.25
Sborník příspěvků XVII. mezinárodní kolokvium o regionálních vědách Hustopeče 18.–20. 6. 2014
51
No. DMU I1
I2
I3
I4
I5
O1
O2
O3
O4
Level 3 (CRS model) – No. of DMU 12
6 BG42 2.84 2.91 2.36 2.63 2.19 3.10 2.52 2.49 2.64
8 CZ00 3.49 3.82 3.72 4.63 4.09 4.47 3.74 4.39 4.51
9 CZ03 3.92 3.51 3.33 3.89 3.92 4.10 3.18 2.77 3.39
10 CZ04 3.13 3.77 2.9 3.84 3.87 3.47 3.48 2.72 3.22
11 CZ05 3.90 3.43 3.59 3.95 4.06 3.98 3.33 2.70 3.43
12 CZ06 3.59 3.54 3.78 3.59 3.99 3.85 3.30 3.21 3.58
20 HU22 3.65 3.53 2.35 3.61 3.60 3.97 3.08 2.75 3.09
33 PL32 3.12 2.89 3.2 3.22 3.09 3.31 2.85 2.22 3.19
37 PL42 3.11 3.16 2.67 3.12 3.30 3.56 2.86 2.95 3.09
40 PL52 3.32 3.25 3.02 3.27 3.32 3.56 3.27 2.56 3.06
41 PL61 3.07 3.01 2.65 3.19 3.20 3.24 3.03 2.59 3.10
55 SK02 3.13 3.52 2.98 3.51 3.78 3.44 3.38 2.63 3.27
Average values 3.36 3.36 3.05 3.54 3.53 3.67 3.17 2.83 3.30
I1
Institutions, I2
Infrastructure, I3
Health, I4
Higher Education and Lifelong Learning, I5
Technological
ReadinessO1
Labour Market Efficiency, O2
Market Size, O3
Business Sophistication, O4
Innovation
Source: own calculation and elaboration, 2014
Conclusion
Measurement and evaluation of efficiency is an important issue for at least two reasons. One is that in
a group of units where only limited number of candidates can be selected, the efficiency of each must
be evaluated in a fair and consistent manner. The other is that as time progresses, better efficiency is
expected. Hence, the units with declining efficiency must be identified in order to make the necessary
improvements (Greenaway, Görg, Kneller, 2008). Nowadays, for many economic entities (also for
international organizations and territories, thus countries and regions) is the focus in increasing
efficiency the most important task. The flagship initiative for an economic efficient Europe under the
Europe 2020 strategy supports the shift towards a resource-efficient economy to achieve sustainable
growth. Continuing our current patterns of resource use and output produce is not an option.
Increasing economic efficiency is key to securing growth and jobs for Europe (Staníčková, 2013). It
will bring major economic opportunities, improve productivity, drive down costs and boost
competitiveness. But within the whole EU, there are two groups of countries – “old” and “new”
Member States having different initial, not only economic, conditions for future development. From
this point of view, it´s firstly necessary to recognize specific characteristics of both groups, their
advantages and disadvantages, for more efficient cooperation. The purpose of this paper was
recognize how to determine the appropriate group exhibiting similar features for subsequent efficiency
evaluation.
Which of two used approaches (CA and DEA) in the formation of homogenous groups, is more
suitable for subsequent efficiency analysis? With respect to characteristic of DEA method and one the
most important requirements on the relation between the number of DMUs on the one side and the
number of inputs and outputs on the other side; then with respect to classification method and
expression similarity, DEA approach based on function of Obtain Levels seems to be more convenient
way how to choose an optimal number of groups covering all evaluated regions coming into further
efficiency analysis. Because obtained results by DEA approach show that targets are the coordinates
of the efficient projection point on the frontier and thus represents levels of operation of inputs and
outputs which would make the corresponding inefficient DMU, thus NUTS 2 region perform
efficiently – but this improvement will be possible to propose based on subsequent efficiency analysis
by DEA method. Efficiency evaluation of NUTS 2 regions within EU13 countries based on above
classified 3 groups/levels is following orientation of further empirical research. Targets can thus play
an important role in practice since they may indicate keys for the inefficient regions to improve their
performance.
Sborník příspěvků XVII. mezinárodní kolokvium o regionálních vědách Hustopeče 18.–20. 6. 2014
52
References
[1] ANNONI, P., DIJKSTRA, L., (2013). EU Regional Competitiveness Index 2013. Luxembourg: Publication
Office of the European Union. ISBN 978-92-79-32370-6.
[2] ANNONI, P., KOZOVSKA, K., (2010). EU Regional Competitiveness Index 2010. Luxembourg: Publication
Office of the European Union. ISBN 978-92-79-15693-9.
[3] COOK, W.D., SEIFORD, L.M. , (2009). Data envelopment analysis (DEA) – Thirty years on. European
Journal of Operation Research, vol. 192, iss. 1, pp. 1-17. ISSN 0377-2217. DOI
10.1016/j.ejor.2008.01.032.
[4] GREENAWAY, D., GÖRG. H., KNELLER. R., (2008). Globalization and Productivity. Cheltenham:
Edward Elgar Publishing Limited. ISBN 978-1845423803.
[5] HAIR, J. F., BLACK, W.C. et al., (2009). Multivariate Data Analysis. New Jersey: Prentice Hall. ISBN
978-0138132637.
[6] MELECKÝ, L., (2013). Comparing of EU15 and EU12 Countries Efficiency by Application of DEA
Approach. In Vojáčková. H. (ed.) 31st International Conference Mathematical Methods in Economics.
Jihlava: College of Polytechnics Jihlava. pp. 618-623. ISBN 978-80-87035-76-4.
[7] MOLLE, W., (2007). European Cohesion Policy. London: Routledge. ISBN 978-0415438124.
[8] STANÍČKOVÁ, M., (2013). Measuring of efficiency of EU27 countries by Malmquist index. In
Vojáčková. H. (ed.) 31st International Conference Mathematical Methods in Economics. Jihlava: College
of Polytechnics Jihlava. pp. 844-849. ISBN 978-80-87035-76-4.
This paper was created under SGS project (SP2014/111) of Faculty of Economics. VŠB-Technical
University of Ostrava and Operational Programme Education for Competitiveness – Project
CZ.1.07/2.3.00/20.0296.
Sborník příspěvků XVII. mezinárodní kolokvium o regionálních vědách Hustopeče 18.–20. 6. 2014
53
Appendix 1: Dendogram using Ward linkage – Clusters of EU 13 NUTS 2 regions
Source: own calculation and elaboration, 2014