Content uploaded by Rosina Moreno
Author content
All content in this area was uploaded by Rosina Moreno on Mar 12, 2014
Content may be subject to copyright.
DOI 10.1007/s00168-005-0021-y
ORIGINAL PAPER
Rosina Moreno .Raffaele Paci .Stefano Usai
Geographical and sectoral clusters
of innovation in Europe
Published online: 17 November 2005
© Springer-Verlag 2005
Abstract In this paper we attempt to provide empirical evidence on the phe-
nomenon of cluster agglomeration of innovation activities throughout time and
space in European regions. More specifically we try to assess whether there are
some forces which support the development of technologically specialised regional
clusters. In particular we want to determine the spatial extent of these forces, their
dynamics along the eighties and nineties and their connection with production
clustering. We have started from a mapping of innovation activity in European
regions by means of an exploratory spatial analysis based on global indicators of
spatial dependence. As a result, in a second step, we check the hypothesis that
innovation concentration can be a result not only of the geographic concentration of
production but also of the development of technologically specialised clusters in
neighbouring regions. The analysis is based on a databank set up by CRENoS on
regional patenting at the European Patent Office spanning from 1978 to 2001 and
classified by ISIC sectors and on the Cambridge Econometrics database on prod-
uction activity. Among the main results, it is shown that specialisation in innovative
activity is positively and significantly influenced by specialisation in production
activity. Additionally, it is obtained that innovation tends to cluster more in sectors
in which the neighbouring regions are also technologically specialised.
JEL Classification R11 .R12 .O31 .C21
R. Moreno (*)
AQR - Regional Quantitative Analysis Research Group,
Department of Econometrics and Statistics, University of Barcelona,
Diagonal, 690, 08034 Barcelona, Spain
E-mail: rmoreno@ub.edu
R. Paci .S. Usai
CRENoS, University of Cagliari, Via S. Ignazio 78, 09123 Cagliari, Italy
E-mail: paci@unica.it
E-mail: stefanousai@unica.it
Ann Reg Sci 39:715–739 (2005)
1 Introduction
The “First Action Plan for Innovation in Europe”, launched by the European
Commission in 1996, clearly states that, in spite of its excellent scientific
capabilities, Europe’s level of innovation is lower than that of its main competitors.
At a time when innovation is the main driving force in economic competitiveness
this has serious implications for employment and economic prosperity in Europe.
Innovation has therefore become a priority of European countries in order to start
and sustain the engine of economic growth. In the spatial context such an engine
may be fuelled both by the amount of technological activity which is carried out
locally and by external technological achievements channeled through information
spillovers (Martin and Ottaviano 2001; Grossman and Helpman 1991). Spillovers
may follow particular patterns depending on economic, technological and geo-
graphical distances among firms and regions, that is, on agglomeration phenomena
which apply both to production and innovation activities.
As a matter of fact, economic growth, technological change and urbanisation
have been, in the past, and are, nowadays, inseparable phenomena (see Baldwin
and Martin 2004). Most importantly, in recent years, an increasing concentration of
innovation activities in and around major urban centers has been noticed
(Audretsch and Feldman 1996) while other studies have highlighted that
innovation is much more polarized than production (see for example Paci and
Usai 2000, for the case of European regions). The increasing costs of conducting
advanced applied research, the concentration of large firms, public research
centers, top universities, and highly skilled human capital in large urban ag-
glomerations are factors that have contributed to this polarisation. Most im-
portantly, learning processes may be facilitated when economic actors have the
possibility to communicate face to face (Von Hipple 1994). At the same time,
according to the line of research started by Coe and Helpman (1995) and refined by
Keller (2002), there may appear important informational spillovers across nations,
due to the fact, for example, that the transmission of knowledge in space is
becoming less costly as a result of advances in telecommunications. As a matter of
fact, Moreno et al. (2005) for Europe and Varga et al. (2005) for the USA have
shown that innovative activity in a certain region may be affected by similar
activities in contiguous regions. Consequently, the spatial extent of such a process
of polarisation is an empirical question which should embrace a much more
complex picture that one of simple concentration or simple delocalisation. A
picture which is a combination of both phenomena, concentration and de-
localisation, which depend on factors such as countries’institutional context and
sectoral characteristics which may affect in different ways, among others, capital
and labour mobility and local human capital accumulation (Mariani 2002).
In particular, in this scenery one has to consider the relationship between the
localisation process of innovation activities and that of production. There are, as a
matter of fact, benefits and costs of doing research close to production. Among the
benefits one may think of the continuous exchange of ideas and information
between the plant and the laboratory and a higher relationship between the
innovation results and the production necessities, whereas among the costs we find
the congestion costs associated to dense agglomerations. The net result may not
compensate the advantages of concentrating production in the areas with strong
local technological economies. Again the trade off is contingent on a number of
716 R. Moreno et al.
factors which attain, for example, to the scientific content of the research and/or the
relative factor intensity of the production process.
This paper aims at studying the phenomenon of agglomeration and
specialisation of innovative activities and its relationship with the agglomeration
of production activities starting from a mapping of innovative performance in
European regions by means of an exploratory spatial investigation. The analysis is
carried out for different time periods starting from the early eighties up to the
beginning of the 21st century and it is implemented for different sectors in order to
evaluate differences and similarities across them. Evidence in favour of the
presence of common specialisation patterns in production and innovation does not
exclude that spillovers may also occur across other than within borders, so that
externalities could cross the geographical borders of regions. We believe that
externalities are neither totally localized nor totally global and we expect them to
depend on the geographical distance among regions.
1
Following these lines, in a
second step, while controlling for the extent to which the specialisation of
production in certain sectors influences the specialisation of innovation, this paper
analyses the role which geographical technological spillovers play in innovation
concentration in some industries. In other words, after the influence that
geographic concentration of production has on innovation concentration has
been controlled for, we address the role that the development of technologically
specialised clusters in neighbouring regions may play. Econometric techniques are
going to allow us to assess the presence and strength of such phenomena.
The analysis is performed at the regional level given that, on the one hand,
innovation policies are often implemented at this territorial level (even though
within a national framework); on the other hand because, as noted above,
technological activities appear strongly localized into clusters of innovative firms.
As argued by Storper (1995, p. 896) this is, as a matter of fact, the geographical
level “at which technological synergies are generated and to which any national
technology policy must therefore be addressed”. As a result, even accepting that
there is need for a global approach to innovation, we try to handle it by considering
important diversities across nations, regions, sectors and time. This aspect is
addressed directly thanks to an original and updated statistical databank on regional
patenting at the European Patent Office spanning from 1978 to 2001 and classified
by ISIC sectors (up to three digits). This database allows the analysis of the spatial
distribution of innovative activity across 175 regions of 17 countries in Europe (the
15 members of the European Union plus Switzerland and Norway) in a set of seven
manufacturing sectors.
The paper is organised as follows. In the following section we provide a
discussion on the quality of the technology indexes used in this paper. The third
section analyses the spatial distribution of innovative activity and specialisation
patterns throughout Europe along the 80s and 90s. In the fourth section we estimate
a model of specialisation patterns in which knowledge interactions are included.
Final remarks are in the last section.
1Additional to the possibility of externalities crossing geographical barriers of regions due to
proximity in space, interregional spillovers may take place due to other reasons such as the
volume of trade between each pair of regions or their economic similarity.
