Content uploaded by Imre Dobos
Author content
All content in this area was uploaded by Imre Dobos on Nov 01, 2020
Content may be subject to copyright.
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
What the overall Digital Economy and Society Index
reveals: A statistical analysis of the DESI EU28
dimensions
Zoltán Bánhidi
Budapest University of Technology
and Economics
E-mail: banhidiz@kgt.bme.hu
Imre Dobos
Budapest University of Technology
and Economics
E-mail: dobos@kgt.bme.hu
András Nemeslaki
Budapest University of Technology
and Economics
E-mail: nemeslaki@finance.bme.hu
Keywords:
digital transformation measurement,
DESI index,
correlation analysis,
cluster analysis,
multidimensional scalin
g
We build on the argument that measuremen
t
of digitalization is essential for effective
public policy strategies in order to to govern
digital transition. Developing this argument,
we investigate the five principal dimensions
of the European Commission´s Digita
l
Economy and Society Index (DESI) using
a
series of multivariate statistics. The analysis
can be divided into three groups. First, we
analyse the linear relationships between
dimensions by correlation analysis, partia
l
correlation analysis, and principal componen
t
analysis. In the partial correlation analysis,
causal relationships between the dimensions
show high correlations. Second, we assign
countries into groups with cluster analysis
and multi-dimensional scaling. The groups
obtained by the two methods are very similar.
Finally, we rank the European Union (EU)
countries using statistical methods and
compare with the results obtained with the
overall DESI index. The correlation between
the two rankings shows a strong linea
r
relationship. Based on these results we dra
w
conclusions on how to effectively use the
DESI data for
p
ublic
p
olic
y
anal
y
sis.
Introduction to digital transformation and problems of
its measurement
During the past 25 years, with the emergence of e-business and the internet
economy, several industries have been restructured with Information and
Communications Technology (ICT) ‘disruption’, and we can hardly recognize some
of them from their past perspectives. For instance, the internet has reformed media
industries such as music, television and films; retail businesses such as books,
newspapers, apparel and electronics; and services such as travel, banking or
What the overall Digital Economy and Society Index reveals:
A statistical analysis of the DESI EU28 dimensions 43
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
insurance. Initially, the e-business revolution only took place in industries related to
information or digital products, while others that were not influenced by the rapid
disruptions in information technology remained intact. However, ICT is now more
pervasive and affordable pressure on innovation has also become wide spread.
Broadly, in this paper, we define this phenomenon as digital transformation.
Pervasiveness may be characterized by penetration, mobility, and access to
computers, devices, and networks. We are very close to the scenario when this
pervasiveness will increase connectedness of the entire world population. Significant
progress has been made towards bridging the global digital divide, achieving
universal and affordable access to the internet in the least developed countries.
Affordability, on the other hand, implies costs and how the costs of hardware,
software, and connectivity compare to incomes and general economic development
per capita. We can expect to witness tremendous progress in this area, and not only
will devices get cheaper but the rates of telecommunication services will also
decline. However, ICT-based business transformation does not happen because of
the economics of ICT, but it is due to the application of such technologies that
allow us to do things fundamentally differently. In a fascinating monograph, former
Google CEO Eric Schmidt and Jared Cohen (2013), a foreign relations and
counterterrorist expert, paint an exciting new world of the future in the digital age –
how ICT reshapes people, businesses, and countries.
In some industries, such as automotive and electronics, competitiveness of
companies is already influenced by the efficient deployment of information and
communication technology, the efficiency of innovation processes, co-operation of
intelligent machines, and of people at all levels (including the shop floor), or by
intertwining artificial intelligence/learning algorithms with everyday processes.
Economic and technological trends, however, indicate that this development will
not only shape the auto and machine industries, but will basically transform all
businesses. A new ecosystem, labelled in several countries as Industry 4.0 or the era
of the industrial internet, is emerging, and it not only encompasses the dilemmas of
how ICT is embedded in blue and white collar processes leading to radical
innovations but also draws attention to future work, new forms of financing, the
changing landscape of international strategic management, organizational culture,
and the need to transform leadership practices.
Notwithstanding the elimination of the classic elements of digital divide, such as
barriers to ICT adoption, use of social media, or the uptake of current e-
government services, new chasms have appeared, e.g. regarding privacy,
cybersecurity, or the major challenge of how to deal with fake news and other forms
of cyber manipulation. The appearance of this new ‘digital divide’ raises the concern
for the need to reconsider the new relationship between governments and
businesses, how public policies and science and technology programs influence
digital transformation and, more importantly, how the new form of leadership must
react to these challenges.
44 Zoltán Bánhidi – Imre Dobos – András Nemeslaki
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
In the context of the European Union (EU), a universal scoring system – the
Digital Economy and Society Index (DESI) – has been developed for measuring
preparedness and progress of digital transformation. For capturing the complexity
of the above socio-economic digital transformation, DESI collects and organizes
the statistical data of 28 EU countries into five main dimensions: Connectivity,
Human Capital, Use of Internet, Integration of Digital Technology, and Digital
Public Services. Notwithstanding that DESI has been around since 2014, we intend
in our paper to highlight two main dilemmas using the latest data available from
DESI Report 2018 (European Commission 2018) for the five indicators. Of course,
in addition to DESI, the EU also sets other similar indices and efficiency indicators
in other areas of the economy and society based on EU databases. For example, the
definition of an international reputation index is presented by Fernandez-Crehuet et
al. (2019). Dudek and Sedefoğlu (2019) use material deprivation rates in the
indicators in a European Union context.
