Download as pdf or txt
Download as pdf or txt
You are on page 1of 25

doi:10.1111/j.1435-5957.2007.00130.

Space versus networks in the geography of


innovation: A European analysis*
Mario A. Maggioni1, Mario Nosvelli2, Teodora Erika Uberti1
1
DISEIS (Dept. of International Economics, Institutions and Development), and Faculty of
Political Science, Catholic University of Milan, L.go Gemelli, 1 I-20123 Milan, Italy
(e-mail: mario.maggioni@unicatt.it; erika.uberti@unicatt.it)
2
CERIS-CNR, Via Bassini n. 15, 20133 Milan, Italy (e-mail: m.nosvelli@ceris.cnr.it)

Received: 20 October 2006 / Accepted: 23 April 2007

Abstract. This paper provides an original framework for the interpretation of


innovative activity among European regions according to traditional ‘geographi-
cal’ spillovers and ‘relational’ spillovers. The focus is on two knowledge-based
relational phenomena: participation in the same research networks (within the EU
Fifth Framework Programme) and EPO co-patent applications. Using two econo-
metric techniques, we investigate the factors that determine patenting activity,
distinguishing structural features, geographical and relational spillovers. In this
way, we are able to test whether hierarchical relationships based on a-spatial
networks between geographically distant excellence centres prevail over diffusive
patterns based on spatial contiguity.

JEL classification: O31, R12, C21

Key words: Networks, innovation, spatial econometrics, spillovers, relational


knowledge exchange

1 Introduction

The European Union has always been concerned about ‘regional disparities’ in
standard macroeconomic indicators such as GDP per capita or the unemployment
rate, and this issue has become more relevant after the recent enlargements of the

* Financial support from D1 Research Project “Relazioni e Suiluppo” is gratefully acknowledged.


Thanks are due to Z. Acs, S. Beretta, K Frenken, A. Gambardella, F. Lissoni, S. Usai and two
anonymous referees for useful comments on previous versions of the paper.

© 2007 the author(s). Journal compilation © 2007 RSAI. Published by Blackwell Publishing, 9600 Garsington Road,
Oxford OX4 2DQ, UK and 350 Main Street, Malden MA 02148, USA.

Papers in Regional Science, Volume 86 Number 3 August 2007.


472 M.A. Maggioni et al.

EU membership (Traistaru et al. 2003; Lackenbauer 2004; Hapiot and Slim 2004).
Traditionally, regional economic disparities have been ascribed to peripherality,
measured via gravity models based on the concept of regional economic potential.
Since the economic potential of a region is “a function both of its proximity to
other economic centres and of its economic size or ‘mass’. [. . .] the influence of
each economic centre on any other centre is assumed to be proportional to its
volume of economic activity and inversely proportional to a function of the
distance between them. The economic potential of a given location is found by
summing the influences on it of all other centres in the system” (Schürmann and
Talaat 2000, p. 4).
More recently, the role of innovative capacities and knowledge endowment in
defining both the level and the GDP growth rate of a region has been stressed in
economics, building on the theoretical results of the endogenous growth theory
(Romer 1986; Lucas 1988). Hence, in order to understand the causes of regional
disparities and to design appropriate policies one should take into account the
relationships between research activity and innovation processes with special
emphasis on the structural features of the European scientific and technological
systems at the regional level, looking in particular at flows of different types of
knowledge and information (Maggioni and Uberti 2007). Having acknowledged
the important ‘trails’ contained in patents, as sustained by Jaffe et al. (1993), in this
approach we move a little further, identifying and detecting two aspects of knowl-
edge flows: participation in the same research networks (funded by the EU within
the Fifth Framework Programme) and EPO co-patent applications. Through these
variables, we attempt to measure the intrinsic relational structure of knowledge
flows, which directly connects people, institutions and, indirectly, regions across
European countries.
The central idea of the paper is that knowledge is created within some crucial
nodes (i.e., firms and universities) which tend to co-locate together in specific
sites, thus determining the birth and development of high-tech clusters, innovative
industrial districts, and excellence centres (Bresnahan et al. 2001; Braunerhjelm
and Feldman 2006). Knowledge is then diffused and exchanged either through a
diffusive pattern based on spatial contiguity, or according to intentional relations
based on a-spatial networks. According to the first pattern, the geographical
selection process leading to a hierarchical structure of the location of innovative
activities goes together with an increasing role of ‘unintended’ spatial knowledge
spillovers that, from excellence centres, extend their positive effects to other
agents (firms, universities, research centres) located in neighbourhood areas. So
relevant regions present both an ‘attractivity’ potential and a ‘diffusive capacity’
(Acs et al. 2002). According to the second pattern, knowledge is mainly exchanged
according to voluntary ‘barter’ and increased through learning by interacting
procedures, within specialised networks which are intentionally established
between crucial nodes (Cowan and Jonard 2004).
The aim of this paper is to verify whether or not hierarchical relationships,
based on a-spatial networks between geographically distant excellence centres,
prevail over diffusive patterns, based on spatial contiguity. To achieve this aim we
perform two empirical exercises focussed on a subset of 109 European regions at

Papers in Regional Science, Volume 86 Number 3 August 2007.


Space versus networks in the geography of innovation: A European analysis 473

NUTS2 level in France, Germany, Italy, Spain and the United Kingdom. In the first
exercise, we consider the process of co-patenting between these regions in order to
compare, within a gravity equation model, the influence of geographical distance
and relational distance in shaping these scientific and technological relationships.
In the second exercise, we analyse the patenting activity of the same subset of
European regions in order to measure and compare, through spatial econometric
techniques, the relative effects of geographical and relational proximity in deter-
mining their innovative performances.
The paper is organised as follows: section 2 describes the theoretical back-
ground, section 3 includes a brief description of the Fifth Framework Programme,
section 4 contains the empirical analyses and section 5 concludes the paper by
summarising the main results.

2 The theoretical framework

Technological activities and knowledge spillovers are considered important factors


in determining the performance of economic systems. More specifically, in the last
two decades, greater attention has been devoted to the spatial and geographical
dimension of the mechanisms of creation and diffusion of technology and numer-
ous studies have emphasized the local and cumulative nature of technology spill-
overs. Starting from the seminal paper by Griliches (1979), several studies have
explored the characteristics of R&D spillovers and their role in the process of the
creation of new knowledge. The empirical investigation of the spatial patterns
however, followed by technology flows has suffered for a long time from the lack
of adequate indicators. Krugman (1991a) stated that knowledge flows are invisible
and cannot be properly measured and tracked. This view has been opposed firstly
by Jaffe et al. (1993) who suggested that knowledge flows do leave a ‘paper trail’,
in the form of patent citations, which can be measured and used to obtain infor-
mation on the spatial dimension of the innovation spillovers mechanism.
In this paper, building on the subsequent composite stream of literature dealing
with space and innovation (Acs et al. 1994; Anselin et al. 1997; Fischer and Varga
2003; Acs et al. 2002; Greunz 2003; Bode 2004; Moreno et al. 2005), we move a
little further by focussing on two knowledge-based relational phenomena (co-
patents and scientific collaborations in the EU Fifth Framework Programme)
which are not linked to patent citations. In a gravity model framework, we measure
and use as dependent variable the number of co-patent applications. Unlike cita-
tions, co-patents detect the localisation of inventors working at the same invention
and thus are a good proxy of scientific and technological collaboration across
space. Similarly, in order to map the structure of relations among scientific insti-
tutions, we use data on research networks within the Fifth Framework Programme
as weight matrix in the spatial econometric exercise. Throughout the paper, our
main theoretical reference is the knowledge production function, firstly identified
by Griliches (1979) and later developed, within a spatial context, by Jaffe (1989),
in order to investigate the process of innovation creation and the existence of
related knowledge flows caused by unintended spatial spillovers and or intentional
knowledge barter exchange.

Papers in Regional Science, Volume 86 Number 3 August 2007.