Geographical and sectoral clusters of innovation in Europe 717
2 Some issues about technology measurement
Several contributions in the past have made extensive use of patent statistics in
order to analyse the spatial distribution of innovation activity. In particular, in the
case of European regions Breschi (2000) and Caniels (2000) have provided an
articulated and extensive analysis of the spatial distribution of innovation in
European regions until the nineties whilst Paci and Usai (2000) have tried to
address the same issue of agglomeration of innovation and production for a smaller
set of countries. These precedents should not let one forget that the use of patents as
indicators of innovative activity implies some inconveniences and shortcomings
which ought to be kept in mind while interpreting the outcome of the analysis, both
descriptive and econometric.
Several economists have been debating the issue of measuring innovative
activity and technological progress, but no universal solution has been found
(Griliches 1990). Based on the concept of knowledge production function two
types of indicators are usually identified: technology input measures (such as R&D
expenditure and employees) and technology output measures (such as patents and
new product announcements).
The main drawback of the former indicators is that they embrace firms’efforts
for invention and innovation together with imitation activities. Moreover, they do
not take into account informal technological activity and, as a consequence, may
underestimate the amount of innovative activity. On the contrary, the latter
represents the outcome of the inventive and innovative process even though there
may be inventions which are never patented as much as patents which are never
developed into innovations. However, the patenting procedures require that
innovations have novelty and usability features and imply relevant costs for the
proponent. This implies that patented innovations, especially those extended
in foreign countries, are expected to have economic value, although highly
heterogeneous.
With respect to the object of our research,
2
that is to study innovative activity
across regions, sectors and time, patent statistics seem particularly suitable, due to
some useful properties compared to R&D data which are summarised below:
(a) They provide information on the residence of the inventor and proponent and
can thus be grouped regionally ( potentially at different territorial units starting
from zip areas), whereas R&D statistics are available just for some regions or at
the national level;
(b)They record the technological content of the invention and can, thus, be
classified according to the industrial sectors whilst R&D data is usually
aggregated, especially at the regional level;
3
2Note that since 2000 there is an important initiative called European trend chart on innovation
which provides several indicators on innovation (based on input and output data and on the CIS
survey) at the regional level and a synthetic measure of them. Unfortunately, the time and the
sectoral dimension of such a database are rather limited. Nevertheless for the time being this
database is going to become more and more a crucial point of reference for the analysis in this
field.
3It should be noted that R&D statistics provide other interesting information concerning the
origin of the expenditure. R&D statistics are, as a matter of fact, usually divided into categories
such as business, university and government.
718 R. Moreno et al.
(c) They are available year by year for a long time span and this allows for a
dynamic analysis. On the contrary regional R&D data is available only for
recent years and discontinuously.
Our proxy for innovative activity refers to patents applications at the European
Patent Office over the period starting from 1978 until 2001 classified by the
inventor’s region in Europe. Applications at EPO should provide a measure of
sufficiently homogenous quality, due to the fact that applying to EPO is difficult,
time consuming and expensive. This indicator, in other words, should prove
particularly effective in order to take into account potentially highly remunerative
innovations which for this reason are patented abroad. The use of the inventor’s
residence, rather than the proponent’s residence, is preferred in order to attribute the
spatial localisation of each innovation. Indeed, the latter generally corresponds to
firms’headquarters and therefore it might lead to an underestimation of peripheral
regions’innovative activity whenever the invention has been developed in a firm’s
subsidiary located in another area.
4
Moreover, differently from previous research
(Bottazzi and Peri 2003) we do not assign patents just to the first inventor, given
that this may bias our result as inventors are usually listed in alphabetical order. For
the case of patents with more than one inventor, therefore, a proportional fraction of
each patent is assigned to the different inventors’regions of residence.
As for the territorial break up we have only partially followed the classification
provided by EUROSTAT through NUTS (Nomenclature des Unités Territoriales
Statistiques).
5
For some countries, this classification turns out to be artificial, based
mainly on statistical concerns while failing to identify uniform regional areas in
terms of economic, administrative and social elements. In fact, we have tried to
select, for each country, a geographical unit with a certain degree of administrative
and economic control.
6
The result is a division of Europe (15 countries of the
European Union plus Switzerland and Norway) in 175 sub-national units (which,
from now on, we will simply call, regions) which are a combination of NUTS 0, 1
and 2 levels (see Appendix for details).
As far as the sectoral classification is concerned, it is known that patent data are
still of limited use for economic analysis due to their mode of classification which
is different from the one used for production: innovations are recorded for
administrative purposes using the International Patent Classification (IPC) system,
which categorizes inventions by product or process. On the contrary, most
4For instance, the headquarter of Enichem, the Italian petroleum and chemical multinational, is
located in Milan (Lombardia) but the innovative activity (as indicated by the residence of the
inventors) is much more dispersed due to the presence of several plants in other regions (e.g.
Veneto, Sicilia, Liguria and Sardegna).
5Eurostat classification list four categories of territorial units: 15 NUTS 0 nations; 77 NUTS 1
regions, 206 NUTS 2 regions and 1031 NUTS 3 regions.
6The perfect territorial unit is difficult to be found since administrative units do not necessarily
reflect economic phenomena. Better territorial units used in the empirical literature are the
functional urban region just for main urban centres at the European level (Cheshire 1990), the
local labour system in Italy (Paci and Usai, 1999) or the basin d’emploi in France (Combes,
2000).
Geographical and sectoral clusters of innovation in Europe 719
economic analyses are interested in the particular sectors of the economy
originating the invention or implementing it. For this reason patent data, originally
classified by means of the IPC, have been converted to the industry of manufacture
thanks to the Yale Technology Concordance
7
. Such a concordance uses the
probability distribution of each IPC or product code across industries of
manufacture in order to attribute each patent proportionally to the different sectors
where the innovation may have originated.
3 The geography of innovative activity
3.1 Mapping of innovation in Europe
At the beginning of the period under consideration (early eighties) a strong central-
periphery distribution of innovation activity is observed (Map 1a).
8
Innovation
activity is mainly concentrated in the very core of Europe, a cluster of regions
which includes the whole of Switzerland, West Germany, Luxembourg and most
regions of Austria. There are also some other hot spots of innovation in the North
and East of France, the South-East part of United Kingdom, in the Netherlands and
in some Scandinavian countries, mostly in Sweden. None or modest technological
activity is documented in most regions of the South of Europe: Spain, Greece,
Portugal and South of Italy. Innovative backwardness is also documented for some
northern countries such as Norway and Ireland.
This picture is confirmed while looking at the innovative activity at the country
level (Table 1) and among the 20 most innovative regions according to the ranking
at the beginning (Table 2a) and at the end of the period under analysis (Table 2b).
At the beginning of the eighties the most innovative country is by far Switzerland,
with 14.5 patents per 100.000 inhabitants, followed by Germany (8.3) and
Luxembourg (7.2). A similar picture appears at the regional level, where, among
the top performers, we find six Swiss regions, nine German regions, two Swedish
regions, Luxembourg and Ile de France (which hosts Paris) and Zuid Nederland
(where Philips HT research center is located).
Looking at the evolution over time of the innovative activity, it is possible to
remark some important elements. First, the intensity to innovate has increased
considerably over the two decades in all countries: the average innovative output
was 3.6 patents per 100,000 inhabitants in the early eighties and it was almost three
times bigger (10.4) at the end of the nineties.