Correlation and causality of the five dimensions
The European Commission accepts that the five principle policy areas of DESI are
not isolated from each other and clearly show that digital transformation can only
be achieved by a concerted effort that must be reflected in the DESI dimensions.
This principle, however contradicts the requirement of avoiding redundancy that
has led to both statistical and interpretation problems. Since the five DESI
dimensions comprise several sub-indicators from statistical databases, we argue
about the importance of investigating, in pure statistical terms, how strong this
correlation is, how many different dimensions DESI measures, and how the
dimensions impact each other. Our first set of research questions aim to answer
these problems.
Problem of country clusters and ranking according to DESI
The typical use of the DESI measurement system – what the European
Commission publishes with every new DESI data recording – is the grouping and
ranking of countries, which is considered as comparison or ‘scoring’ of EU
countries related to their digital transformation. It is interesting that while the term
‘cluster’ is used in this context, DESI reports do not apply a statistical form of
cluster analysis or scaling methods; instead they prefer weighting of the five
dimensions and publish country grouping accordingly. For instance, in one of the
latest reports, the European Commission identified three so-called clusters, namely
high, medium, and low performing EU countries with 9, 10 and 9 members
belonging to them. We argue that without exploring how statistical clusters and
scaling separates and ranks countries according to the raw DESI data, an
unnecessary bias may distort member states´ position. In other words, by finding
out how naturally DESI ranks and groups countries by using appropriate
What the overall Digital Economy and Society Index reveals:
A statistical analysis of the DESI EU28 dimensions 45
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
multivariate statistical methods, we can reveal the inherent structure of EU
countries and juxtapose it with existing reports. Our second set of research
questions intend to contribute to this problem by analysing DESI data with cluster
analysis and multidimensional scaling.
The paper is organized as follows. Section 2 presents a brief review of the
existing literature on the analysis of DESI data. Section 3 presents the DESI overall
index with the five indicators. Please note that the sub-dimensions and individual
indicators of DESI dimensions are not examined in this paper. Section 4 covers the
statistical analysis. The linear connections of the variables are examined with
classical correlational analysis, partial correlation analysis, and principal component
analysis. The countries are grouped with cluster analysis and multidimensional
scaling. The following section is an application of the multidimensional scaling
model to rank countries. These statistical results are then compared with the scoring
model developed by the European Commission. Section 5 concludes the paper by
presenting the results.
A conceptual review of the Digital Economy and Society Index
Measurement of digital transformation: introduction to DESI and other
systems
There are several indices, scores, indicators, and measurement units that describe
the status of the digital economy, society, public administration, and they are used as
descriptors of digital transformation.
First, some of these scoring systems describe and compare the impact of global
digitization. For instance, reports from the United Nations (UN), Organization for
Economic Cooperation and Development (OECD), World Bank, or the
International Telecommunication Union (ITU) serve similar objectives as the
regular research projects of major consulting firms such as Forrester, International
Data Corporation (IDC), Gartner or McKinsey.
The second category of these measures focus on regional or well-defined
country clusters belonging to a geopolitical area. Typical surveys of this type are the
EU scoring boards: Digital Skills Indicator (DSI), Consumer Conditions Scoreboard
(CCS) indices, or the Digital Economy and Society Index (DESI).
Finally, the third set of data collected for describing the impact of ICT are
country-specific compilations conducted by the respective National Statistical
Offices or domestic research firms.
The DESI report tracks the progress made by EU member countries with
respect to their digitization. It is structured around five chapters (Table 1):
46 Zoltán Bánhidi – Imre Dobos – András Nemeslaki
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
Table 1
Dimensions of DESI
Connectivity Fixed broadband, mobile broadband, and prices
Human Capital Internet use, basic and advanced digital skills
Use of Internet Services Citizens' use of content, communication, and online transactions
Integration of Digital Technology Business digitization and e-commerce
Digital Public Services eGovernment and eHealth
Source: https://digital-agenda-data.eu
It is a widely used and quoted measurement system by experts and policy
makers, however it has certain advantages and serious limitations. The main
advantage is that it is measured in 28 countries, enables comparison, it is accepted
by the European Union, allows compliance and provides the big picture of the
digital ecosystem in the EU and its member countries.
Disadvantages are rooted in the same sources as the advantages. The fact that
measurements are collected in 28 different countries entails that the methodology is
determined to be general and applicable to all. Therefore the results are also general
and not suitable for deep analysis and to explain certain phenomena. Specifically,
the major drawbacks are that the measurement factors often have the impression of
improvised choice in a given year and they change frequently. It often seems biased
by industry lobbies, the period between data collection and publication is very long,
thereby frequently resulting in outdated assessments. The composition of
dimensions changes year-by-year, and this makes it difficult to compare time series
performances because these corrections are often not emphasized. There are also
significant differences between the statistical offices and data collection methods
among countries.
