KIBS and the Dynamics of Industrial Clusters: a

Complex Adaptive Systems Approach

Benoı̂t Desmarchelier, Faridah Djellal, Faı̈z Gallouj

To cite this version:

Benoı̂t Desmarchelier, Faridah Djellal, Faı̈z Gallouj. KIBS and the Dynamics of Industrial
Clusters: a Complex Adaptive Systems Approach. 2015. <halshs-01188568>

HAL Id: halshs-01188568

https://halshs.archives-ouvertes.fr/halshs-01188568
Submitted on 1 Sep 2015

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 1

KIBS and the Dynamics of Industrial Clusters: a

Complex Adaptive Systems Approach
B. Desmarchelier*, F. Djellal**, F. Gallouj**
* Xi’An Jiaotong-Liverpool University, China

** University Lille 1, Clersé-CNRS

Abstract

An important and highly debated question in economic geography is how to

explain the dynamics of industrial clusters, i.e. their emergence and evolution
through time. Two main theories are generally explored, without being
confronted: the cluster life cycle theory - which mainly adopts an aggregate
point of view - and the network-based approach. Although KIBS are an
important actor of industrial clusters, these two theories pay little attention to
them as a potential driver of clusters’ dynamics.

We show in this paper that properly taking KIBS into account requires
considering an alternative and integrative approach that conciliates these two
theories. In particular, we argue that complex adaptive systems (CAS)
constitute a promising basis for such a synthesis. We then operationalize the
CAS approach by studying an existing industrial cluster - Skywin (aeronautics in
Wallonia region, Belgium) - within this framework. For this purpose, we use an
exhaustive list of the innovation projects undertaken within this cluster between
2006 and 2014 and we build temporal innovation networks linking the agents of
the cluster. It appears that Skywin’s innovation networks exhibit a small-world
effect. This implies that any agent who takes part into an innovation project of
this cluster can easily benefit from knowledge and information generated within
another ongoing project. We argue that this effect is an interesting proxy of a
cluster’s attractiveness and an appropriate aggregate variable for studying
clusters’ dynamics as it shows cluster’s potential for further growth. We also
demonstrate that KIBS are the main responsible for the emergence of this
small-world effect in Skywin’s innovation networks.

Introduction
This paper aims at intertwining – in a theoretical and operational way – three strands
of literature: (i) innovation through knowledge intensive business services (KIBS
thereafter), (ii) industrial clusters‟ dynamics, and (iii) complex adaptive systems.

KIBS are services which are processing, generating, and diffusing knowledge within
the economy, and as such they are largely regarded as important (co-)producers of
 2

innovations (Miles et al., 1995; Gadrey and Gallouj, 1998; Den Hertog, 2000; 2002;
Gallouj, 2002), as well as a promising engine for economic growth (Desmarchelier et
al., 2013a) and a key component of regional and national innovation systems (Muller
and Zenker, 2001) and of technological and sectoral systems of innovation alike.
Typical activities are training services, R&D, engineering services and consultancy in
its various forms (technical or not). KIBS include both traditional professional services
(such as legal services, audit and accountancy, market research, personnel services,
management consultancy, etc.) and new technology based services. According to
Miles et al. (1995), regarding “their relation to new technology”, compared to the latter,
the former are “users rather than agents in development and diffusion” (p. 27) of new
technologies. Universities are often not included into the broad category of KIBS (ex.
Muller and Zenker, 2001; Miles et al., 1995). They have indeed many functions (e.g.
teaching and fundamental/academic research), which are not directly oriented
towards businesses‟ technological (and non-technological) needs. However, some of
their functions clearly fit into KIBS purposes, especially but not exclusively new
technology based KIBS (ex. technical training, technical consultancy, business
funded R-D, establishment of research centers in partnership with businesses) and
industrial clusters‟ studies very often highlight their central role in explaining clusters
emergence (Saxenian, 1994; Audretsch and Feldman, 1996a; 1996b). The present
study itself also underlines universities‟ role in favoring the emergence of new
technologies within an industrial cluster. In this paper we therefore include
universities and research bodies within the KIBS category and the empirical part is
mainly focused on such types of KIBS.