9
As regards the level of inequality in
the spatial distribution of the innovative activity, this is clearly very high: the ratio
between the most innovative country (Switzerland) and the least (Portugal) in the
period 1999–2001 is still equal to 93. The coefficients of variation reported in
the two last rows of Table 1, both for nations and regions respectively, shows that the
7The original YTC was conceived by Evenson et al. (1991). Updates to the YTC have been
programmed by Daniel Johnson who kindly provides downloadable conversion tables and
detailed explanations on the procedures at the Internet address: http://faculty1.coloradocollege.
edu/~djohnson/jeps.html.
8Throughout the paper patents per capita are used, even though main results do not change if one
uses the absolute value of patents.
9This phenomenon is partly due to a shift of patent applications by European firms from National
patenting offices to the European one.
720 R. Moreno et al.
former has gone from 1.05 to 0.71 and the latter from 1.46 to stabilize around 1.04 in
the last period. Most importantly, the innovations have been spreading to some more
regions starting from the core described above. It is clear that such a phenomenon
has involved mainly the whole of France, Belgium, Denmark, the North of Italy, a
few Northern regions in Spain and most importantly the South Finland and almost
the whole of Sweden (see Map 1b, c, d).
Figure 1presents some detail concerning the process of divergence/conver-
gence of innovative activity across the 175 regions both for the total of
manufacturing and for some sectors. In general, the coefficient of variation in the
patenting activity among the 175 regions for the Manufacturing and the energy
sector is around 1.6 in 1981 but descends gradually to around 1.00 at the end of the
period (see the top-left panel in Fig. 1). Such a regular decline in the geographical
concentration of innovative activity is a common feature of some macro-sectors,
Map 1 Distribution of innovative activity in the European regions (patents per 100,000
inhabitants, annual average). Panel a (1981–1983), Panel b (1989–1991), Panel c (1994–1996),
Panel d (1999–2001)
Geographical and sectoral clusters of innovation in Europe 721
such as Electronics and Fuels, chemical and rubber. In some other sectors, such as
Food, beverages and tobacco, Mining and energy supply and Transport equipment,
there appears to exist some changes over time, without a clear explanation although
with a common feature of lower values for dispersion at the end of the period, while
Textiles and clothing and the residual sector of Other manufacturing (together to
the one of Construction) show a rather constant pattern throughout the period. It
seems, therefore, that the innovation carried out in sectors with a high technological
component was much localised in some specific regions at the beginning of the
eighties but have experimented a more clear spread to other regions over time than
in sectors with lower technological component.
One other way to look at the dynamics of spatial diffusion of technological
activity is to analyse the distributions of the patents per capita through the kernel
density functions for the four periods under examination, as reported in Fig. 2.Itis
clear that the distribution is skewed to the lower values of patents during all the
periods, whereas the outliers are in the upper band of patents (basically some
regions in Switzerland and Germany). However, the kurtosis is much stronger at
Table 1 Innovative activity in European countries (patents per 100,000 inhabitants, annual
average)
Nation Num. of
regions
Period
1981–1983 1988–1990 1994–1996 1999–2001
Pat pc ranking Pat pc ranking Pat pc ranking Pat pc ranking
Switzerland 7 14.5 1 20.9 1 19.7 1 27.8 1
Germany 40 8.3 2 14.7 2 12.2 2 19.9 2
Sweden 8 6.5 4 8.3 4 11.7 3 18.7 3
Finland 6 1.4 11 4.7 10 9.6 4 18.3 4
Netherlands 4 4.1 5 8.3 3 8.3 5 14.5 5
Denmark 1 2.5 9 4.8 9 7.6 6 12.9 6
Luxembourg 1 7.2 3 5.0 8 6.4 10 12.7 7
Austria 9 3.3 8 6.8 6 6.8 8 10.5 8
Belgium 3 2.2 10 4.5 11 6.6 9 10.1 9
France 22 3.9 6 6.8 5 7.1 7 9.8 10
United
Kingdom
12 3.4 7 5.4 7 5.1 11 7.3 11
Norway 7 0.9 13 2.1 13 3.0 13 5.1 12
Italy 20 1.1 12 3.0 12 3.4 12 5.0 13
Ireland 2 0.5 14 1.3 14 1.9 14 4.2 14
Spain 15 0.1 15 0.5 15 0.8 15 1.5 15
Greece 13 0.1 16 0.1 16 0.2 16 0.4 16
Portugal 5 0.0 17 0.1 17 0.1 17 0.3 17
EU 175 3.6 6.5 6.7 10.4
CV across
nations 1.05 0.91 0.75 0.71
CV across
regions 1.46 1.17 1.05 1.04
CV refers to coefficient of variation
722 R. Moreno et al.
Table 2 Innovative activity in top 20 regions (patents per 100,000 inhabitants, annual average)
Region Nation Period
1981–1983 1988–1990 1994–1996 1999–2001
Pat pc ranking Pat pc ranking Pat pc ranking Pat pc ranking
a
Nordwestschweiz CH 34.13 1 38.9 1 32.8 1 42.4 3
Zurich CH 18.40 2 27.4 3 24.7 5 36.6 5
Oberbayern DE 18.08 3 29.1 2 26.9 2 50.4 1
Rheinhessen-Pfalz DE 18.01 4 24.9 5 26.0 4 32.5 7
Darmstadt DE 17.90 5 26.8 4 26.0 3 32.2 8
Koln DE 14.97 6 19.9 10 17.4 13 24.8 16
Region Lemanique CH 12.61 7 14.3 19 14.8 20 21.8 25
Karlsruhe DE 12.09 8 21.0 8 19.9 8 29.6 11
Ile de France FR 10.74 9 15.9 16 16.1 16 22.1 23
Dusseldorf DE 10.37 10 20.1 9 15.9 17 22.7 21
Stockholm SE 10.28 11 13.0 21 20.3 7 30.8 10
Mittelfranken DE 9.95 12 22.3 6 17.8 12 31.0 9
Stuttgart DE 9.50 13 21.8 7 23.9 6 43.3 2
Ostschweiz CH 9.44 14 19.0 11 17.2 14 25.5 15
Espace Mittelland CH 9.18 15 14.7 17 15.1 19 21.9 24
Sydsverige SE 9.13 16 9.2 32 12.4 25 23.1 20
Freiburg DE 8.50 17 16.3 15 18.7 11 29.3 12
Zuid-Nederland NL 8.15 18 18.4 12 15.5 18 36.8 4
Zentralschweiz CH 7.39 19 17.4 13 19.9 9 24.5 17
Luxembourg LU 7.16 20 5.0 66 6.4 59 12.7 44
b
Oberbayern DE 18.08 3 29.1 2 26.9 2 50.4 1
Stuttgart DE 9.50 13 21.8 7 23.9 6 43.3 2
Nordwestsschweiz CH 34.13 1 38.9 1 32.8 1 42.4 3
Zuid-Nederland NL 8.15 18 18.4 12 15.5 18 36.8 4
Zurich CH 18.40 2 27.4 3 24.7 5 36.6 5
Uusimaa FI 3.48 49 9.4 30 19.5 10 35.5 6
Rheinhessen-Pfalz DE 18.01 4 24.9 5 26.0 4 32.5 7
Darmstadt DE 17.90 5 26.8 4 26.0 3 32.2 8
Mittelfranken DE 9.95 12 22.3 6 17.8 12 31.0 9
Stockholm SE 10.28 11 13.0 21 20.3 7 30.8 10
Karlsruhe DE 12.09 8 21.0 8 19.9 8 29.6 11
Freiburg DE 8.50 17 16.3 15 18.7 11 29.3 12
Tubingen DE 6.39 25 14.4 18 16.7 15 28.6 13
Vorarlberg AT 2.52 64 11.3 25 13.3 22 25.8 14
Ostschweiz CH 9.44 14 19.0 11 17.2 14 25.5 15
Koln DE 14.97 6 19.9 10 17.4 13 24.8 16
Zentralschweiz CH 7.39 19 17.4 13 19.9 9 24.5 17
Geographical and sectoral clusters of innovation in Europe 723
the beginning of the 80s, with a clear smoothing process in late 80s and mid-90s, so
that the right-hand tale becomes thicker, in other words, more regions are obtaining
output in the innovative activity.