Regardless of the problems, DESI´s method and collection system is still a
robust approach, unavoidable in many instances, and it is considered as the best
choice for mapping Europe’s progress on digitalisation.
Literature review on conceptual questions of DESI
Stoica and Bogoslov (2017) compared the five indicators of DESI with the available
data for Romania and the EU and analysed them over time. They concluded that
Romania had undergone significant development during the period under review
(2014–2017). However, they did not analyse what type of statistical correlations
exist between the available data. Using data from Greece, Kontolaimou and Skintzi
(2018) published a similar study. Their research was methodologically deeper,
extended to human capital, including gender. Dynamic effects are presented in this
analysis; in this case however it is difficult to test dynamic datasets due to limited data
What the overall Digital Economy and Society Index reveals:
A statistical analysis of the DESI EU28 dimensions 47
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
availability. Mirke et al. (2019) analysed human capital elements of the Czech Republic
and Latvia, and assessed the adult population’s digital competence through learning.
Nikolov and Krumova (2019) considered the fifth element of DESI in their
study, that is e-Government, and examined the countries of the European Union
(EU) in their model. Scupola (2018) provided an overview of the state of digital
transformation in Denmark. The study also reported experiences in e-Government.
This paper appears to be very insightful as Denmark has been one of the best-
performing countries in DESI rankings for several years. Urs (2017) also
investigated the development of e-Government, focusing on local municipalities in
Romania and highlighted that improving their service quality can be a daunting task
for their understaffed and underfinanced IT departments, hampered by the lack of
central coordination and deficient national infrastructure.
Russo (2020), following the European guidelines on the DESI index, applied
them on the Italian region of Abruzzo to provide a local framework of
technological development. Alonso–García (2018) analysed the digital economy in
the context of entrepreneurship using DESI. Their research focused on how
digitalization affects the entrepreneurial ecosystem and had found that digitalization
changes not only the size of the entrepreneurial market but also its profitability.
Curko et al. (2017) considered DESI from the perspective of smart business and
examined the impact of Industry 4.0 on the digital economy. Similarly, Götz (2017)
analysed the impact of Industry 4.0 on the economic relations between Germany
and Poland, concluding that the digital economy can have a positive effect on
German-Polish relationships.
These illustrative overviews indicate that the DESI has been used in academic
research in order to understand how digital transformation works across different
European countries, sectors, and social and economic dimensions. Thus, we argue
that it is important to understand the statistical robustness of the DESI
methodology, specifically the conclusions we can draw using the five dimensions in
such contexts.
Research questions and methodology – statistical analysis of
DESI 2018 indicators
To analyse the two problem areas, we propose five research questions (RQ): three
related to the problem of correlation (RQ1-RQ3), and two assigned to the problem
of country clustering and ranking (RQ4-RQ5).
RQ1: What are the linear relationships between the DESI dimensions?
RQ2: Based on their relationships can the dimensions be reduced? That is, do
the dimensions measure different latent variables?
RQ3: Can we reveal any information on the causal relationship between the
DESI dimensions?
48 Zoltán Bánhidi – Imre Dobos – András Nemeslaki
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
RQ4: How can we cluster the European Union countries – as data points –
using the DESI dimensions?
RQ5: By using the DESI dimensions how can we arrange the country data
points – by creating a natural ranking – and compare this with the published DESI
overall index?
For data analysis, our DESI dataset of 2018 was compiled from the EU
Commission’s website https://digital-agenda-data.eu (Table 2). This raw dataset for
the five dimensions – synonymously referred to as variables in this context – is
presented in Table 2 and used for the multivariate statistical analysis (Tabachnick et
al. 2007):
Table 2
Raw DESI 2018 data for analysis
Country Connectivity Human
Capital
Use of
Internet
Integration
of Digital
Technology
Digital
Public
Services
Denmark DK 78.482 70.413 75.085 61.283 73.219
Sweden SE 75.985 74.227 73.370 56.367 70.765
Finland FI 66.053 79.240 65.406 60.879 78.642
Netherlands NL 81.132 74.254 66.458 52.338 70.550
Luxembourg LU 80.053 71.311 65.864 33.175 56.208
Ireland IE 65.118 61.706 52.316 60.017 64.719
United Kingdom UK 68.848 71.631 62.395 39.962 58.244
Belgium BE 75.128 57.452 53.310 54.560 57.872
Estonia EE 64.112 61.375 61.607 37.065 78.103
Spain ES 64.708 54.578 49.431 49.791 72.362
Austria AT 63.725 64.365 47.568 44.145 66.468
Malta MT 73.140 51.633 63.281 38.904 61.262
Lithuania LT 64.895 48.526 56.845 47.455 68.217
Germany DE 64.745 62.855 52.743 41.326 50.233
Slovenia SI 60.337 52.009 44.938 47.866 57.350
Portugal PT 67.383 45.824 46.344 41.928 59.643
Czech Republic CZ 63.861 55.100 46.460 40.444 50.165
France FR 56.404 59.069 42.247 37.828 58.371
Latvia LV 65.932 43.840 54.777 27.027 65.185
Slovakia SK 55.071 51.882 51.285 37.430 50.401
Cyprus CY 60.632 43.027 51.136 37.674 54.815
Croatia HR 49.436 49.807 54.075 35.439 44.351
Hungary HU 61.728 47.978 53.638 25.082 40.390
Poland PL 58.810 48.283 42.080 23.532 48.208