474 M.A. Maggioni et al.

The basic formalisation of the knowledge production function (Griliches 1979)


specifies a function in which knowledge depends on the current and past level of
public and private research expenditure and a set of unmeasured indexes of the
accumulated level of knowledge. We implement it by regressing the number of
patents or co-patents over a series of attributive and relational variables, and being
aware of the advantages, on the one hand – “patents statistics remain a unique
resource for the analysis of technical change” (Griliches 1990, p. 1702) – whereas
on the other hand, “. . . patents are a flawed measure particularly since not all new
innovations are patented and since patents differ greatly in their economic impact”
(Pakes and Griliches 1980, p. 378), thereby showing their limitations. In the
empirical analyses, independent variables are tuned to our specific research objec-
tives focussing on the role played by both the internal structure of each regional
innovation system (public and private R&D and industrial structure) and the
influence of ‘neighbouring’ regions, where neighbourhood may be spatially or
relationally defined. Further variables are added to specific regression models.
The crucial distinction between geographical and relational proximity refers to
the existence of two different channels through which knowledge flows between
different institutions across space. A similar distinction was first proposed by
Hägerstrand (1965 and 1967) who postulated that knowledge diffusion might
follow two different patterns: contagious and hierarchical. In the first pattern,
knowledge diffusion is more effective in the vicinity of the knowledge source and
greatly diminishes across distance. In the second pattern, knowledge diffuses
firstly among central places, and at later stages it trickles down to lower order
places. From a hierarchical perspective, knowledge diffusion between central
places is faster than among lower order ones because central places display a
higher level of knowledge infrastructures and better connections to transportation
and communication networks. Florax and Folmer (1992) expanded this line of
reasoning and empirically tested the impact of academic knowledge production at
a regional level on the investment behaviour of the manufacturing industry in the
Netherlands by modelling two alternative knowledge diffusion patterns: conta-
gious and hierarchical. The hypothesis tested in their paper is the following: “the
division of labour with regard to universities and private companies requires
intensive interaction. This interaction may take place via contagious and/or hier-
archical diffusion. If the former dominates, a clustering of private firms around
universities may be expected. If hierarchical diffusion dominates, a clustering
around central places instead of around universities will appear” (Florax and
Folmer 1992, p. 441). Their results show that for investments in buildings, near-
ness to universities is not an important factor, thus contrasting with the contagious
pattern, whereas investments in equipments are higher for firms located nearer to
central places, supporting the hierarchical diffusion pattern.
According to Polanyi (1958 and 1966), one may also associate these different
knowledge diffusion patterns to the existence of two different kinds of knowledge:
codified and tacit knowledge. Codified knowledge is knowledge that has been
converted into symbols. For this reason it is easily transferable and can be trans-
mitted through ICT technologies and infrastructures over long distances and across
organisational boundaries. Tacit knowledge however, is inarticulate, embedded and

Papers in Regional Science, Volume 86 Number 3 August 2007.


Space versus networks in the geography of innovation: A European analysis 475

not explicitly stated. For this reason, it is not easily transferable and its transfer is
extremely costly and sensitive to social context. Cowan et al. (2000) suggest that
technological change and economic growth have had the effect of tipping the
balance between tacit and codified knowledge in the economic process of contem-
porary advanced economies. Maggioni and Nosvelli (2005) study the role played by
tacit and codified knowledge in the employed labour force in the Italian provinces
and find that they play a complementary role everywhere apart from industrial
districts, in which they behave as substitute goods. Breschi and Lissoni (2001)
raised some criticism to the empirical literature on localised knowledge spillovers
emphasising that this concept is too narrow to embrace the broad range of knowl-
edge transmission mechanisms that may, or may not, facilitate the dissemination of
ideas and expertise while keeping the diffusion process bounded in space.
Since in this paper we are working with regional data at NUTS2 level, it is
difficult to distinguish between contagious and hierarchical knowledge diffusion
similarly to Florax and Folmer (1992). Furthermore since we are dealing with
knowledge flows connected to patenting and co-patenting activities on the one hand
and to institutionalised research networks on the other, it is almost impossible to
disentangle the tacit from the codified elements of knowledge flows running across
European regions. It is reasonable however, to assume that, since tacit knowledge
needs face-to-face contacts and these contacts are inversely related to geographic
distance (Maggioni and Uberti 2007), long-distance relations imply a greater role
played by codified knowledge than the relationships between nearer regions.

3 The 5 Framework Programme (5FP): An overview

Searching for ‘paper trails’ recording evidence of scientific and technological


cooperation and networking between institutions and organisations located in five
European countries, we focussed on the research contracts within the Fifth Frame-
work Programmes (henceforth, 5FP). The main aim of European Framework
Programmes is to promote scientific research and technological development
within the EU. The mission of such programmes is to solve problems and answer-
ing the challenges of the integration process by investigating the socio-economic,
technological, industrial, economic, social and cultural aspects of the EU. The 5FP
is a five-year programme that finances research projects within a time intervals of
5 years (1998–2002) and differentiates from the previous ones according to its
targets, organisational structure and types of projects. 5FP has a multi-theme
structure that covers all soft and hard sciences; contracts are grouped according to
the aim of the research, as indicated by the official definition of ‘Thematic’ and
‘Horizontal’ Programmes, and are divided into various typologies depending on
the type of actors involved (European Commission 1998). Being interested in the
networks of scientific and technological relationships established between institu-
tions and organisations belonging to the 5 largest EU countries (France, Germany,
Italy, Spain and United Kingdom – which represent about 80 percent of the GDP
and population and 77 percent of the total R&D expenditure), we focus our
analysis on a subset of research contracts, whose organisational structure is a
‘network’ composed of one coordinator and several participants. While the average

Papers in Regional Science, Volume 86 Number 3 August 2007.


476 M.A. Maggioni et al.

membership of all 5FP contracts is slightly less than 5, the average dimension of
the selected sample exceeds 8.
Since we focus our analysis on a network-type organisational structure of
contracts, it is important to define the network nodes: coordinators and participants.
Coordinators are, in EU jargon, those legal entities (i.e., natural persons, legal
persons, international organisations and joint research centres) in charge of the
contracts both in legal and in scientific terms, since they are ‘legally’ responsible for
the project in front of the Commission; but their role reflects important managing
and power skills with respect to the other members of the research project. 5FP
research contracts involve several type of actors involved at different levels. In
general, research and education institutions coordinate 4,316 contracts (27 percent),
while firms only 1,010 contracts (6 percent). Institutions grouped under a residual
label ‘Other institutions’ coordinate 8,512 contracts (53 percent).1 A similar analy-
sis, conducted on the categories of participants included in the dataset (62,617 in
total), confirms the relative marginality of the industry sector, representing only 11
percent of the total participants, while universities and research institutions (44
percent) and ‘Other’ actors (41 percent) play a more relevant role.
Another interesting analysis of the 5FP research contracts deals with its
geographical distribution. The map of coordinators of 5FP contracts is naturally
euro-centered (89.4 percent of the total projects are coordinated by institutions
located in Europe 15). In general, coordinators are not evenly distributed across
Europe: larger countries, with well-developed national innovation systems, are
over-represented and the presence of EU offices on the national territory seems to
have a positive impact, both as coordinator and as participants. The five countries
analysed in this work account for the 62.5 percent of the 5FP coordinators. British
institutions coordinate about 18 percent of total financed contracts, followed by
Germany (14.1 percent) and France (13 percent). Italy (9.8 percent) and Spain (7.2
percent) are lagging behind. The same countries represent 55.6 percent of total
participants, but German institutions are more present (13.9 percent) followed by
British (12.3 percent), and French (12.1 percent), whereas Italy and Spain show
percentages similar to the coordinator’s case (9.1 percent and 6.1 percent). Similar
percentages characterise the distribution of total funds: the United Kingdom
receives nearly 18 percent of total funds, Germany nearly 17 percent; France 14.5
percent; Italy and Spain receive less funds, respectively 9 percent and 6 percent. In
per capita terms, the figures are completely different: smallest countries (i.e.,
Denmark, Belgium and Netherlands) bear on average nearly three times more than
the 5 big countries.
In this section, we have shown that 5 major countries account for about 2/3 of
the entire budget of the 5FP. These countries display a lot of internal variances in
the structure and performance of their regional innovation systems and in partici-
pation rates to joint research networks. Therefore, an empirical analysis conducted
on the basis of our selected sample may give interesting insights into the relative
1
This label defines private consultancy firms specialised in dealing with complex EU funding
application procedures. Hence these coordinators may not have the relevant skills in research and
technological development activities, but rather in the applications procedures and in managing
research funding.