This pattern is the result of different performances by countries and regions:
there has been some catching up as much as some falling behind. For example
Region Nation Period
1981–1983 1988–1990 1994–1996 1999–2001
Pat pc ranking Pat pc ranking Pat pc ranking Pat pc ranking
Unterfranken DE 5.42 29 10.7 26 14.7 21 23.2 18
Braunschweig DE 3.12 55 8.0 39 6.6 57 23.2 19
Sydsverige SE 9.13 16 9.2 32 12.4 25 23.1 20
Pat pc refers to patents per capita
In Table 2a the ranking refers to the first period, whereas in Table 2b the ranking refers to the last
period
Table 2 (continued)
0.00
0.05
0.10
0.15
0.20
-10 0 10 20 30 40 50
1981-83
-10 0 10 20 30 40 50
1989-91
-10 0 10 20 30 40 50
1994-96
-10 0 10 20 30 40 50 60
1999-01
Fig. 2 Kernel density function for innovation activity (patent per 100,000 inhabitants, annual
average)
Fig. 1 Coefficient of variation for innovation activity across European regions (1981–2001)
724 R. Moreno et al.
among the catching up process, it is worth highlighting the most brilliant one
shown by Finland, which in the nineties manages to reach the fourth position in the
country ranking (Table 1) and to put its capital region Uusimaa among the first
producers of innovation in Europe. This region was 49th at the beginning of the
eighties and sixth at the end of the nineties, originating one of the most remarkable
catching up performance in Europe in the last 20 years (see Table 2b).
The comparative examination of Table 2a and 2b, moreover, is rather
informative about the relatively great reshuffle of regions. Table 2a for example
tells us that even though 14 out of 20 innovative regions have managed to keep a
ranking among the 20 most innovative regions from 1981–1983 to 1999–2001,
there have been some remarkable declines (that of Luxembourg which goes from
the 20th place to 44th). Most interestingly, Table 2b tells us that 15 regions which
are in the top 20 in the latest period were already there in the early eighties.
However there are some interesting stories to pinpoint, other than that of Uusimaa.
Stutgart and Zuid Nederalnd, for example, were in the 13th and 18th position and
are now second and fourth. Voralberg (that is the most western Austrian region in
between Switzerland and Germany) was 64th and it is now 14th. All in all, Table 2
illustrates that among the top regions the German leadership has been strengthened
(11 regions out of 20 are German) whilst the Swiss regions have lost some ground
(they were six and they are now four).
Among the declining countries the most remarkable cases are the one of the
United Kingdom which goes from the seventh to the eleventh position and the one
of France which moves from the sixth to the tenth ranking. It should be, however,
noted that the two cases are different since in the latter there are still one champion
region, that of Ile de France which has the 23rd rank. On the contrary the first
British region in the ranking is Eastern which features in 39th position. Finally, no
notable improvement is shown by the followers, in other words, countries such as
Italy, Norway, Spain, Portugal and Greece.
3.2 Innovation clusters in Europe
All the evidences gathered in the tables and maps analysed in the previous section
show that innovative activity is relatively concentrated in few areas in Europe. We
examine now whether the spatial concentration of innovation activity observed
from the maps generates a process of spatial dependence. In other words, to what
extent the technological activity performed in one region is associated to the one in
neighbouring regions. The degree of spatial association can be analysed by means
of the Moran’s I statistic, which is defined as:
I¼N
S0P
N
iP
N
j
wij xi
xðÞxj
x
P
N
i¼1
xi
xðÞ
2
where x
i
and x
j
are the observations for region iand jof the variable under analysis,
patents in our case;
xis the average of the variable in the sample of regions; and w
ij
is the i–jelement of the row-standardised Wmatrix of weights. S0¼PiPjwij is a
Geographical and sectoral clusters of innovation in Europe 725
standardisation factor that corresponds to the sum of the weights. The most general
specification for the weight matrix is the physical contiguity one, given rise to a
binary and symmetric matrix where its elements would be 1 in the case of two
regions sharing a boundary and 0 otherwise. In the case of a row-standardised W
matrix, in which each element in a row is divided by the total sum of the row, S
0
equals the number of observations, N, so that N/S
0
is equal to 1.
The values for the Moran’s index for the seven manufacturing sectors as well as
for different physical contiguity matrices (1st, 2nd and 3rd order neighbours) are
presented in Table 3. The use of the Moran index for the total manufacturing sector
(see first rows in Table 3) shows a clear rejection of the null hypothesis with a
positive value of the statistic: there appears a strong positive spatial autocorrelation,
confirming the visual impression of spatial clustering given by the maps. If one
also considers the spatial correlogram, this rejection is observed till the third order
of contiguity—1st, 2nd and 3rd order neighbours—as reported also in Table 3.
Nonetheless, there also appears a pattern of decreasing autocorrelation with
increasing orders of contiguity typical of many spatial autoregressive processes.
Table 3 Spatial autocorrelation in the innovative activity (Moran’s I test, normal approximation)