Italy IT 52.826 40.810 37.364 36.821 52.503
Bulgaria BG 54.906 34.832 41.684 24.424 49.717
Greece EL 43.104 38.185 45.218 26.940 39.239
Romania RO 58.119 32.065 34.964 17.756 41.362
Source: https://digital-agenda-data.eu
What the overall Digital Economy and Society Index reveals:
A statistical analysis of the DESI EU28 dimensions 49
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
We used the IBM SPSS 20 Statistical Program Package to perform the following
methodological calculations. For analysing the linear and causal relationships (RQ1
and RQ3), simple Pearson and partial correlation measures were used. Pearson
correlation shows the strength and direction of the linear relationship between two
variables, while the partial correlation provides information on the causal
relationship between the two variables, but without the ability to deduce its
direction. For RQ2, that is, basically for compressing information, we performed
principal component analysis on the datasets, based on which we examined the
latent variables describing the best returned variance. Basically, this calculation
shows how many new variables characterize the existing five dimensions. In order
to cluster the EU countries (RQ4) we used hierarchical cluster analysis. This method
shows which countries are positioned relatively close to each other in the five-
dimensional space determined by the original DESI variables. Multidimensional
scaling was used to assess the stability of the cluster analysis; a method in which the
elements of the five-dimensional space are represented in lower dimensions. We
preferred the two-dimensional space for visualization, in accordance with the results
of the principal component analysis. Finally, for answering our last research
question (RQ5), we applied an extreme case of multidimensional scaling by mapping
our country data points from the five-dimensional space to a one-dimensional
space, which may be called a straight numerical order. This order can be turned into
ranking, when and if this mapping shows a good fit.
Results and discussion
RQ1: Correlation analysis
Table 3 shows that there is a moderate to strong correlation between the variables.
Each of these correlation coefficients has a positive sign, suggesting that the
movement of variables is in a single direction; importantly, all the correlation ratios
are significant at 99% confidence level. We consider this as an important
consequence to any further use of the DESI dimensions, in which case the dilemma
is about how orthogonal these dimensions are in terms of our understanding the
digital economy and society. In other words, how will each of these dimensions add
value to our assessment of the status of digital transition in the EU? In any case, a
high correlation could be alarming in this respect, pointing towards little additional
value for some variables in the DESI model.
50 Zoltán Bánhidi – Imre Dobos – András Nemeslaki
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
Table 3
Correlation matrix between DESI variables
Human
Capital
Use of
Internet
Integration of
Digital
Technology
Digital Public
Services
Connectivity
Pearson
Correlation 0.683** 0.736** 0.547** 0.605**
Sig. (2-tailed) 0.000 0.000 0.003 0.001
Human Capital
Pearson
Correlation 0.765** 0.709** 0.647**
Sig. (2-tailed) 0.000 0.000 0.000
Use of Internet
Pearson
Correlation 0.568** 0.616**
Sig. (2-tailed) 0.002 0.000
Integration of
Digital Technology
Pearson
Correlation 0.726**
Sig. (2-tailed) 0.000
**Correlation is significant at the 0.01 level (2-tailed).
RQ2: Application of principal component analysis
Principal component analysis reveals how the five – strongly correlated – DESI
variables may be reduced. We used the principal component analysis method,
without rotation, with VARIMAX rotation to answer the dilemmas raised as a result
of RQ1, that is, to identify the group of variables that describe different aspects of
the digital economy.
The Kaiser-Meyer-Olkin measure of sampling adequacy shows that our model is
meritorious with 0.834 value, confirming its fitness. At the same time, Bartlett's test
of sphericity is proven significant, further underscoring the goodness of our model.
These tests support the acceptance of our model’s results.
First, we report communalities values that are between 0.8 and 0.9, suggesting
that the calculated components explain a large proportion of the variance. In our
analysis, two components are sufficient for this purpose and they returned 84.280
percent of variance. Specifically, in the principal component model without rotation,
the first component yielded 72.859 percent of variance, and the value of component
weights was above 0.830, indicating that each variable strongly correlated with this
component. In the case of our second component, each variable showed weak
correlation, except Integration of Digital Technology, which showed weak-medium
correlation of 0.451. The second component displayed 11.422 percent of variance.
We obtained a rather insightful result, concluding that the second part of the
principal component analysis, which was computed with the VARIMAX rotation of
What the overall Digital Economy and Society Index reveals:
A statistical analysis of the DESI EU28 dimensions 51
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
the above results, yielded 84.280 percent of the variance on the two components.
Table 4 summarizes the two components and their corresponding variables with the
respective component weights.
The first component is strongly determined by Connectivity and Use of Internet
variables, while the second with strong component weights such as Integration of
Digital Technology and Digital Public services. We can observe that Human Capital
shows moderate weights with both components, slightly higher with the first one.