Papers in Regional Science, Volume 86 Number 3 August 2007.


Space versus networks in the geography of innovation: A European analysis 477

effects of geographical and relational proximity in determining the innovative


activity of a region.

4 The empirical analyses

Scientific and technological knowledge is both created and diffused through some
crucial nodes (i.e., universities, research institutions, firms, etc.), which tend to
spatially concentrate within excellence centres and high-tech clusters (Swann et al.
1998; Bresnahan et al. 2001; Maggioni 2002; Braunerhjelm and Feldman 2006).
This co-location process however, may have two distinct effects on the ‘geography
of innovation’ at the regional level. On the one hand, each cluster may extend its
influence over the neighbouring territories through a trickling down process of
spatial diffusion (underlining the role of face-to-face contacts, the local mobility of
the labour force, and other forms of localised knowledge spillovers). According to
this perspective, space matters most and knowledge flows following unintended
patterns. On the other hand, technological and scientific knowledge, developed
within the cluster, may be diffused and exchanged through a set of a-spatial
networks (often structured in formal and contractual agreements between institu-
tions) connecting each cluster with other clusters, irrespectively of the geographi-
cal contiguity. According to this perspective, relational networks matter most and
knowledge spreads following intentional patterns, not directly traced by geo-
graphical contiguity.
The aim of the empirical analyses is to verify the relative importance of the two
above-mentioned phenomena in order to test whether formal relationships based
on a-spatial networks between geographically distant clusters prevail over diffu-
sive patterns based on spatial contiguity. In order to do this we focus on two
knowledge-based phenomena: EPO co-patents applications and joint participa-
tions in 5FP research networks. Through these variables, we attempt to measure
the intrinsic relational structure of knowledge flows, which directly connects
people, institutions and, indirectly, regions across the European countries. In
particular, the first one measures the networking activity as output and the second
one detects the impact of the networking activity as an input of the innovative
process. Throughout the analysis, ‘relational’ refers to a subset of all 5FP con-
tracts, selected according to two criteria. First of all, we chose those contracts*
which involved (directly or indirectly) the creation of a ‘network’, and whose
coordinator was an organisation/institution localised in one of the 109 European
regions belonging to the above mentioned 5 large countries, irrespective of the
type of coordinator. Secondly, we selected only those applications whose record
was complete (i.e., containing information on region’s coordinator; on typology of
coordinator and on the total amount of EU funds). This selection process produced
a final number of 4,566 contracts, about 28 percent of all contracts financed within
the 5FP and 64 percent of the total funding. Finally, we transformed all selected
contracts into a single 109 ¥ 109 network (see section 4.2 for details).
* Cooperative research contracts, coordination of research actions, cost-sharing contracts, demon-
stration contracts, preparatory, accompanying and support measures, research infrastructure trans-
national access, research networks contracts, thematic network contracts.

Papers in Regional Science, Volume 86 Number 3 August 2007.


478 M.A. Maggioni et al.

4.1 A gravity model of co-patenting

Economists have seen patents as one of the best output indicators of innovative
activities (Griliches 1990; Acs et al. 2002). Since the seminal contributions of
Scherer (1965) and Griliches (1981), patents have also been used in the economic
literature in order to measure knowledge spillovers. The constitution of the Euro-
pean Patent Office (EPO) in Munich in 1977 allowed researchers to use a common
dataset to analyse the innovative performances of different European countries and
regions. In particular Paci and Usai (2000), Caniëls (2000) and Breschi and Lissoni
(2004) developed systematic analyses of patenting activity throughout Europe at
different NUTS levels, showing the existence of significant clustering phenomena
within a core-periphery geographical pattern and whose agglomeration indexes are
even higher than those registered by high-tech manufacturing. Later studies inves-
tigate patent data as relational variables. Breschi and Lissoni (2004 and 2006) use
patent citations in order to compare the relevance of spatial proximity as opposed to
social proximity in determining the spillover effect of scientific research. Maggioni
et al. (2007) look at patents as relations between inventors and applicants in the
Italian provinces searching for industry-specific patterns and testing the hypotheses
of a ‘brain-drain’ dynamic between peripheral and metropolitan areas.
Following Maggioni and Uberti (2007), we study an important relational
aspect of patents: co-patents, in relation to the co-invention process. Co-invention
(and thus co-patenting) is a process involving exchanges of both tacit and codified
knowledge. Thus, it implies a series of both ‘face-to-face’ and ‘over the distance’
relationships between inventors. That is why it is interesting to analyse the relative
importance of ‘geographic’ versus ‘functional’ distance as forces shaping the
interregional (international) structure of knowledge flows networks (as done in
Maggioni and Uberti 2007). For this purpose, out of a total of more than 170,900
patent applications belonging to every IPC section (coming from inventors located
in the above mentioned 5 countries in the period 1998–2002) – extracted by the
CRENOS files based on the original EPO database – we selected only those
patents whose applications were recorded by more than one inventor. Next we split
each patent into equal shares attributed to each inventor; and we added these data
for each NUTS2 regions in order to build a matrix in which a generic cell ij
represents the share of patents recorded jointly by inventors located in regions i
and j (where region i and region j might belong to different nations). Finally, a total
of nearly 30,000 co-patents was identified.
The gravity equation model is an extremely successful tool for empirical
analyses which explain interactions according to the existence of ‘attractive’ and
‘impeding’ forces. In economic literature, gravitational models are commonly
used to explain the actual pattern of international trade flows. Bilateral trade
between two countries is proportional to their economic mass (i.e., GDP and/or
population) and inversely related to their geographical distance. In this paper we
build a gravitational model which explains knowledge flows embedded in the
realisation of a patentable innovation by (at least) two inventors living in two
generic regions i and j, CO_PATij, co-patenting during the period 1998 and 2002,
as a function of a series of attributional and relational variables. Firstly, we include

Papers in Regional Science, Volume 86 Number 3 August 2007.


Space versus networks in the geography of innovation: A European analysis 479

the most elementary inputs in technical knowledge creation: business R&D expen-
diture as a percentage of GDP (BizRD) and government R&D expenditure as a
percentage of GDP (GovRD),2 as defined by the Frascati Manual (OECD 2002).
Secondly, because it is a gravity equation framework, we include a variable to
measure the role played by geographical distance (between the regions and from
the regions to the European ‘centre’).
We use geographical distances (GEODISTij) between 109 European regions,
calculated according to the shortest road distance (in kilometres) between regional
‘capitals’. In this paper, the notion of ‘regional capital’ is arbitrary since NUTS2
levels are administratively meaningful entities in Italy, Germany, Spain and France,
but not in the UK. In this last case we used the size of the population as the selective
criteria to identify the most relevant city (which we called ‘capital’), irrespective of
the presence of an administrative capital. Another important geographical aspect is
related to the centrality, or peripherality, of a region in respect to a geographical
European centre identified with Brussels: regions distant from Brussels are much
more disadvantaged and show poorer economic performance. Hence, we include
the PERIPHij variable to detect this aspect. This variable simultaneously considers
the distance of each couplet of regions from Brussels and we also consider the
minimum value between the two to detect the advantage of the most central
region.The similarity of the innovative specialisation of two regions (TECHSIMILij)
is measured as the correlation coefficient between the sectoral composition of patent
application registered by region i and those registered by region j at EPO during the
period 1997–2000 (Moreno et al. 2005). The relational aspect is grasped through the
co-membership in the same 5FP project. The variable MEMBij counts the number of
joint memberships of regions within the considered subset of 5FP contracts. This
variable captures the influence of research networking on the co-patenting activity.
Finally, we add some dummy variables: CONTIGij, the geographical contiguity
which takes value 1 for contiguous regions (i.e., which share a border), 0 elsewhere;
and dI, a dummy variable which is used to control for fixed national effects both on
the ‘emitting’ and the ‘receiving’ regions.
By adopting an OLS procedure, we estimate a double-log specification so that
coefficients can be interpreted as elasticities. It is defined as follows:

Co _ PATij = α 0 + α1 BizRDi + α 2 BizRD j + α 3GovRDi + α 4GovRD j +


α 5TECHSIMILij + α 6GEODISTij + α 7CONTIGij + α 8 PERIPHi + (1)
α 9 MEMBij + δ I + ε ij

where dI indicates country dummies variables and eij is the standard error term.
The results of the regression in Table 1 show that business R&D is more
relevant than government R&D in explaining the variance of the co-patenting
activity and that the co-patenting activity of two regions is positively correlated to
the degree of technological similarity of their innovation systems. In particular a 1
percent increase of private R&D influences co-patenting activity to about half of a

2
Due to data availability, either BizRD or GovRD are calculated as averages for years 1995, 1997,
1999 and 2001.

Papers in Regional Science, Volume 86 Number 3 August 2007.


480 M.A. Maggioni et al.

Table 1. The gravity equation of co-patents. Dependent variable:


Co-patentingij – OLS estimation

Variables Model

BizRDi 0.393***
(0.029)
BizRDj 0.393***
(0.029)
GovRDi 0.042***
(0.020)
GovRDj 0.042***
(0.020)
TECHSIMILij 0.321***
(0.154)
GEODISTij -1.001***
(0.040)
CONTIGij 0.701***
(0.091)
PERIPHij -0.591***
(0.047)
MEMBij 0.315***
(0.023)
Number of observations 4,518
F-Test 383.43
Adj-R2 0.5153

Notes: Double log specification. Robust standard errors in parenthe-


sis; *** significant at 1%.
Country dummies and a constant term were included in all regres-
sions. Their coefficients are always significant but not reported in the
table.

percentage point, but the same increment in government R&D affects co-patenting
activity 10 times less (about 0.04 percent).
The technological similarity of the innovation systems of two regions posi-
tively affects the co-patenting activity between them, confirming the relevance of
the technological specialisation of a region in the identification of its innovative
partners. Space does matter in the innovation activity: geographical distance
between regions is negatively related to co-patenting (doubling the distance
between two regions, reduces the co-patenting activity by one-half); and contigu-
ous regions show a higher propensity to co-invent. Furthermore, geographical
peripherality does negatively influence the co-patenting activity: two equally
distant couplets of regions are likely to have different co-patenting figures accord-
ing to their distance from Brussels.3 Relational activities (proxied by joint research
networks) play a smaller, but significantly positive, role in the co-patenting activ-
ity. In fact, a 1 percent increase in co-membership affects the co-patenting activity

3
We replicated the estimation with an alternative measure of peripherality (based on remoteness:
i.e., the distance of a region from every other region in the sample) and obtained very similar results a
part from stronger coefficients of the peripherality variable.

Papers in Regional Science, Volume 86 Number 3 August 2007.


Space versus networks in the geography of innovation: A European analysis 481

by 0.3 percent. This confirms that relational innovative activities (e.g., being part
of the same European research network) positively affect the innovative capacity
of a region and determine the increase of common innovative results.
We can therefore conclude that this first empirical exercise suggests the promi-
nence of spatial spillover dynamics over formal relational networking in deter-
mining the knowledge exchange and interactions needed to jointly produce a
patentable innovation. One may suspect that most research contracts are estab-
lished between contiguous (or proximate) regions, thus weakening the above-
mentioned results by blurring the difference between spatial spillovers and the
relational exchange of knowledge. This suspicion however, proves to be wrong as
shown by the correlation coefficient between GEODISTij and MEMBij which is
almost null (-0.0744) and the average distance between two randomly selected
members of a research network, which is equal to about 800 km.

4.2 Spatial dependence vs. relational dependence in patenting activity

The previous econometric exercise focussed on co-patenting which account for a


minimal part of all patenting activity in the 5 largest European countries (30,000
out of 170,900 patents). The sample is certainly small but not biased, since the
‘co-patenting’ and the ‘patenting’ activity are similarly distributed across Euro-
pean regions.4 In order to have a more complete description of the relevance of
relational versus geographical proximity, we estimate a second model whose
object is the patenting activity of European regions focussing on the geographical
and relational dependence in the data. In doing so we follow a stream of literature,
which examines the knowledge production function from a spatial perspective.
Jaffe (1989) analyses the existence of spatial knowledge spillovers originated by
universities and finds that these positive externalities are intrinsically local due to
relevance of tacit knowledge in academic relations, which makes scientific and
technological knowledge transmissible only through face-to-face contacts.5
The relevance of localised knowledge spillovers has been confirmed by a series
of empirical analyses for the US (Acs et al. 1994; Audretsch and Feldman 1996;
Anselin et al. 1997). These papers however, do not explicitly model the mecha-
nism of knowledge transfer and therefore, as highlighted by Breschi and Lissoni
(2001), run the risk of amalgamating local knowledge spillovers with other sources
of pecuniary externalities. In order to distinguish between these two effects we use
a knowledge production function where the dependent variable is the number
of patent applications per million labour force registered by inventors located in
109 European regions in 2002. The set of independent variables is composed
of various types of research and development expenditures (R&D), different

4
The Spearman rank correlation (0.85) and Pearson correlation (0.74) between the variables
distributions are both very high.
5
More recent papers such as Bottazzi and Peri (2003), Greunz (2003), Varga et al. (2003), Bode
(2004), Abreu et al. (2005) and Moreno et al. (2005) investigate the mechanisms and determinants of
the process of creation and diffusion of innovative knowledge, explicitly taking into account both
temporal and spatial dynamics.

Papers in Regional Science, Volume 86 Number 3 August 2007.


482 M.A. Maggioni et al.

measures of the specialisation in patenting and production, and two different


measures for spatial and relational dependence. Due to the fact that there exists a
time lag between inputs and outputs in the production of new knowledge between
R&D expenditure and patent applications, we consider the number of patent
applications registered at EPO in 2002, while for all structural independent vari-
ables we calculated the average value of the period 1995–2001. The use of
different sources of R&D funding (private and public) enables us to detect which
sector (private or public) is most involved in the process of creating potentially
‘marketable’ knowledge leading to an active patenting activity. In the regression
we include business R&D expenditure (BizRDi) and government R&D expenditure
(GovRDi) expressed as a percentage of the regional GDP.6
Following Moreno et al. (2005), we also want to test whether or not the
high-tech specialisation (of both the innovation and production systems at the
regional level) influences the patenting activity in general. We therefore first
calculate absolute (denoted by SP) and relative (denoted by LQ) specialisation
indexes. SP_INNi corresponds to the share of high-tech patents (HT) with respect
to the total number of patents (TOT) of region i as follows:

PatiHT
SP _ INN i = .
PatiTOT

Secondly, in order to test the influence of the relative specialisation of the


regional innovation system, as compared to the national one, we calculate a
traditional location quotient for high-tech patents (LQ_INNi) as follows:

PatiHT Pat IHT


LQ _ INN i =
PatiTOT Pat ITOT

where i identifies the region, I the nation, HT the patents in high-technology


industries and TOT the total number of patents in all sectors.
In order to measure the relevance of the specialisation of the production
structure of each region included, we calculate an absolute and a relative speciali-
sation index, SP_PRODi and LQ_PRODi, by measuring economic activity in terms
of the number of local units. The first index (SP_PRODi) is the absolute speciali-
sation in high-tech activity of regions, and is calculated as the share of the number
of local units in high-tech sectors (HT)7 in region i compared to the total number
of local units (TOT). Formally:

6
In the regression we exclude university R&D for two main reasons. Firstly, the empirical literature
showed that this variable is insignificant when regressed against patents if not used together with an
“index of geographic coincidence of university and industrial research activity” (Jaffe 1989, p. 958),
and these data for NUTS2 are not available. Secondly, and more importantly, data showed a high
correlation between university and government R&D at the regional level, implying the risk of
multicollinearity should both variables be used as regressors.
7
In this analysis, the high-tech sector includes the following sectors of NACE classification Rev.
1.1: DL30 Manufacture of office machinery and computers; DL32 Manufacture of radio, television and
communication equipment and apparatus; DL33 Manufacture of medical, precision and optical instru-
ments, watches and clocks.