Period 1981–1983 1988–1990 1996–1996 1999–2001
Sector Contiguity
matrix
Z-value Prob Z-value Prob Z-value Prob Z-value Prob
Total
manufacturing
1st order 3.435 0.00 4.111 0.00 4.327 0.00 4.493 0.00
2nd order 2.850 0.00 3.581 0.00 4.170 0.00 4.256 0.00
3rd order 3.357 0.00 3.424 0.00 3.672 0.00 3.527 0.00
Mining and
energy
1st order 6.789 0.00 5.135 0.00 5.604 0.00 0.835 0.40
2nd order 4.825 0.00 3.036 0.00 3.680 0.00 0.822 0.41
3rd order 1.283 0.20 0.402 0.69 0.888 0.37 -0.004 1.00
Food 1st order 8.878 0.00 10.313 0.00 10.407 0.00 11.317 0.00
2nd order 8.176 0.00 9.430 0.00 9.263 0.00 5.349 0.00
3rd order 5.777 0.00 8.346 0.00 7.224 0.00 -1.002 0.32
Textile and
clothing
1st order 7.482 0.00 7.923 0.00 5.670 0.00 2.783 0.01
2nd order 5.450 0.00 5.836 0.00 3.801 0.00 5.582 0.00
3rd order 3.814 0.00 4.621 0.00 3.399 0.00 3.569 0.00
Chemicals
and plastic
1st order 3.567 0.00 3.809 0.00 3.304 0.00 3.375 0.00
2nd order 2.162 0.03 2.383 0.02 2.394 0.02 2.619 0.01
3rd order 2.492 0.01 3.300 0.00 3.501 0.00 3.255 0.00
Electronics 1st order 3.335 0.00 2.409 0.02 3.013 0.00 2.785 0.01
2nd order 2.793 0.01 1.835 0.07 2.418 0.02 2.251 0.02
3rd order 2.305 0.02 1.606 0.11 1.725 0.08 1.803 0.07
Transport
equipment
1st order 10.404 0.00 10.308 0.00 9.365 0.00 3.496 0.00
2nd order 8.532 0.00 8.290 0.00 7.162 0.00 3.245 0.00
3rd order 5.484 0.00 6.079 0.00 5.221 0.00 1.457 0.15
Other
manufacturing
1st order 4.453 0.00 5.649 0.00 7.924 0.00 4.911 0.00
2nd order 3.959 0.00 4.682 0.00 6.683 0.00 4.466 0.00
3rd order 3.750 0.00 3.858 0.00 4.260 0.00 2.493 0.01
Number of observations: 175
726 R. Moreno et al.
We have also constructed the scatter maps in order to assess the sign of the
spatial association in the different areas and its evolution along time (see Map 2,
panel a and b). The scatter maps show that there is a clear association of high–high
values in the centre, and low–low values in the South. This positive association
remains true throughout the period, with few exceptions: some regions in the North
of Italy initially showed high value of patents surrounded by low values whilst in
the nineties became a cluster of high values. Additionally, Finland has performed
remarkably well along this period, presenting low values at the beginning
surrounded by low values, but changing to high values. This pattern shows almost
no difference over time.
10
The presentation of the aggregate geographic distribution of innovative activity in
Europe does not give information of the propensity for innovation to cluster spatially
within specific sectors. However, the database on patenting allows one to investigate
the geographical distribution of innovative activity also sector by sector in order to
see if agglomeration forces depend on sectoral characteristics. In Map 3the sector
with the highest revealed technological advantage index is used to define the
specialisation in European regions at the beginning of the 80s (panel a) and at the end
of the 90s (panel b).The technological specialisation index is measured as follows:
ISTij ¼
Pij,P
M
j¼1
Pij
P
N
i¼1
Pij,P
M
j¼1P
N
i¼1
Pij
where iindexes the region (i=1,..., N), jindexes the industrial sector ( j=1,...,M)and
Pstands for patents in the considered period. The mapping, among other interesting
evidences, shows that there seem to be some clusters of common technological
Map 2 Scatter for innovative activity in the European regions (patents per 100,000 inhabitants,
annual average). Panel a (1989–1991); Panel b (1999–2001)
10 Scatter maps for other periods not reported in the paper are available on request.
Geographical and sectoral clusters of innovation in Europe 727
specialisation patterns: textiles and clothing in Italy, Fuels, chemicals and rubber in
Germany, Food and beverages in Northern Europe.
Also, the distribution of innovative activity for the seven macro-sectors under
analysis is given in Table 3where we have reported the Moran’s tests for spatial
autocorrelation in each of them. The sectoral results confirm the presence of spatial
association up to the third contiguity order for all sectors. Another interesting issue is
to analyse in which sectors the autocorrelation of innovation is considerably greater
or lower than that for the Total manufacturing sector. At the beginning of the period
under analysis, the sectors of Mining and energy, Food, Textile and clothing and
Transport equipment presented a higher value of the Moran’s statistic than that for
the sector of Total manufacturing, that is, concentration in space in these sectors was
more important than for the entire manufacturing industry. The opposite is obtained
in the cases of Chemicals and plastic and Electronics, although the spatial
autocorrelation encountered in those cases is also significant. However, at the end of
the period, the value of the Moran’s index becomes more similar in the different
sectors, with prevalence of significant values of the test. All in all, this means that
patenting activity in a certain sector tends to be correlated to patenting performed in
the same sector in contiguous areas, determining the creation of specialised
clustering of innovative regions in different sectors. This suggests that the analysis
of technological spillovers and sectoral interdependences across regions is a
promising way forward in the study of the specialisation of innovation. In the next
section a first attempt in this direction is done by means of an empirical model.
4 Model and results
In this section we investigate the phenomenon of agglomeration of innovation
activities throughout time, space and sectors in European regions. We try to assess
which are the forces which support the development of technologically specialised
regional clusters. In particular we would like to assess the spatial extent of these
Map 3 Index of technological specialisation (top sector) in the European regions (annual
average). Panel a (1989–1991), Panel b (1999–2001)
728 R. Moreno et al.
forces, their dynamics along the eighties and nineties and their connection with
production clustering.
In the paper by Jaffe et al. (1993) it is highlighted that one possible explanation
why innovation in some sectors tends to cluster geographically more than in other
sectors is that the location of production is more concentrated spatially. This being
true, whenever one analyses why the propensity for innovative activity to cluster
geographically changes across sectors, it is needed to control for the geographic
concentration of the location of production activity. However, even after accounting
for the geographic concentration of the production specialisation, as done in the re-
gression analysis below, an interesting point to be analysed is to what extent the
specialisation of innovative activity in one region is influenced by the specialisation
pattern in neighbouring regions. In other words, which is the role played by inter-
regional technological spillovers in sectoral specialisation in the geographical space.
Following the ideas above, we want to analyse the extent to which the
innovation specialisation of a given sector in a given region is influenced by the
level of specialisation in the production activity in the same region and sector and
the level of technological specialisation in the same sector in the nearby regions.
The model to be estimated is therefore as follows:
ISTijt ¼0þ1ISPijt1þ2WðrÞISTijt1þX
17
N¼1
NNATN
þX
7
S¼1
SSECTSþ"ijt
(1)
where IST
ij
represents the relative technological specialisation index of region iin
sector jas presented in Section 3.2, which is the result of a double weighting of the
regional sectoral innovation activity (measured through patents), with respect to the
total innovation in the region and with respect to the European quota of that sector.
As outlined above, such an indicator is considered to be a function of the presence
of production specialisation in the same sector within the same region by means of
ISP
ij
which is the relative production specialisation index of region iin sector j. The
same indicator as described above is used to measure this production specialisation
index, with employment as the variable used. However, in the specification given
in (1) innovation concentration is expected to be influenced by other variables.
Specifically, based on the theoretical ideas given in the introduction, we include a
variable proxying for the influence of interregional technological spillovers in the
same sector, this variable being a weighted average of the specialisation index in
the same sector of nearby regions (W(r)IST
ij
), where rindicates different order of
lags for the weight matrix.
Moreover, our general framework given in (1) introduces an additional vector
of factors which may also have a significant effect on the specialisation of the
innovative activity and that take into account potential omitted variables. So, firstly
we attempt to control for institutional environment and other structural factors
common to all the regions belonging to a nation, which may affect innovation
specialisation, through the use of a set of national dummies, NAT. Additionally,
with the aim to control for the different technological opportunities of the sectors
under consideration, a set of sectoral dummies, SECT, referred to the seven
manufacturing industries is also included.