This suggests that both components have a medium linear relationship with the
variable Human Capital. We will clarify this in more detail by calculating the partial
correlation values between our dimensions. Table 4 is an important and somewhat
intuitive outcome of our quantitative analysis, namely the five DESI dimensions can
be reduced to two different variables describing the status of digitalization: one
component provides the measure of ‘preparedness’ by integrating infrastructure and
its use; while the other, more obvious component, measures what is actually
happening with digital technologies, namely how it is integrated into businesses,
social institutions, and public administration. However, the special status of Human
Capital is somewhat counter intuitive, which one may have predicted as a
‘preparedness’ variable, but principal component analysis results do not provide
strong proof on this issue. Table 4
Rotated Component Matrix
Component
1 2
Connectivity 0.862 0.300
Use of Internet 0.867 0.339
Human Capital 0.699 0.562
Integration of Digital Technology 0.302 0.895
Digital Public Services 0.387 0.818
Extraction Method: Principal Component Analysis.
Rotation Method: Varimax with Kaiser Normalization.
These conclusions lead us to the third part of the correlation problem, that is,
whether the data show any further relationship between the DESI dimensions in
order to disclose the causality underlying our components.
52 Zoltán Bánhidi – Imre Dobos – András Nemeslaki
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
RQ3: Partial correlational analysis
We report the results of partial correlation analysis in Table 5 by italicizing values
that are significant at 5% level. Table 5 shows that six relationships do not affect
each other by eliminating the impact of the other variables, and the values of the
four significant polar correlations are between 0.4 and 0.5, which is considered a
weak-medium linear relationship between these variables after filtering out the
effects of the other three variables. Table 5
Partial correlation matrix between variables after filtering
Human
Capital
Use of
Internet
Integration of
Digital
Technology
Digital Public
Services
Connectivity
Pearson
Correlation 0.177 0.411 0.005 0.181
Sig. (2-tailed) 0.397 0.041 0.983 0.387
Human Capital
Pearson
Correlation 0.454 0.407 0.032
Sig. (2-tailed) 0.023 0.044 0.880
Use of Internet
Pearson
Correlation –0.073 0.151
Sig. (2-tailed) 0.727 0.470
Integration of
Digital Technology
Pearson
Correlation 0.490
Sig. (2-tailed) 0.013
Numerical data from Table 5 can be translated into a graphical visualization shown
in Figure 1. By acknowledging that partial correlation analysis only reveals causal
relationships but does not provide information on its direction, we carefully avoided
drawing arrows on the connecting lines. The answer to this research question
suggests that DESI has two ‘independent variables’ (Connectivity and Human
Capital), and three ‘dependent variables’, which are Use of Internet, Integration of
Digital Technology, and Digital Public Services.
What the overall Digital Economy and Society Index reveals:
A statistical analysis of the DESI EU28 dimensions 53
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
Figure 1
Causal relationships between variables
Figure 1 shows a somewhat unexpected result that needs further investigation
and has ample consequences on the use of DESI for policy making. The data
suggests that penetration and level of digital public services are only dependent on
how digital technology is integrated into other business processes and it is not
directly connected with internet use. Similarly, the integration of digital technology
into business processes is in causality with human capital not significantly with
connectivity, which we may consider as the basis for digital transformation.
Elaborating on the consequences of these results – RQ1, RQ2, and RQ3 – is
beyond the scope of this paper but causality mapping may be explored for more
elaborate policy design in digital transformation.
RQ4: Grouping countries with hierarchical cluster analysis
Cluster analysis is a statistical method for classifying observations, in this case
regarding the EU 28 countries, in the space of the variables, that is, the five DESI
dimensions. We argue that it is an important incumbent grouping of data points
instead of choosing artificially defined ‘leaders and laggards’ categories according to
predetermined DESI composite values. We must assign a group centre to determine
the distance between countries in cluster analysis based on data metrics, which is the
squared Euclidean distance in this paper, and our scale of measurement is an
interval scale. We selected the hierarchical cluster methodology because it can be
supported with the dendrogram visualization, and is an expressive demonstration of
how distinctly different groups emerge as a result of creating clusters. Dendrograms
enable the monitoring of cluster attachment and cluster spreading. The first result
describes how closely data points belong to a given group, and the second shows
how many different groups the data points may be divided into. The components of
our dendrogram are shown in Figure 2, and its interpretation is presented with the
Connectivity
Human Capital
Use of Internet
Integration of
Digital Technology
Digital Public
Services
54 Zoltán Bánhidi – Imre Dobos – András Nemeslaki
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
help of Table 6, which summarizes cluster membership and the sequence how
clusters are formed. Figure 2
Dendrogram of cluster analysis
First, the algorithm has separated the two clusters, basically the top performing
countries in the digital economy from the remaining ones. These countries are
Denmark, Sweden, Finland, Netherlands, Luxembourg, and the United Kingdom.