More generally, KIBS‟ central role within successful industrial clusters has been
emphasized since the birth of this latter concept. Indeed, in Porter‟s words “clusters
are geographic concentrations of interconnected companies, specialized suppliers,
service providers, firms in related industries, and associated institutions (e.g.
universities, standard agencies, trade associations) in a particular field that compete
but also cooperate” (Porter, 2000, p.15). In this definition, KIBS enter mainly into the
“associated institutions” category, as it includes “universities, think-tanks, vocational
training providers” (p.17).

An important and highly debated question in economic geography is how to explain

the dynamics of these clusters, i.e. their emergence and evolution through time
(Frenken et al., 2015; Boschma and Fornhal, 2011). Two main theories are generally
explored, without being confronted1: the cluster life cycle theory (Menzel and Fornahl,
2010; Shin and Hassink, 2011; Audretsch and Feldman, 1996) - which adopts mainly
an aggregate 2 point of view - and a network-based approach (Saxenian, 1994).
Surprisingly, these two theories pay little attention to KIBS as a potential driver of
clusters‟ dynamics.

1
A notable exception is to be found in Martin and Sunley (2011), whose contribution will be discussed
in this paper.
2
This means that the age of the cluster is proxied by just one (or a limited number of) variable(s),
which can be for example the number of employees or the number of firms, etc.
 3

We show in this paper that taking KIBS into account requires considering an
alternative and integrative approach that conciliates these two theories. In particular,
we argue that complex adaptive systems – or CAS - (Martin and Sunley, 2011;
Holland, 2012) constitute a promising candidate for such a synthesis. Since these
systems are mainly encountered into the theoretical literature (Dilaver et al., 2014;
Albino et al., 2003; Squazzoni and Boero, 2002; Boero et al., 2004), we choose to
justify our theoretical stance by studying KIBS‟ leading role within an existing
industrial cluster, conceived as a CAS: Skywin (aeronautics in Wallonia region,
Belgium).

The remaining of the paper is organized in two parts: we begin by discussing about
the competing theories of clusters dynamics and we advocate for the CAS approach,
then we conduct our empirical analysis in order to illustrate the usefulness of this
theoretical stance.

1. Life Cycle Theory vs. Network-Based Approach: the

need for a synthesis
a. From product and industry life cycles to cluster life cycle
The life cycle hypothesis in economics and management literature is, in its original
form, a descriptive model aiming at synthesizing in a coherent manner a wide variety
of stylized facts about the evolution through time of the marketed products.

Pioneers of the Product Life Cycle (PLC) theory, Utterback and Abernathy (1975)
portray product evolution in three successive steps: (i) the “uncoordinated process”,
within which firms undertake mainly product innovations aiming at improving their
technical performance, (ii) the “segmental process”, where firms modify minor
characteristics for increasing product variety and earning market shares, then (iii) the
„systemic process” where innovation efforts focus on reducing production costs.
These three phases are also and more often labeled: “fluid”, “transition” and “specific”
phases (Abernathy and Utterback, 1978). Even though authors claim that “there is
reason to believe that in any cases the progression may stop for long periods, or
even reverse” (Utterback, and Abernathy, p. 645), they insist on the high degree of
predictability/determinism in the way products evolve through time. Klepper (1996,
1997) systematizes this PLC into a model of Industry Life Cycle within which firms‟
entries, exit, growth and innovations in “technologically progressive industries” (p.
564) are the driving forces behind the PLC.

The general character of this theory has been challenged by the advent of service
economies. In particular, Barras (1986) points out the existence of a “reverse product
cycle” (RPC) within service sectors. In this view, service firms acquire innovations
(mainly information technologies) coming from manufacturing sectors firstly for
improving the efficiency of service operations. Then comes the stage of service
 4

quality improvement, and eventually the production of totally new services. The PLC
is supposed to be reversed as far as process innovations precede product innovation
in the cycle. Likewise, this life cycle theory of the innovation dynamics in services
proved to be incomplete. Gallouj (1998) argues that it reflects a “technological bias”
(p. 128): indeed in Barras model (1986), services cannot innovate by themselves as
their innovations mainly come from the use of the so-called “enabling technologies”,
i.e. information technologies. Non-technological forms of innovation which are
important in services and which concern not only the organization and the process
but also the product are not taken into account by the RPC theory. In contrast,
Gallouj (1998) finds that many KIBS perform several types of non-technological
innovation including “ad hoc” innovations, i.e. custom made innovations adapted to
their clients‟ needs.