Geographical and sectoral clusters of innovation in Europe 729
The regression analysis is performed as a cross-section for three different time
periods, that is, tis equal to 1989–1991, 1994–1996 and 1999–2001 in each case,
so that one can assess the evolution of the parameter under examination, if any. In
order to avoid endogeneity problems we consider independent variables at time t
−1, which refer to periods 1981–1983, 1989–1991, 1994–1996, respectively.
In the spatial econometrics literature the “classical”specification search
approach (specific to general or bottom-up approach) has been used almost
exclusively, while null attention has been paid to the so-called Hendry approach
(backward step-wise regression approach). Additionally, Florax et al. (2003)
demonstrate that the classical approach is found to slightly outperform the Hendry
approach in the case of the estimation of linear spatial models. All this leads us to
follow the classical specification search approach in which the initial model as in (1)
is estimated by means of OLS and a subsequent check for spatial dependence is
made. The tests for spatial autocorrelation in the residuals-the Moran’s I given by
Moran (1948), the LM-ERR suggested by Burridge (1980) and the LM-LAG
proposed by Anselin (1988)—are used to assess the degree to which remaining
unspecified spatial autocorrelation may be present in the regression. If the null
hypothesis of non-spatial dependence is rejected, our proposal is to correct such
misspecification. On the contrary, if the tests lead to the non-rejection of the null of
no spatial correlation among the residuals the ultimate model is the one given in (1).
Table 4summarises econometric results. We have estimated three equations for
each period with a pool for 175 regions and seven sectors, so that the final set up
refers to an estimation model with 1,225 observations. The first column in the
estimation of each period refers to the case in which the weighted average of the
index of specialisation of technology in the neighbours is obtained with a weight
matrix referring to the first order of contiguity. The second and third columns refer
to the second and third order of contiguity, respectively.
Some results are interesting to highlight. First of all, the relationship between
production and innovation specialisation proves positive and significant. Most
importantly such a link is getting stronger over time, the value of the coefficient being
0.11 in the first period under considerationand 0.20 in the last one. This result would
be an indication that research labs tend to stay closer to production plants, confirming
some previous findings on Europe (Paci and Usai 2000) and as a consequence
against those obtained by Audretsch and Feldman (1996) for the case of US states.
Secondly, the positive and significant coefficient of the spatial lag of the index of
technological specialisation suggests that the innovation specialisation of one region
is related to the specialisation of close-by regions. Thus, even after controlling for the
influence of sectoral specialisation in production in the same region, innovation
tends to cluster more in sectors in which the neighbouring regions are also
technologically specialised. In other words, technological spillovers play a decisive
role in the geographical configuration of industrial specialisation patterns.
Since we have considered second- and third-order lags (second and third
columns in the estimation for each time period) of the variable reflecting the
interregional effect on specialisation, it is observed that such a relationship is
significant until the second order of contiguity. So, technological specialisation in
one region depends not only on the technological specialisation of first-order
neighbouring regions but also on the technological specialisation of the regions
sharing a border with these first-order neighbours, although with a considerably
lower magnitude of this influence. Spillovers stop at this level given that the third-
730 R. Moreno et al.
Table 4 Econometric results
Variables OLS estimation
1989–1991 1994–1996 1999–2001
ISP
t−1
0.108 0.097 0.095 0.175 0.172 0.174 0.197 0.178 0.177
0.007 0.015 0.019 0.000 0.000 0.000 0.000 0.000 0.000
W (1) IST
t−1
0.182 0.174 0.175 0.281 0.259 0.259 0.187 0.152 0.151
0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.002 0.002
W (2) IST
t−1
0.022 0.021 0.016 0.018 0.034 0.033
0.003 0.005 0.063 0.039 0.000 0.000
W (3) IST
t−1
0.003 −0.009 0.002
0.602 0.233 0.846
LIK −397.3 −392.8 −392.7 −342.0 −340.3 −339.5 −339.8 −332.3 −332.6
AIC 844.5 837.7 839.4 734.1 732.5 733.1 729.6 717.2 719.1
Moran’s I 3.350 3.178 3.205 0.909 0.994 1.304 1.478 1.530 1.544
0.001 0.001 0.001 0.364 0.320 0.301 0.140 0.126 0.123
LM-ERR 6.650 5.721 5.770 0.032 0.061 0.073 0.543 0.598 0.596
0.010 0.017 0.016 0.859 0.805 0.787 0.461 0.439 0.440
LM-LAG 17.572 8.752 8.669 1.383 1.080 1.168 2.819 2.062 2.034
0.000 0.003 0.003 0.240 0.299 0.280 0.093 0.151 0.154
Dependent variable: technological specialisation index (IST)
Number of observations: 1,225 (175 regions and seven sectors). National and sectoral dummies included. P-values are given in italics
Geographical and sectoral clusters of innovation in Europe 731
order contiguity lag is non-significant. This result confirms other indirect evidence,
within the setting of the knowledge production function, on the relationship among
technological activities performed by contiguous regions (see for example Bottazzi
and Peri 2003; Moreno et al. 2005).
The tests for spatial autocorrelation in the residuals, the Moran’s I and the LM
tests, are computed for a physical contiguity matrix, that is, a binary and symmetric
matrix with elements equal to 1 in case of two regions having a common border. As
observed by the values of these statistics, spatial correlation is a problem which
seems to be present in the first period whilst it does not appear as a problem in the
residuals of the estimation for the last two periods.
In order to assess the robustness of our results to the problem of spatial
autocorrelation and to the applied estimation method we have also estimated a
second set of regressions. In such a set the spatial lags of the technological
specialisation indexes are considered with respect to the same period of the
dependent variable. In other words, the estimated equation is as follows:
ISTijt ¼0þ1ISPijt1þ2WðrÞISTijt þX
17
N¼1
NNATN
þX
7
S¼1
SSECTSþ"ijt
(2)
Accordingly, this spatial lag term has to be treated as an endogenous variable and
proper estimation methods have to account for this endogeneity. The most widely
used alternative method is Maximum Likelihood (ML) since OLS estimators are
biased and inconsistent due to the simultaneity bias.
Results are reported in Table 5and, on the whole, confirm those reported in
Tab l e 4. The relationship between the specialisation patterns of production and
innovation activity within regions is positive and increasing over time, with values
ranging from 0.11 to 0.20, which are very similar to those obtained from the
estimation of Eq. 1. The relationship between innovative activities in contiguous
regions is also positive and significant, whereas the introduction of second order
contiguities, as expected, implies a reduction of the coefficient on the first order lag
variable. As in the estimation of Eq. 1, the spatial lag of IST when the third-order
contiguity neighbours are considered is never significant, whereas the strength of the
relationship among second order contiguous regions is remarkably stable along time.
11
The significant value of the Likelihood Ratio test for spatial lag dependence points
to the statistical adequacy of the estimation of this type of model, while spatial
autocorrelation as a spatially correlated error does not appear to be a remaining
problem according to the non-significance of the Lagrange Multiplier test.
11 Other specifications have been estimated to assess for the presence of a relationship between
innovative specialisation of one region and productive specialisation in contiguous regions but
results were not significant. Similarly, some attempts to evaluate the presence of different
coefficients for each macro-sector by means of interactive dummies have not provided interesting
results, probably due to the aggregate nature of our data.