Second, a more refined picture is seen when the rest of the population falls into two
distinct groups, separating the low performers, namely Bulgaria, Romania, and
Greece. Finally, in steps three and four a wider cluster spread can be defined by
lowering the distance between the group centres. As a result, we can observe that
What the overall Digital Economy and Society Index reveals:
A statistical analysis of the DESI EU28 dimensions 55
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
two countries are separated from the leading group, that is, Luxemburg and the UK,
and the populated middle group falls into two subgroups comprising a somewhat
more and a rather less developed one. In the former cluster, we have Ireland,
Belgium, Estonia, Spain, Austria, Malta, Lithuania, Germany, Slovenia, Portugal,
Czech Republic, and France, while Latvia, Slovakia, Cyprus, Croatia, Hungary,
Poland, and Italy are grouped in the latter. Table 6
Changes in cluster membership
Country 2 Clusters 3 Clusters 4 Clusters 5 Clusters
Denmark DK 2 3 3 4
Finland FI 2 3 3 4
Netherlands NL 2 3 3 4
Sweden SE 2 3 3 4
Luxembourg LU 2 3 4 5
United Kingdom UK 2 3 4 5
Austria AT 1 1 1 1
Belgium BE 1 1 1 1
Czech Republic CZ 1 1 1 1
Estonia EE 1 1 1 1
France FR 1 1 1 1
Germany DE 1 1 1 1
Ireland IE 1 1 1 1
Lithuania LT 1 1 1 1
Slovenia SI 1 1 1 1
Spain ES 1 1 1 1
Croatia HR 1 1 1 3
Cyprus CY 1 1 1 3
Hungary HU 1 1 1 3
Italy IT 1 1 1 3
Latvia LV 1 1 1 3
Malta MT 1 1 1 3
Poland PL 1 1 1 3
Portugal PT 1 1 1 3
Slovakia SK 1 1 1 3
Bulgaria BG 1 2 2 2
Greece EL 1 2 2 2
Romania RO 1 2 2 2
It may be observed that hierarchical cluster analysis provides a deeper insight
into understanding country performance based on the Euclidean distance in the
five-dimensional DESI space without setting artificial thresholds or group numbers.
Depending on the scale of difference, we can identify country groups that are really
and consistently similar – these are the results of the first two steps of the algorithm
56 Zoltán Bánhidi – Imre Dobos – András Nemeslaki
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
– and we refine grouping also, by choosing lower Euclidean distance values as in the
third and fourth step, resulting in more clusters but with less distinct difference
between the middle group. Based on the structure of DESI data, we conclude that
the EU 28 countries fall into five separate and relevant clusters as presented in
Table 6:
1. Cluster (#4): Denmark, Finland, Netherlands, and Sweden,
2. Cluster (#5): Luxembourg and United Kingdom,
3. Cluster (#1): Austria, Belgium, Czech Republic, Estonia, France, Germany,
Ireland, Lithuania, Slovenia, and Spain,
4. Cluster (#3): Croatia, Cyprus, Hungary, Italy, Latvia, Malta, Poland, Portugal,
and Slovakia, and
5. Cluster (#2): Bulgaria, Greece, and Romania.
In the next section, clustering is performed using another method, the
multidimensional scaling.
RQ5: Grouping countries by multidimensional scaling
When applying multidimensional scaling (MDS), we consider the smallest
dimensional space in which the distances between observations can be best
returned. Thus, our data are derived from the five-dimensional space into a smaller
dimensional one, but preferably to the size that we can visualize. In our case, we
transform our data into the plane, a two-dimensional space. This corresponds with
the results of the principal component analysis that yielded two components
condensing the five DESI dimensions. One of these components is ‘digital
application’ and we may summarize the other as ‘readiness’ or ‘digital competence’,
as indicated in Figure 3. The adequacy of dimension reduction is robust, since the
stress value is very low at 0.103, which is the sum of the squares of the difference
between the five-dimensional and two-dimensional distances.
What the overall Digital Economy and Society Index reveals:
A statistical analysis of the DESI EU28 dimensions 57
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
Figure 3
Location of countries on the plane of principal components
Digital applications
0.0
0.5
1.0
0–3 –2 –1 1 2
Digital competences
–1.5
–1.0
–0.5
3
The adequacy of the model is further confirmed by the R-squared value
(R2=0.958), which shows a strong linear relationship between the points in the two
dimensional space, as R-squared over 0.900 linearity is considered to be very good.
For a very simple interpretation of Figure 3, we may conclude that the DESI
measurement distributes the EU 28 countries into four quadrants based on their
level of competency and application of digital technologies. The first quadrant in the
upper right corner contains countries where both components are higher than
average, that is, they perform well in both competencies (connectivity, use) and
application (integration and public services). Moving clockwise, the lower right
corner collects countries where applications lag the opportunities of competencies –
the value of this latter component is negative. Countries which are lower than
average in both components can be found in the lower left corner, which
corresponds to the fifth and fourth clusters of the hierarchical cluster analysis
showing the refined situation, as in Romania and Bulgaria; although lagging in terms
of the competency component, they perform better in terms of the application
dimension. Finally, the last quadrant is the upper left segment with countries that
exploit applications of technology to a certain extent, that is, beyond the level of
competency or their actual readiness.