Papers in Regional Science, Volume 86 Number 3 August 2007.


Space versus networks in the geography of innovation: A European analysis 483

LUiHT
SP _ PRODi = .
LUiTOT

The relative specialisation of the regional production system in high-tech


industries is calculated as a location quotient relative to local units in high-tech
sectors (LQ_PRODi) is calculated as follows:

LUiHT LU IHT
LQ _ PRODi =
LUiMAN LU IMAN

where all variables are as above except LUMAN, which identifies the total number of
local units in manufacturing in the region (i) (or in the nation if I).
Furthermore, we include a variable, MEMBi that corresponds to the total
number of 5FP networks in which region i was involved as a member. Finally, in
order to detect not only the role of geographical distance, but also the role of
peripherality, we included a geographical variable, PERIPHi, that indicates the
distance of region i from Brussels.
To measure the relative importance of geographical versus relational depen-
dence in the patenting activity we use spatial econometrics techniques that
control for spatial dependence and heterogeneity problems (Anselin 1988). As
we want to identify the role of geographical and/or relational contiguity, for the
spatial econometric analysis we adopt two different spatial weight matrices, a
geographical (indicated as geo) and a relational one (indicated as rel), both
defining the neighbouring structure of each observation. The geographical
weight matrix (geo), indicated with W g, is a binary matrix defined by the geo-
graphical contiguity of 109 regions according to the rook criterion. Hence each
element wgij is equal to 1 if regions i and j share a common border and 0
otherwise. The relational weight matrix (rel), indicated as W r, is defined accord-
ing to the scientific relations within the 5FP. The identification of this weight
matrix is obtained in a 5-step procedure: firstly we geographically localised
(according to NUTS2 classification) each single actor involved in the selected
contracts; secondly we summed up, for each region, all contracts dealing with
actors located there, distinguishing between coordinators and participants;
thirdly we identified an asymmetric square 109 ¥ 109 matrix (Z) containing the
number of times each couplet of regions was jointly involved in 5FP contracts
and we symmetrised it according to the minimum value: if zij>zji, the value on
the symmetric matrix, Zmin, is zji. Using this procedure, we were able to capture
the least joint research activity. Fourthly, since the Zmin matrix was a non-binary
matrix, we dichotomised it according to a threshold value, equal to 7, equivalent
to average number of partners in all selected contracts. Hence, the final matrix,
W r, is a binary matrix where each element wrij is equal to 1 if region i and j
jointly participate in 5FP contracts more than 7 times.
Before concentrating on the results of the regressions, we checked for the
presence of spatial autocorrelation in innovative activity using the two weight
matrices. The most diffused tests for the existence of spatial autocorrelation are the

Papers in Regional Science, Volume 86 Number 3 August 2007.


484 M.A. Maggioni et al.

global index Moran’s I and the local index LISA. The Moran’s I, a measure of
global spatial autocorrelation or overall clustering, allows us to estimate the
strength of spatial autocorrelation in data distribution. The LISA index, which
indicates the presence, or absence, of significant spatial clusters or outliers for each
location allowing the identification of ‘hot-spots’ (as well as ‘cold-spots’) areas
(Anselin 1995).
Moran’s I statistics, calculated considering patent activity of each region and
patent activity of ‘spatially’ and ‘relationally’ lagging regions, show positive and
significant (at 1 percent) coefficients, both for W g and W r. Looking at Figure 1a,
we can clearly see that while high-patent and low-patent tend to form isolated
homogenous clusters (suggesting the existence of a polarised structure of the
innovation system within our sample), the European scientific networks financed
within the 5FP are composed of a mixture of high and low patenting regions
(Figure 1b). This preliminary analysis seems to suggest that geographical spill-
overs within contiguous regions seem to be more effective when explaining the
patenting activity of European regions, than explicit scientific collaboration
expressed through research networks.
The local autocorrelation (LISA) maps in Figures 2 and 3, highlight the dif-
ferences in the results obtained using the two different weight matrices. In
Figure 2, when a geographical definition of neighbourhood is used, a ‘stronger’
cluster characterises some German regions, whereas the two ‘weak’ clusters
involve Italian and Spanish regions. In Figure 3, when the 5FP relational neigh-
bourhood is selected as a spatial weight matrix, the ‘stronger’ relational clusters
regard a subset of advanced European regions which, even if they are geographi-
cally dispersed, are connected by a large number of joint research projects. There-
fore, from this rough and preliminary analysis it is evident that 5FP has determined
the emergence of a very dense relational and ‘intentional’ network between
advanced regions from both a technological and an economic perspective which

Fig. 1. Moran scatterplots of patenting activity:


a) Geographical matrix (Wg): Moran’s I’: 0.6672; b) Relational matrix (Wr): Moran’s I’: 0.1822

Papers in Regional Science, Volume 86 Number 3 August 2007.


Space versus networks in the geography of innovation: A European analysis 485

Fig. 2. LISA ‘geographic’ cluster map of patenting activity (Wg)

Fig. 3. LISA ‘relational’ cluster map of patenting activity (Wr)

neglects peripheral ones. Furthermore, one may note that scientific collaboration
apparently disregards geographical contiguity.
Standard spatial econometric theory offers a well-consolidated methodology
for the detection of spatial dependence in error terms, which, in particular, captures
neighbouring effects on the dependent variable. With spatial error models we
tested two different equations that contain the traditional input factors of a knowl-
edge production function (and some measurements of innovative and production
specialisation). The equation, in a double-log specification, is defined as follows:

Papers in Regional Science, Volume 86 Number 3 August 2007.


486 M.A. Maggioni et al.

PATi = β0 + β1 BizRDi + β2 GovRDi + β3 INNi + β4 PRODi + β5 PERIPHi +


(2)
β6 MEMBi + λWi ε + ξi

where explicative variables are those defined above, and INNi is either the absolute
(SP_INNi) or relative index (LQ_INNi) to verify the specialisation of the innova-
tion system; PRODi is either the absolute (SP_PRODi) or relative index
(LQ_PRODi) to verify the specialisation of the production system; spatial effects
are accounted for through the autoregressive coefficient (l) and the spatial lag for
the errors (We). xi is the standard error term. Each equation is first estimated by
using the geographical weight matrix and then the relational one.
Table 2 shows the results of the spatial econometric analysis of the patenting
activity of the 109 European regions included in our sample. The coefficient of
BizRDi is always positive and significant as Table 1, while the coefficient of
GovRDi is generally negative and often insignificant. One may conclude that
publicly funded R&D primarily addresses basic research which rarely produces
patentable (or patented) results, and that the main target of public R&D is the
fostering of cooperation and collaboration between different institutions.