Gallouj and Weinstein (1997) proposed a more complete view of firms‟ innovations
(whether they originate from industry or services). Adopting a characteristics-based
approach of the product – good or service – conceived as a set of technical and
service characteristics 3 , they identify six different modes of innovation: radical,
improvement, incremental, ad hoc, recombinative and formalization (see Gallouj and
Weinstein, 1997 for details). These six modes are not exclusive to each other nor a
priori ordered in a pre-determined sequence. Another important point as regards the
life cycle theory is that, according to the authors, the PLC encompasses only “one
point of entry” for innovations: the technical characteristics of the product. It follows
that the Life Cycle conception offers a limited and deterministic view of innovation
dynamics. Nevertheless, it is a popular metaphor for reporting industrial clusters‟
evolutions.

A first exploration of the Life Cycle theory applied to industrial clusters is undertaken
by Audretsch and Feldman (1996a; 1996b). According to them, the main driver of
firms‟ agglomeration is the low transferability of tacit knowledge through long
distances. Following Klepper‟s Industry Life cycle, they postulate that tacit/ localized
knowledge is important in early developments of a given industry, fostering a certain
degree of firms‟ agglomeration. However, as the industry becomes mature, a
dominant design emerges and the product becomes standardized. Firms thus mainly
rely on codified knowledge and information, which are easy to share in long distances.
The initial clustering is thus replaced by a movement of firms‟ dispersion when the
industry reaches maturity.

Even though KIBS are not explicitly mentioned by Audretsch and Feldman (1996a;
1996b), universities are seen by these authors as an important source of tacit
knowledge and are thus a key focal point for early clusters‟ developments. Moreover,
knowledge codification process appears to be the driving force of the life cycle.
Arguably, services and KIBS in particular are major actors in knowledge processing
(Gallouj, 2002) and transmission (Miles et al., 1995; Lau and Lo, 2015) and should

3
In the Lancasterian tradition (Lancaster, 1966, Saviotti and Metcalfe, 1984)
 5

thus be regarded as key actors in explaining clusters‟ evolution. This service-friendly

theory is questioned by Klepper (2010), who rather emphasizes the role of a “spinoff
process”, that is of the emergence of “firms with one or more founders that previously
worked at another firm” (p. 16).

Menzel and Fornahl (2010) for their part also propose a knowledge-driven clusters‟
life cycle theory, summarized in Figure 1. Two dimensions of the cluster are
considered: the number of employees and the heterogeneity of “accessible
knowledge”. The main driver of the cluster life cycle, addressed in terms of number of
employees, is a gradual process of knowledge homogenization among the members
of the cluster. Although similar to Audretsch and Feldman‟s approach (1996a; 1996b)
by the role it attributes to the nature of knowledge in the dynamics of the cluster, this
theory has the advantage to avoid too deterministic evolutions from emergence to
death, since clusters can always enter into loops of self-sustainment, successive
cycles of growth and decline, or even re-orient themselves through a process of
“transformation”. This adaptability is determined by the degree of knowledge
heterogeneity and by the openness to new comers of incumbent firms‟ networks.

Figure 1: Knowledge-based cluster life cycle (from Menzel and Fornahl, 2010 p. 218)

Interestingly, cluster life cycle theory has moved away from the original product and
industry life cycles, as it becomes less deterministic and more influenced by local
drivers, notably clusters‟ ability to maintain a healthy degree of knowledge
heterogeneity (especially through new intrant firms or new “imported” technologies).
At the opposite of Audretsch and Feldman (1996a; 1996b), for whom clusters‟ life
cycles are “shaped” by the industries that they belong to, Menzel and Fornahl (2010)
consider clusters as more independent entities. However, we argue that this
approach remains too restrictive and deterministic since, at least for incumbent
cluster agents, it considers only one kind of innovation trajectory, i.e. only one kind of
knowledge processing mode, namely: formalization (Gallouj, 2002). Actually,
“renewal”, “adaptation” and “transformation”, i.e. the reverse innovation trajectory or
 6

knowledge processing mode (namely differentiation/localization) (Figure 1) are only

possible through “external knowledge” (Menzel and Fornahl, 2010 p. 229), thus
through exogenous/unexplained factors.