58 Zoltán Bánhidi – Imre Dobos – András Nemeslaki
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
In the second part of this research topic we apply MDS to generate ranking of
data points measured by the DESI variables. Basically, this is an extreme case when
the original data are projected into a one-dimensional space using MDS, that is, we
assign values for each country lying on a numerical line.
It is important to note that this method is fundamentally different from the
present logic of DESI ranking (or overall DESI), since applying a multidimensional
scoring method involves weighing of the five dimensions and aggregating them
accordingly. Hence DESI overall is a scalar multiplication of the DESI variables
presented in Table 2, and a predetermined weight vector defined by the EU
Commission is shown in Table 7. Table 7
Weights of the variables for DESI overall index
Connectivity Human Capital Use of
Internet
Integration of
Digital
Technology
Digital Public
Services
0.25 0.25 0.15 0.2 0.15
The EU Commission´s website (https://digital-agenda-data.eu) provides a
simulation tool where weight values may be changed and DESI ranking can be
calculated by altering the weight of Connectivity for the benefit of Integration or
Digital Public Services. No matter which values are set, however, the logic of
multidimensional scoring always reflects the competence of the decision maker or
the requestor for the ranking. On the other hand, MDS provides incumbent ranking
based on the structure of data that are independent of any external utility
assessment of the DESI dimensions. The results of MDS and its comparison with
the existing scoring is presented in Table 8. The stress value in this case is 0.187,
which is worse than that of the plane model in Figure 8, although still acceptable.
Furthermore, the value of R-squared is still very high at 0.908, as in our two-
dimensional model.
Table 8 shows that the result of the scoring model does not differ significantly
from MDS. The correlation between the two rankings is 0.991, indicating a strong
linear relationship between them. On the other hand, consistent with our previous
cluster analysis results, we identify some important findings regarding the
assessment of a few countries. Among the high performers, Finland´s position is
different according to the two methods, which may indicate some interesting best
practice observations and transfer. Also, there are slight positional differences
between countries in the middle and lower groups.
What the overall Digital Economy and Society Index reveals:
A statistical analysis of the DESI EU28 dimensions 59
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
Table 8
DESI overall indicators with ranking
Country DESI overall
index
Ranking with
DESI MDS values Ranking with
MDS
Denmark 71.726 1 1.801 1
Sweden 70.447 2 1.638 3
Finland 70.106 3 1.780 2
Netherlands 69.865 4 1.493 4
Luxembourg 62.787 5 0.905 5
Ireland 61.264 6 0.760 6
United Kingdom 61.208 7 0.659 8
Belgium 60.735 8 0.544 9
Estonia 59.741 9 0.728 7
Spain 58.048 10 0.427 10
Austria 57.957 11 0.343 11
Malta 57.655 12 0.255 12
Lithuania 56.606 13 0.254 13
Germany 55.612 14 0.021 14
Slovenia 53.003 15 –0.186 15
Portugal 52.585 16 –0.283 16
Czech Republic 52.323 17 –0.312 17
France 51.527 18 –0.331 18
Latvia 50.843 19 –0.485 19
Slovakia 49.477 20 –0.487 20
Cyprus 49.342 21 –0.528 21
Croatia 46.663 22 –0.823 22
Hungary 46.547 23 –1.010 23
Poland 45.023 24 –1.062 24
Italy 44.253 25 –1.084 25
Bulgaria 41.029 26 –1.416 26
Greece 38.379 27 –1.675 27
Romania 37.546 28 –1.926 28
60 Zoltán Bánhidi – Imre Dobos – András Nemeslaki
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
Figure 4
DESI overall indicators
Conclusions
Measuring the processes and results of digital transformation is important for policy
making and for improving local, regional, and European level competitiveness.
From 2014 onwards, the EU Commission has been using the Digital Economy and
Society Index (DESI) as a complex measurement scheme for collecting data from
member states in order to assess the business and social implications of
digitalization status in the EU. Over the years, DESI has emerged as a reference
point and a major source for policy making and comparing the performance of
different countries. In our paper we analysed two major problems by defining five
research questions of DESI that need more attention regarding its use.
Three research questions are used to analyse the first problem area covering the
issues of correlation and causality of the five DESI dimensions. After running
correlation tests, principal component analysis, and partial correlation computations,
we concluded that the five DESI dimensions are strongly correlated, they can be
reduced to two principal components covering the same scope of measurement, and
there is a causality mapping the variables´ relationship. According to these findings
there are two independent dimensions (Connectivity and Human Capital) and three
DESI overall index
37,546–41,000
41,001–49,000
49,001–55,000
55,001–62,000
62,001–71,726
What the overall Digital Economy and Society Index reveals:
A statistical analysis of the DESI EU28 dimensions 61
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
dependent ones, of which Use of Internet shows causality with both independent
dimensions, and Integration of Digital Technology shows causality only with
Human Capital. Furthermore, Digital Public Services clearly show a relationship
with the Integration dimension, underlying the intuitive experience that applications,
both in business and public services are creating a strong principal component while
capabilities such as connectivity and human readiness create another.