Table 2. Patenting activity: geographical versus relational dependence. Dependent variable: Patenti.
Weight matrices: geo = geographical contiguity matrix Wg; rel = relational contiguity matrix Wr

Variables Specification I Specification II

OLS SEM (ML) OLS SEM (ML)

(I geo) (I rel) (II geo) (II rel)

CONST 10.82*** 10.94*** 10.37*** 10.26*** 9.52*** 10.20***


BizRDi 0.51*** 0.38*** 0.51*** 0.63*** 0.52*** 0.62***
GovRDi -0.04 -0.09 -0.06 -0.15 -0.22** -0.17*
SP_INNi 0.41*** 0.33*** 0.41***
SP_PRODi -0.28* -0.13 -0.30*
LQ_INNi 0.18 0.16 0.18*
LQ_PRODi 0.54* 0.24 0.53**
PERIPHi -0.58*** -0.64*** -0.54*** -0.73*** -0.75*** -0.73***
MEMBi -0.19 -0.13 -0.17 -0.04 0.07 -0.02
Lambda 0.56*** 0.36* 0.60*** 0.25
Observations 109 109 109 109 109 109
R2 0.68 0.53
Log-Likelihood -125.45 -116.57 -123.32 -146.75 -136.94 -144.93
Akaike info 264.9H1 247.15 260.64 307.51 287.87 303.87
criterion

(geo) (rel) (geo) (rel)

Moran’s I 0.27*** 0.11*** 0.28*** 0.14***


Robust LM Lag 2.44 0.44 0.64 0.06
Robust LM Error 6.61** 6.24** 7.52*** 10.29***
Likelihood ratio test 17.76*** 4.27** 19.63*** 3.64*

Note: Double log specification. * significant at 10%, ** significant at 5%, *** significant at 1%.

Papers in Regional Science, Volume 86 Number 3 August 2007.


Space versus networks in the geography of innovation: A European analysis 487

The two sets of specialisation indexes of the innovation and production


systems at the regional level show that the absolute (SP_INNi), rather than the
relative specialisation (LQ_INNi), of the regional innovation systems in high-tech
sectors produces a higher number of patents per head. This result does not hold for
the specialisation of the production systems: having a larger share of the national
high-tech industries (SP_PRODi) does not grant a region a higher per capita
patenting activity (coefficients are negative or insignificant), whereas a relative
specialisation (LQ_PRODi) in the same sectors gives a comparative advantage in
terms of patent propensity and activities. The negative and always significant
coefficient of PERIPHi confirms the existence of a polarised ‘core-periphery’
structure of the European innovation system. The farther the region from Brussels,
the lower the number of registered patents per million labour force. This confirms
the results showed in Figures 1, 2 and 3.
The coefficient of the MEMBi variable is not significant. This shows that it is
not really important for a region to participate in a lot of networks in order to be
innovative; what is important however, is to be part of the most relevant networks,
i.e., to be connected to other relevant regions.
To detect the possible spatial dependence and heterogeneity according to
different weight matrices, W g and W r we computed the Moran’s I on the residu-
als of the OLS regressions, which detects the general presence of spatial depen-
dence in the errors in our model. In order to choose the best model for spatial
dependence correction we follow Anselin and Florax (1995) considering
Lagrange Multiplier (LM) tests for spatial error and spatial lag dependence and
their robust counterparts and we decided to use the spatial error models speci-
fication in our regressions. From Table 2 it quite clearly emerges that the
Moran’s I statistic is significant for all four model specifications, thus signalling
the presence of spatial and relational dependence and the OLS estimation pro-
cedures are inconsistent and biased. The Koenker-Basset and White tests
exclude the presence of heteroskedasticity in the specifications with the spatial
matrix but not with the relational matrix. These results may depend on the very
particular nature of the relational weight matrix since one must bear in mind that
heteroskedasticity is “frequently encountered in spatial analysis when there are
irregular units and relationship being modelled are different in different places”
(Law and Haining 2004, p. 37).
Models (I geo) and (II geo) take into account the geographical interaction
existing between the value of the dependent variable in a given region and the
value of that same variable registered in the neighbouring regions. The lambda
coefficient (which derives from the spatial error specification of the model) is
always positive and highly significant, thus confirming the empirical results of
the established literature: patenting activity in a given region benefits from the
positive performance of its geographically defined neighbours. Models (I rel)
and (II rel) are derived including a relational ‘geography’ in order to test whether
or not there is significant interaction existing between the value of the dependent
variable in a given region and the value of that same variable registered in the
relationally connected regions. The lambda coefficient is positive, but smaller
than in previous cases and not significant in the (II rel) specification. Therefore,

Papers in Regional Science, Volume 86 Number 3 August 2007.


488 M.A. Maggioni et al.

relationally defined neighbourhood seems to have a lower influence on the pat-


enting activity of the European regions.
As we are fully aware however, that one cannot compare the size of the
coefficients of two regressions based on two different weight matrices and that, if
one assumes that a true data generation process is influenced both by geographical
and relational contiguity, any estimation based on a model specifying only one out
of the two possible definitions of contiguity would result in an irremediably biased
estimation, due to omitted variables specification. Hence, we decided to perform
one additional spatial econometric exercise using another different ‘spatial’ weight
matrix. The attempt to disentangle geography from relations implied the definition
of a new ‘spatial’ weight matrix, W r-g, obtained as a difference between W r and W g.
In other words we subtracted geographical contiguity to relations, so that the
surviving neighbourhoods are ‘purely relational’, connections established between
geographically non-contiguous regions which should better proxy ‘intentional’
knowledge barter exchange dynamics. Hence, the wijr-g is equal to 1 if region i and
j are relationally contiguous but not geographically contiguous, and 0 otherwise.
In the next Table 3, we show the results relative to a regression conducted using
W r-g as spatial weight, limiting the investigation to the first specification model
(which includes only the absolute specialisation indexes). This further analysis
confirms the previously obtained results, as far as the independent variables and

Table 3. Patenting activity: purely relational dependence.


Dependent variable: Patenti

Variables OLS Specification I Spatial Error model

Rel-Geo (W r-g)

CONST 10.82*** 10.59***


BizRDi 0.51*** 0.51***
GovRDi -0.04 -0.05
SP_INNi 0.41*** 0.41***
SP_PRODi -0.28* -0.27*
PERIPHi -0.58*** -0.54***
MEMBi -0.19 -0.19
Lambda 0.36*
Observations 109 109
R2 0.68
Log-Likelihood -125.45 -123.75
Akaike info criterion 264.91 261.49

Rel-Geo

Moran’s I 0.09***
Robust LM Lag 0.21
Robust LM Error 4.26**
Likelihood ratio test 3.42*

Note: Double log specification. * significant at 10%, ** significant


at 5%, *** significant at 1%.

Papers in Regional Science, Volume 86 Number 3 August 2007.


Space versus networks in the geography of innovation: A European analysis 489

the model specification (SEM model) are concerned. The lamba term is still
positive and significant. When geography and relations are disentangled, there is
still a ‘spatial’ autocorrelation between ‘geographically distant’ by ‘relationally
near’ regions captured by the lambda term and spatial dependence arises.8
One should also consider that, since one of the main objectives of EU policies
is to reduce geographical imbalances and strengthen the degree of regional cohe-
sion, it might well be the case of a strong selection bias within the 5FP aimed at
supporting research networks involving rich and poor regions, advanced and
laggard ones, central and peripheral. If this is true, then relational contiguity may
be underestimated by our analysis. What is certain is that two counteracting forces
are at play. Autonomous decisions of research institutions and firms at the regional
level will push actors located in central and advanced regions to establish exclusive
networks with other actors located in other central regions, but in order to maxi-
mise the probability of being funded, they look for partners located in peripheral
and laggard regions. The net effect is, in theory, uncertain and has to be empirically
measured.