732 R. Moreno et al.
5 Conclusions
In this paper we attempt to provide empirical evidence on the phenomenon of
cluster agglomeration of innovation activities throughout time and space in
European regions. More specifically we try to assess whether there are some forces
which support the development of technologically specialised regional clusters. In
particular we want to determine the spatial extent of these forces, their dynamics
along the eighties and nineties and their connection with production clustering.
We have started from a mapping of innovation activity in European regions by
means of an exploratory spatial analysis based on a global indicator of spatial
dependence. The analysis has been carried out for different time periods and sectors
in order to evaluate differences and similarities. Two main outcomes are worth
remarking. First, the presence of a strong central-periphery distribution of in-
novation activity at the beginning of the period. Innovation activity is concentrated
in regions in North and centre Europe, while none or modest technological activity
is performed in most Southern European regions. Second, this concentration tends
to decrease over time while innovation activity has been spreading to some more
regions in Scandinavia and in the South of Europe. The analysis of the global
indicator of spatial association confirms the presence of a strong and positive
spatial autocorrelation process in the innovative activity. This means that patenting
activity in a certain region tends to be correlated to patenting performed in
contiguous areas. Spatial association is also found at the sectoral level, even at a
higher degree than at a aggregated level, determining the formation of specialised
clustering of innovative regions in different sectors.
The second step concerns the analysis of the characteristics of the geography of
innovation specialisation modes across regions and across time. So, we follow the
Table 5 Economic results
Variables ML estimation
1989–1991 1994–1996 1999–2001
ISP
t−1
0.112 0.104 0.108 0.184 0.167 0.168 0.194 0.174 0.173
0.004 0.008 0.006 0.000 0.000 0.000 0.000 0.000 0.000
W (1) IST
t−1
0.163 0.140 0.140 0.119 0.101 0.102 0.130 0.104 0.103
0.000 0.000 0.000 0.002 0.008 0.007 0.001 0.006 0.007
W (2) IST
t−1
0.030 0.033 0.032 0.032 0.036 0.034
0.004 0.000 0.001 0.000 0.000 0.000
W (3) IST
t−1
−0.013 −0.007 0.003
0.101 0.402 0.699
LIK −396.8 −390.6 −389.2 −354.0 −348.2 −347.9 −341.5 −333.1 −333.0
AIC 843.5 833.1 832.4 758.1 748.5 749.8 733.1 718.2 720.0
LR test 18.512 13.136 13.154 10.125 7.096 7.252 11.614 7.129 7.035
0.000 0.000 0.000 0.001 0.008 0.007 0.001 0.008 0.008
LM spatial error 0.904 0.189 1.534 0.027 0.125 0.001 0.009 0.016 0.084
0.342 0.664 0.216 0.870 0.724 0.981 0.923 0.900 0.772
Dependent variable: technological specialisation index (IST )
Number of observations: 1,225 (175 regions and SEVEN sectors). National and sectoral dummies
included. P-values are given in italics
Geographical and sectoral clusters of innovation in Europe 733
idea that innovation specialisation in one region is highly dependent on spe-
cialisation of production in the same region. However, even after accounting for the
geographic concentration of the production specialisation, an interesting point
analysed in this paper is that we provide evidence on to what extent the specialisation
of innovative activity in one region is influenced by the specialisation pattern in
neighbouring regions. In other words, we analyse the role played by interregional
technological spillovers in sectoral specialisation in the geographical space.
Among the main results, it is shown that specialisation in innovative activity is
positively and significantly influenced by specialisation in production activity, a
pattern which seems to be increasing over time. Moreover, the positive and
significant coefficient of the weighted average of the index of technological
specialisation in the neighbouring regions suggests that innovation tends to cluster
more in sectors in which the neighbouring regions are also technologically spe-
cialised. In other words, technological specialisation patterns follow a geographical
pattern which links contiguous regions. All in all, the results suggest that the
propensity for innovation to cluster in some specific sectors in a region is attributable
not only to the geographic concentration of production in those sectors but also on
the role played by technological spillovers. So, the results in this paper raise some
policy issues. Among others, according to the evidence showed of the existence of
technological spillovers, it seems that coordinated actions among different regions
in favour of spurring technology could be more successful than isolated actions.
Acknowledgements We thank Barbara Dettori for excellent research assistance. We have
benefited from useful comments by participants at the 2004 ERSA conference, COST Action 17
meetings in Prague and Kaunas and seminars in Barcelona and Cagliari. This paper is the result of
a joint research project developed within the COST-Action 17. Financial support by MIUR
(COFIN 2002 project n. 2002138187_02) and DGICYT SEC2002-00165 are gratefully
acknowledged.
Appendix
European Regions in CRENoS database (Id-CRENoS; Id-Nuts; Region; Nuts
level)