In the second problem area two research questions examined the clustering and
ranking of EU28 countries and juxtaposed the results of hierarchical cluster analysis
and multidimensional scaling with the DESI overall indices. With the introduction
of these two multivariate statistical methods we created an incumbent grouping of
EU28 countries and mapped the groups on a two dimensional and one-dimensional
(ranking) space. Our results confirm the usability and effectiveness of the DESI
overall index – or weighted aggregate score – and also present a more subtle
mapping and ranking of EU countries in the DESI dimensions.
Our paper has its limitations that require future research. A key limitation is that
we ran the analyses based only on 2018 data. Therefore, in order to generalize our
conclusions analysis should be extended to previous datasets as well. Furthermore,
we focused on the five highest level dimensions, a more detailed picture can be
gained using second-level data before aggregating them into the five top variables.
Regardless we believe our contribution is relevant from at least two points. First, by
extending the multivariate statistical analysis of DESI, the entire measurement
system can be amended and improved – especially the correlation dilemma can be
resolved. Second, based on our findings, a more refined public policy analysis,
design and execution may be projected, as adequate knowledge of causality between
variables helps decision makers to identify the root causes and actions for
improvement.
REFERENCES
ALONSO, M. A. P.– GARCÍA, J. C. S. (2018): Digitalization as Push and Pull Factor Redefining
the Entrepreneurship Concept Conference Proceedings Chapters. In: TODOROV,
K.–KOLAROV, K. (ed.): The International Entrepreneurship: Trends, Challenges,
Achievements Proceedings of the Eighth International Conference, 6–9 June 2017,
Varna, Bulg, edition 1, chapter 127, pp. 175–191., Bulgarian Association for
Management Development and Entrepreneurship, Varna.
CURKO, K.–CURIC, T.–VUKŠIC, V. B. (2017): Perspective of smart business development
International Journal of Renewable Energy Sources 2 (2): 40–47.
DUDEK, H.–SEDEFOĞLU, G. (2019): Modelling severe material deprivation rates in EU
regions using fractional response regression Regional Statistics 9 (2): 130–147.
https://doi.org/10.15196/RS090210
FERNANDEZ-CREHUET, J. M.–ROSALES-SALAS, J.–BARRAGÁN, G. G. (2019): Construction
and validation of an international reputation index: the European case Regional
Statistics 9 (2): 105–129. https://doi.org/10.15196/RS090205
62 Zoltán Bánhidi – Imre Dobos – András Nemeslaki
Regional Statistics, Vol. 10. No. 2. 2020: 42–62; DOI: 10.15196/RS100209
GÖTZ, M. (2017): Industry 4.0–the perspective of international economics. The case of
Polish-German relationships Przegląd Zachodni 365 (4): 169–185.
KONTOLAIMOU, A.–SKINTZI, G. (2018): 4.2. Digitisation patterns of the Greek economy
and society Greek Economic Outlook 37: 41–48.
MIRKE, E.–KAŠPAROVÁ, E.–CAKULA, S. (2019): Adults’ readiness for online learning in the
Czech Republic and Latvia (digital competence as a result of ICT education
policy and information society development strategy) Periodicals of Engineering and
Natural Sciences 7 (1): 205–215. http://dx.doi.org/10.21533/pen.v7i1.366.g255
EUROPEAN COMISSION (2018): DESI Report:
https://ec.europa.eu/digital-single-market/en/des (downloaded: 05.06.2019)
NIKOLOV, H. S.–KRUMOVA, M. Y. (2019): Hofstede’s model in the context of e-government
and open government in EU countries: countries clustering based on similarities
and differences Smart Cities and Regional Development (SCRD) Journal 3 (1): 29–46.
RUSSO, V. (2020): Digital Economy and Society Index (DESI). European Guidelines and
Empirical Applications on the Territory. In: SARASOLA SÁNCHEZ-SERRANO, J.
L.–, MATURO, F.– HOŠKOVÁ-MAYEROVÁ, Š. (eds.): Qualitative and Quantitative
Models in Socio-Economic Systems and Social Work pp. 427–442., Springer, Cham.
SCHMIDT, E.–COHEN, J. (2013): The New Digital Age: Reshaping the Future of People, Nations and
Business Alfred A. Knopf, New York.
SCUPOLA, A. (2018): Digital Transformation of Public Administration Services in Denmark:
A Process Tracing Case Study Nordic and Baltic Journal of Information and
Communications Technologies (1): 261–284.
https://doi.org/10.13052/nbjict1902-097X.2018.014
STOICA, E. A.–BOGOSLOV, I. A. (2017): A Comprehensive Analysis Regarding DESI
Country Progress for Romania Relative to the European Average Trend Balkan
Region Conference on Engineering and Business Education 3 (1): 258–266.
https://doi.org/10.1515/cplbu-2017-0034
TABACHNICK, B. G.–FIDELL, L. S.–ULLMAN, J. B. (2007): Using multivariate statistics (5th
Edition). Pearson, Boston, MA.
URS, N. (2018): E-government development in Romanian local municipalities: A
complicated story of success and hardships Transylvanian Review of Administrative
Sciences 14 (55): 118–129. http://dx.doi.org/10.24193/tras.55E.8