5 Conclusion

Between the end of the ’80’s and the beginning of the ’90’s, in the field of
economic theory endogenous growth literature highlighted the role played by
human capital and R&D in determining growth performance of economic systems
(Romer 1986; Lucas 1988). Similarly, new economic geography showed that
market forces might produce heavily polarised spatial distribution of economic
activities (Krugman 1991b). In March 2000 the European Council held a special
meeting in Lisbon to “agree to a new strategic goal for the Union in order to
strengthen employment, economic reform and social cohesion as part of a
knowledge-based economy” (European Commission 2000). To achieve this goal,
the European Council put forward a threefold strategy whose first and main pillar
aimed at preparing the EU for a knowledge-based economy and to adopt the
necessary structural reforms to improve competitiveness and innovation. Nowa-
days, although the celebrated Lisbon strategy’s claim has been seriously criticised
and partially reconsidered, it is not possible to deny its merit: having focussed our
attention on the relation between scientific excellence, technological primacy,
trade and production performance, on the one hand, and the trade-off between
internal cohesion and external competitiveness, on the other hand.
Recent empirical literature demonstrates that scientific and technological
knowledge, which leads to, and is partly embedded in patents, is both created and

8
Another way to take into account both relations and geography in the same analytical framework
is to divide the original (non-binary) relational matrix (Z) by the distance matrix (D), which is also
non-binary. According to this procedure, we obtain a new spatial (non-binary) weight matrix that can
be used, after an appropriate row standardisation, to run a new spatial econometric analysis. The results
obtained by this specification confirm the existence of positive spatial dependence (in a SAR specifi-
cation). However, in this paper, in order to maintain consistency among all spatial econometric
exercises we decided to show in Table 3 only results obtained through binary spatial weight matrices.

Papers in Regional Science, Volume 86 Number 3 August 2007.


490 M.A. Maggioni et al.

diffused through crucial nodes like universities, research institutions, firms, etc.,
which tend to concentrate spatially within excellence centres and high-tech clus-
ters (Swann et al. 1998; Bresnahan et al. 2001; Maggioni 2002; Braunerhjelm and
Feldman 2006). This co-location process however, may shape the ‘geography of
innovation’ at the regional level, according to two distinct patterns: an ‘unin-
tended’ geographical effect and an ‘intentional’ relational effect. On the one hand,
each cluster might extend its influence over the neighbouring territories through a
trickling down process of spatial diffusion (underlining the role of face-to face
contacts, labour force, local mobility and other forms of localised knowledge
spillovers). According to this perspective, space matters most. On the other hand,
technological and scientific knowledge developed in the cluster may be diffused
and exchanged through a set of a-spatial networks (often structured in formal and
contractual agreements between institutions) connecting each cluster with other
clusters, irrespective of the geographical contiguity. According to this perspective,
relational networks matter most.
This paper investigates both phenomena and tests whether or not formal
relationships based on a-spatial networks between geographically distant clusters
prevail over diffusive patterns based on spatial contiguity. Two empirical exercises
have been performed. In the first exercise, we build a gravitational model which
explains knowledge flows embedded in the realisation of a patentable innovation
by two inventors living in two generic regions (i.e., co-patents) as a function of a
series of attributional and relational variables. According to this test, the significant
variables in explaining the structure of co-patents are business and public R&D
expenditure and the similarity of innovative structure of the regions. Geography
plays a relevant role in this relational activity: spatial proximity and geographical
centrality (with respect to Brussels) are always significant in determining the
co-patenting activity. Joint collaborations seem to be another important factor in
the shaping of co-patents activity. In the second exercise, using spatial economet-
rics techniques, we tested a knowledge production function, which explains the
number of patent applications as a function of a set of variables (i.e., different
sources of R&D expenditures, various measures of specialisation of the regional
innovation and production system, and spatial and relational regressors) to
measure the degree of spatial and relational autocorrelation of the dependent
variable. The estimations confirm the main results of the literature on innovation
activity: the crucial role played by the R&D expenditure. However, in this exercise
the private expenditure is much more relevant in patenting activity. This is prob-
ably due to the market oriented activity of the industrial sector as opposed to the
basic research activity performed by the public sector and by the relative speciali-
sation in high-technology sectors by private R&D expenditures.
In all the empirical exercises, we introduced different variables, derived from
the 5FP database, to measure the influence of scientific networking on regional
innovation outcomes. In all cases, the participation in 5FP research networks have
some impact on regional innovative activity, but the spatial autocorrelation
depends on variables not included in the model. When membership is measured by
the number of research network participations and it is included as a regressor, its
coefficient is not significantly different from zero. These EU funded research

Papers in Regional Science, Volume 86 Number 3 August 2007.


Space versus networks in the geography of innovation: A European analysis 491

networks may have not fully supported European competitiveness and innovative
performance. This could be partially explained, on the one hand, by the fact that
most collaborative research has been undertaken in the ‘pre-competitive’ phase of
the research process9 and, on the other, by the fact that the private sector is only
marginally involved in the 5FP.10 But the most relevant function of Framework
Programmes lies in the creation of dynamic networks, bringing together research-
ers from laboratories scattered throughout European firms, universities and other
research institutions, providing access to complementary skills, and reducing the
degree of excessive competition among researchers and the duplication of research
efforts.
We should also consider the incentives structure that affects the choice of
partners in a 5FP application. In fact, two important and counteracting forces are
involved: on the one hand, actors want to collaborate with institutions located in
central and advanced regions to establish exclusive research networks and to create
excellence centres; but on the other hand, in order to maximise the probability of
being funded, actors also look for partners located in peripheral and laggard
regions. In conclusion, the paper has shown that relational networks, here proxied
with 5FP membership, influence the behaviour of regional innovation systems, but
that spatial proximity plays a more relevant role in determining their performance.

References

Abreu M, De Groot H, Florax J (2005) Space and growth: A survey of empirical evidence and
Mmethods. Région et Développement 21: 13–44
Acs ZJ, Anselin L, Varga A (2002) Patents and innovation counts as measures of regional production
of new knowledge. Research Policy 31: 1069–85
Acs ZJ, Audretsch D, Feldman M (1994) R&D spillovers and recipient firm size. Review of Economics
and Statistics 76: 336–340
Anselin L (1988) Spatial econometrics: Methods and models. Kluwer Academic Publisher, Norwell
Anselin L (1995) Local indicators of spatial association – LISA. Geographical Analysis 27: 93–115
Anselin L, Florax R (1995) Small sample properties of tests for spatial dependence in regression
models. In: Anselin L, Florax R (eds) New directions in spatial econometrics. Springer, Berlin
Anselin L, Varga V, Acs Z (1997) Local geographic spillovers between university research and high
technology innovations. Journal of Urban Economics 42: 422–448
Audretsch D, Feldman M (1996) R&D spillovers and the geography of innovation and production.
American Economic Review 86: 641–652
Bode E (2004) The spatial pattern of localised R&D spillovers: An empirical investigation for
Germany. Journal of Economic Geography 1: 43–64
Bottazzi L, Peri G (2003) Innovation and spillovers in regions: Evidence from European patent data.
European Economic Review 47: 687–710
Braunerhjelm P, Feldman M (2006) Cluster genesis: The origins and emergence of technology-based
economic development. Oxford University Press, Oxford

9
Too often successful projects did not produce marketable results, either because they have been
isolated from market and social considerations despite their technical excellence, or because the means
by which they were to be exploited were not specified or even thought about at the earliest stages of
work (PREST 2000).
10
As confirmed by the percentage of industrial coordinators involved (equal to 6 percent), a
negligible percentage when compared to the incidence of the educational and research sectors (27
percent).

Papers in Regional Science, Volume 86 Number 3 August 2007.