1 AT11 Burgenland 2
2 AT12 Niederosterreich 2
3 AT13 Wien 2
4 AT21 Karnten 2
5 AT22 Steiermark 2
6 AT31 Oberosterreich 2
7 AT32 Salzburg 2
8 AT33 Tirol 2
9 AT34 Vorarlberg 2
10 BE1 Bruxelles-Brussel 1
11 BE2 Vlaams Gewest 1
12 BE3 Region Walonne 1
13 CH01 Region Lemanique 2
14 CH02 Espace Mittelland 2
15 CH03 Nordwestschweiz 2
734 R. Moreno et al.
16 CH04 Zurich 2
17 CH05 Ostschweiz 2
18 CH06 Zentralschweiz 2
19 CH07 Ticino 2
20 DE11 Stuttgart 2
21 DE12 Karlsruhe 2
22 DE13 Freiburg 2
23 DE14 Tubingen 2
24 DE21 Oberbayern 2
25 DE22 Niederbayern 2
26 DE23 Oberpfalz 2
27 DE24 Oberfranken 2
28 DE25 Mittelfranken 2
29 DE26 Unterfranken 2
30 DE27 Schwaben 2
31 DE3 Berlin 2
32 DE4 Brandenburg 2
33 DE5 Bremen 2
34 DE6 Hamburg 2
35 DE71 Darmstadt 2
36 DE72 Giessen 2
37 DE73 Kassel 2
38 DE8 Mecklenburg-Vorpomm 2
39 DE91 Braunschweig 2
40 DE92 Hannover 2
41 DE93 Luneburg 2
42 DE94 Weser-Ems 2
43 DEA1 Dusseldorf 2
44 DEA2 Koln 2
45 DEA3 Munster 2
46 DEA4 Detmold 2
47 DEA5 Arnsberg 2
48 DEB1 Koblenz 2
49 DEB2 Trier 2
50 DEB3 Rheinhessen-Pfalz 2
51 DEC Saarland 2
52 DED1 Chemnitz 2
53 DED2 Dresden 2
54 DED3 Leipzig 2
55 DEE1 Dessau 2
56 DEE2 Halle 2
57 DEE3 Magdeburg 2
58 DEF Schleswig-Holstein 2
59 DEG Thuringen 2
60 DK DENMARK 0
61 ES11 Galicia 2
62 ES12 Asturias 2
63 ES13 Cantabria 2
Geographical and sectoral clusters of innovation in Europe 735
64 ES21 Pais Vasco 2
65 ES22 Navarra 2
66 ES23 Rioja 2
67 ES24 Aragon 2
68 ES3 Madrid 2
69 ES41 Castilla-Leon 2
70 ES42 Castilla-la Mancha 2
71 ES43 Extremadura 2
72 ES51 Cataluna 2
73 ES52 Com. Valenciana 2
74 ES61 Andalucia 2
75 ES62 Murcia 2
76 FI13 Ita-Suomi 2
77 FI14 Vali-Suomi 2
78 FI15 Pohjois-Suomi 2
79 FI16 Uusimaa 2
80 FI17 Etela-Suomi 2
81 FI2 Aland 2
82 FR1 Ile de France 2
83 FR21 Champagne-Ard 2
84 FR22 Picardie 2
85 FR23 Haute-Normandie 2
86 FR24 Centre 2
87 FR25 Basse-Normandie 2
88 FR26 Bourgogne 2
89 FR3 Nord-Pas de Calais 2
90 FR41 Lorraine 2
91 FR42 Alsace 2
92 FR43 Franche-Comte 2
93 FR51 Pays de la Loire 2
94 FR52 Bretagne 2
95 FR53 Poitou-Charentes 2
96 FR61 Aquitaine 2
97 FR62 Midi-Pyrenees 2
98 FR63 Limousin 2
99 FR71 Rhone-Alpes 2
100 FR72 Auvergne 2
101 FR81 Languedoc-Rouss 2
102 FR82 Prov-Alpes−Cote d’Azur 2
103 FR83 Corse 2
104 GR11 Anatoliki Makedonia 2
105 GR12 Kentriki Makedonia 2
106 GR13 Dytiki Makedonia 2
107 GR14 Thessalia 2
108 GR21 Ipeiros 2
109 GR22 Ionia Nisia 2
110 GR23 Dytiki Ellada 2
111 GR24 Sterea Ellada 2
736 R. Moreno et al.
112 GR25 Peloponnisos 2
113 GR3 Attiki 2
114 GR41 Voreio Aigaio 2
115 GR42 Notio Aigaio 2
116 GR43 Kriti 2
117 IE01 Border 2
118 IE02 Southern and Eastern 2
119 IT11 Piemonte 2
120 IT12 Valle d’Aosta 2
121 IT13 Liguria 2
122 IT2 Lombardia 2
123 IT31 Trentino-Alto Adige 2
124 IT32 Veneto 2
125 IT33 Fr.-Venezia Giulia 2
126 IT4 Emilia-Romagna 2
127 IT51 Toscana 2
128 IT52 Umbria 2
129 IT53 Marche 2
130 IT6 Lazio 2
131 IT71 Abruzzo 2
132 IT72 Molise 2
133 IT8 Campania 2
134 IT91 Puglia 2
135 IT92 Basilicata 2
136 IT93 Calabria 2
137 ITA Sicilia 2
138 ITB Sardegna 2
139 LU LUXEMBOURG 0
140 NL1 Noord-Nederland 1
141 NL2 Oost-Nederland 1
142 NL3 West-Nederland 1
143 NL4 Zuid-Nederland 1
144 NO01 Oslo og Akershus 2
145 NO02 Hedmark og Oppland 2
146 NO03 Sor-Ostlandet 2
147 NO04 Agder og Rogaland 2
148 NO05 Vestlandet 2
149 NO06 Trondelag 2
150 NO07 Nord-Norge 2
151 PT11 Norte 2
152 PT12 Centro 2
153 PT13 Lisboa e V.do Tejo 2
154 PT14 Alentejo 2
155 PT15 Algarve 2
156 SE01 Stockholm 2
157 SE02 Ostra Mellansverige 2
158 SE04 Sydsverige 2
159 SE06 Norra Mellansverige 2
Geographical and sectoral clusters of innovation in Europe 737
160 SE07 Mellersta Norrland 2
161 SE08 Ovre Norrland 2
162 SE09 Smaland med oarna 2
163 SE0A Vastsverige 2
164 UKC North East 1
165 UKD North West 1
166 UKE Yorkshire and the Humber 1
167 UKF East Midlands 1
168 UKG West Midlands 1
169 UKH Eastern 1
170 UKI London 1
171 UKJ South East 1
172 UKK South West 1
173 UKL Wales 1
174 UKM Scotland 1
175 UKN Northern Ireland 1
References
Anselin L (1988) Spatial econometrics: methods and models. Kluwer, Dordrecht
Audretsch D, Feldman M (1996) R&D spillovers and the geography of innovation and
production. Am Econ Rev 86:631–640
Baldwin RE, Martin P (2004) Agglomeration and regional growth, In: V Henderson, JF Thisse
(eds) Handbook of regional and urban economics: cities and geography. North-Holland
Bottazzi L, Peri G (2003), Innovation and spillovers in regions: evidence from European patent
data. Eur Econ Rev 47:687–710
Breschi S (2000) The geography of innovation: a cross-sector analysis. Reg Stud 34:213–229
Burridge P (1980) On the Cliff–Ord test for spatial autocorrelation. J R Stat Soc B 42:107–108
Caniels M (2000) Knowledge spillovers and economic growth, Edward Elgar, Cheltenham
Cheshire P (1990) Explaining the performance of the European Community’s major urban
regions. Urban Stud 27:311–333
Ciccone A (2002) Agglomeration effects in Europe. Eur Econ Rev 46:213–27
Coe D, Helpman E (1995) International R&D spillovers. Eur Econ Rev 39:859–87
Combes P (2000) Economic structure and local growth: France, 1984–1993. J Urban Econ
47:329–355
Evenson RE, Putnam J, Kortum S (1991) Estimating patent counts by industry using the Yale–
Canada concordance, Final Report to the National Science Foundation
Florax RJ, Folmer H, Rey S (2003) Specification searches in spatial econometrics: the relevance
of Hendry’s methodology. Reg Sci Urban Econ 33:557–79
Griliches Z (1990) Patent statistics as economic indicators: a survey. J Econ Lit 28:1661–1707
Grossman G, Helpman E (1991) Trade, knowledge spillovers, and growth. Eur Econ Rev
35:517–526
Jaffe AB, Trajtenberg M, Henderson R (1993) Geographic localization of knowledge spillovers
as evidenced by patent citations. Q J Econ 63:577–598
Keller W (2002) Geographic localization of international technology diffusion. Am Econ Rev
92:120–142
Mariani M (2002) Next to Production or to technological clusters? The economics and
management of R&D location. J Manag Gov 131–152
Martin P, Ottaviano G (2001) Growth and agglomeration. Int Econ Rev 42:947–968
Moran P (1948) The interpretation of statistical maps. J R Stat Soc 10:243–251
Moreno R, Paci R, Usai S (2005) Spatial spillovers and innovation activity in European regions,
Environ Plann A (forthcoming)
Paci R, Usai S (1999) Externalities, Knowledge Spillovers and the Spatial Distribution of
Innovation. GeoJournal 49:381–390
Paci R, Usai S (2000) Technological enclaves and industrial districts. An analysis of the regional
distribution of innovative activity in Europe. Reg Stud 34:97–114
738 R. Moreno et al.
Storper M (1995) Regional technology coalitions an essential dimension of national technology
policy. Res Policy 24:895–911
Varga A, Anselin L, Acs Z (2005) Regional Innovation in the US over Space and Time, In: G
Maier, S Sedlacek (eds) Spillovers and Innovation, Springer, Wien
Von Hipple E (1994) Sticky information and the locus problem solving: implications for
innovation. Manag Sci 40:429–439
Geographical and sectoral clusters of innovation in Europe 739