492 M.A. Maggioni et al.

Breschi S, Lissoni F (2001) Localised knowledge spillovers vs. innovative milieux: Knowledge ‘tac-
itness’ reconsidered. Papers in Regional Science 80: 255–273
Breschi S, Lissoni F (2004) Knowledge networks from patent data: Methodological issues and research
targets. In: Moed H, Glänzel W, Schmoch U (eds) Handbook of quantitative science and tech-
nology research: The use of publication and patent statistics in studies of S&T systems. Springer,
Berlin Heidelberg New York
Breschi S, Lissoni F (2006) Cross-firm inventors and social networks: Localised knowledge spillovers
revisited. Annales d’Economie et de Statistique (forthcoming)
Bresnahan TF, Gambardella A, Saxenian AL (2001) ‘Old Economy’ inputs for ‘New Economy’
outcomes: Cluster formation in the new Silicon Valleys. Industrial and Corporate Change 10:
835–860
Caniëls M (2000) Knowledge spillovers and economic growth. Edward Elgar Publishing, Cheltenham
Cowan R, David PA, Foray D (2000) The explicit economics of knowledge codification and tacitness.
Industrial and Corporate Change 9: 211–253
Cowan R, Jonard N (2004) Network structure and the diffusion of knowledge. Journal of Economic
Dynamics and Control 28: 1557–1575
European Commission CORDIS (1998) Fifth framework programme. http://cordis.europa.eu/fp5/
European Commission (2000) Lisbon European Council. March 2000
Fischer MM, Varga A (2003) Spatial knowledge spillovers and university research: Evidence from
Austria. Annals of Regional Science 37: 303–322
Florax R, Folmer H (1992) Knowledge impacts of universities on industry: An aggregate simultaneous
investment model. Journal of Regional Science 32: 437–466
Greunz L (2003) Geographically and technologically mediated knowledge spillovers between Euro-
pean regions. Annals of Regional Science 37: 657–680
Griliches Z (1979) Issues in assessing the contribution of research and development to productivity
growth. The Bell Journal of Economics 10: 92–116
Griliches Z (1981) Market value, R&D and patents. Economic Letters 7: 183–187
Griliches Z (1990) Patent statistics as economic indicators: A survey. Journal of Economic Literature
28: 1661–1707
Hägerstrand T (1965) Aspects of the spatial structure of social communication and the diffusion of
information. Papers of the Regional Science Association 16: 27–42
Hägerstrand T (1967) Innovation diffusion as a spatial process. University of Chicago, Chicago
Hapiot A, Slim A (2004) Regional gaps and regional policies. The example of the Czech Republic and
Slovakia. Paper presented at the conference “Institutions and Policies for a New Europe”,
Portoroz-Koper, Slovenia, June 17–19
Jaffe AB (1989) Real effects of academic research. American Economic Review 79: 957–970
Jaffe AB, Henderson R, Trajtenberg M (1993) Geographic localization of knowledge spillovers as
evidenced by patent citations. Quarterly Journal of Economics 108: 557–598
Krugman P (1991a) Geography and trade. The MIT Press, Cambridge Massachusetts
Krugman P (1991b) Increasing returns and economic geography. Journal of Political Economy 99:
483–499
Lackenbauer J (2004) Catching-up, regional disparities and EU cohesion policy: The case of Hungary.
Managing Global Transitions 2: 123–162
Law J, Haining R (2004) Spatial regression. University of Cambridge, Department of Geography,
www.geog.cam.ac.uk/people/law/spatialregression.pdf
Lucas RE Jr (1988) On the mechanics of economic development. Journal of Monetary Economics 22:
3–42
Maggioni MA (2002) Clustering dynamics and the location of high-tech firms. Springer Verlag,
Heidelberg
Maggioni MA, Nosvelli M (2005) Conoscenze tacite e codificate nell’apprendimento dei sistemi locali.
In: Bruzzo A, Occelli S (eds) Le relazioni tra conoscenza ed innovazione nello sviluppo dei
territori. Franco Angeli, Milano
Maggioni MA, Uberti TE (2007) International networks of knowledge flows. In: Frenken K
(ed) Applied evolutionary economics and economic geography. Edward Elgar Publishing,
Cheltenham

Papers in Regional Science, Volume 86 Number 3 August 2007.


Space versus networks in the geography of innovation: A European analysis 493

Maggioni MA, Uberti TE, Usai S (2007) Treating patent as relational data: Knowledge transfers and
spillovers in the Italian provinces. Mimeo, March
Moreno R, Paci R, Usai S (2005) Spatial spillovers and innovation activity in European regions.
Environment and Planning A 37: 1793–1812
OECD (2002) Frascati manual 2002. OECD, Paris
Paci R, Usai S (2000) Technological enclaves and industrial districts. An analysis of the regional
distribution of innovative activity in Europe. Regional Studies 34: 97–104
Pakes A, Griliches, Z (1980) Patents and R&D at the firm level: A first report. Economics Letters 5:
377–381
Polanyi M (1958) Personal knowledge: Towards a post-critical philosophy. University of Chicago
Press, Chicago
Polanyi M (1966) The tacit dimension. Doubleday, New York
PREST (2000) European Union science and technology policy and research joint venture collabora-
tion. University of Manchester, Manchester
Romer P, (1986) Increasing returns and long-run growth. Journal of Political Economy 94: 1002–1037
Scherer FM (1965) Firm size, market structure, opportunity, and the output of patented inventions.
American Economic Review 55: 1097–1125
Schürmann C, Talaat H (2000) Towards a European peripherality index: Final report for the General
Directorate XVI Regional Policy of the European Commission. Institut für Raumplanung, Uni-
versität Dortmund, Dortmund
Swann GMP, Prevezer M, Stout D (1998) The dynamics of industrial clustering. Oxford University
Press, Oxford
Traistaru I, Nijkamp P, Longhi S (2003) Specialization of regions and concentration of industries in EU
accession countries. In: Traistaru I, Nijkamp P, Resmini L (eds) The emerging economic geogra-
phy in EU accession Countries. Ashgate, Aldershot
Varga A, Anselin L, Acs Z (2003) Regional innovation in the US over space and time. Discussion
Papers on Entrepreneurship, Growth and Public Policy No. 1804. Max Planck Institute of Eco-
nomics, Jena

Papers in Regional Science, Volume 86 Number 3 August 2007.


doi:10.1111/j.1435-5957.2007.00130.x

Abstract. Este artículo proporciona un marco original para la interpretación de


actividades innovadoras entre regiones europeas de acuerdo con spillovers (efectos
derrame) tradicionales, “geográficos” y “relacionales”. Se pone atención sobre dos
fenómenos relacionales de tipo conocimiento: participación en las mismas redes
de investigación (dentro del Quinto Programa Marco de la UE) y solicitudes de
copatentes en la EPO. Usando dos técnicas econométricas, investigamos los fac-
tores que determinan el proceso de patentes, diferenciando características estruc-
turales, y spillovers geográficos y relacionales. De este modo, somos capaces de
analizar si ciertas relaciones jerárquicas basadas en redes no espaciales entre
centros de excelencia geográficamente distantes predominan sobre patrones de
difusión basados en contigüidad espacial.

Palabras clave: Redes, innovación, econometría espacial, spillovers, intercambio


de conocimiento relacional.

© 2007 the author(s). Journal compilation © 2007 RSAI. Published by Blackwell Publishing, 9600 Garsington Road,
Oxford OX4 2DQ, UK and 350 Main Street, Malden MA 02148, USA.

Papers in Regional Science, Volume 86 Number 3 August 2007.


Dichiarazione sostitutiva di certificazione
(art. 46 D.P.R.28 dicembre 2000 n. 445 )

Il/la Sottoscritto/a Teodora Uberti c.f. BRTTDR72C62G264F

nato/a a Palazzolo sull’Oglio (BS) il 22 /marzo /1972 ,

residente a Milano (MI) in via Ercole Ferrario n° 12

consapevole che chiunque rilascia dichiarazioni mendaci è punito ai sensi del codice penale e delle

leggi speciali in materia, ai sensi e per gli effetti dell'art. 46 D.P.R. n. 445/2000

DICHIARA

che nell’articolo

MAGGIONI M.A, NOSVELLI M, UBERTI T.E. (2007). Space versus networks in the geography of
innovation: A European analysis. PAPERS IN REGIONAL SCIENCE, vol. 86, p. 471-493, ISSN:
1056-8190

sebbene frutto di un lavoro congiunto, le single sezioni siano così attribuibili

section 1 Mario A. Maggioni; section 2 Teodora Erika Uberti; section 3 Mario Nosvelli; section 4.1
Mario Nosvelli; section 4.2 Teodora Erika Uberti; section 5 Mario A. Maggioni.

Luogo, Milano, 30 ottobre 2013

Firma del dichiarante


(per esteso e leggibile)
Ai sensi dell’art. 10 della legge 675/1996 e successive modificazioni, le informazioni indicate nella
presente dichiarazione verranno utilizzate unicamente per le finalità per le quali sono state
acquisite.

You might also like