More generally (and beyond Menzel and Fornahl‟s contribution) another point of
criticism towards the life cycle theory is that it gives too few importance to cluster‟s
actors in explaining aggregate dynamics. Indeed, this approach generally considers
the cluster – i.e. an aggregation of heterogeneous actors – as a relevant decision
maker. Following Martin and Sunley (2011), one can wonder whether “products,
technologies, industries and clusters [can] be treated as if they are the economic
equivalent of biological organisms” (p. 1301). Besides, even though universities are
sometimes cited in early clusters‟ dynamics (Audretsch and Feldman, 1996a; 1996b),
the main actors mentioned are very often the “firms”, but we neither know which
primary sector of activity they belong to (or whether they all belong to the same
sector) nor the nature of the interactions they entertain between each other. Klepper
(2010) “spinoff process” is clearer on this point, since spinoffs generally belong to the
same sector as the original company – or as the research team in the case of
university spin-offs - and are, at first, of smaller size. The exclusive focus on firms is
not satisfactory for addressing the clusters dynamics. Indeed, according to Porter
(2000), firms are also supported by a number of “associated institutions” within
clusters, mainly “universities, think-tanks, vocational training providers” (p. 17).
Regarding the account for the diversity of the actors involved, the network-based
approaches are obviously more appropriate.

b. Networks and clusters dynamics

An alternative explanation of clusters‟ dynamics is focusing on their internal
organization in the form of networks of interacting entities. According to Newman
(2003, p. 2), “ a network is a set of items [called] vertices or sometimes nodes, with
connections between them, called edges. Systems taking the form of networks
abound in the world”. Within clusters, nodes are companies and supporting
institutions and the edges are all kind of relations between these actors: common
investments in R&D, involvements in the same production processes, common
patents or shared resources, etc. However, the network is not only a structure, it is
also a mode of coordination that fits between market and hierarchy. From an
innovation perspective, the network is considered as a coordination mode that is
more effective than both market and hierarchy. Indeed resorting to the market
assumes the establishment of explicit contracts, while in the field of research and
innovation, projects are highly complex and uncertain. This makes it difficult to
establish explicit contracts, which furthermore raise the risk that strategic secrets
might be divulged (Hakansson, 1989; Callon, 1991; Hakansson and Johansson,
1993). The hierarchy for its part reduces transaction costs but involves the risk of
bureaucratisation, which (as already foreseen by Schumpeter) may be prejudicial to
innovation. In this network tradition, one can mention here Saxenian‟s (1994) seminal
work, comparing the “network” or system-based Silicon Valley and the “independent
 7

firms-based” Route 128. Saxenian argues that this is the prevalence of horizontal
networks between firms and research institutions (e.g. Stanford) in Silicon Valley that
allowed this cluster to successfully switch from semiconductors to microcomputers
during the 1980s, whereas independent firms in Route 128 failed to adapt to the new
technological conditions of that time. An horizontal network is in Saxenian‟s words a
set of actors who “deepen their own capabilities by specializing” (p. 4), thus whose
links are different from just input-output flows.

An interesting observation here is that, in a network perspective, clusters exist and

develop because of a specialization process, which is the opposite of the knowledge
homogenization generally invoked by the Life cycle literature. However, KIBS can be
drivers of both dynamics: formalization of existing tacit knowledge or generation of
custom-made (specialized) knowledge (Gallouj, 2002). It follows that if we recognize
that KIBS are active members of industrial clusters, we have to acknowledge that
both dynamics are possible. This remark about KIBS advocates for an integrative
approach recognizing the influence of actors‟ interactions on the direction taken over
time by the cluster as a whole. We argue in the following that Complex Adaptive
Systems (CAS) allows integrating, within a single framework, both the aggregate
perspective of the life cycle theory and the micro (or multi-agents based) perspective
of the network-based approach, without falling into deterministic predictions.

c. Complex Adaptive Systems: towards and integration of

network and life cycle perspectives
Martin and Sunley (2011) recently proposed to consider industrial clusters as a
particular type of CAS. According to them, a CAS is a system “made up of numerous
components with functions and inter-relationships that imbue the system as a whole
with a particular identity and a degree of connectivity or connectedness” (p. 1303).
Furthermore, a CAS is “characterized by non-linear dynamics because of
various feedbacks and self-reinforcing interactions amongst component (…). It
is also characterized by emergence and self-organization”. However, reading from
these authors, it is not clear what improvements these CAS bring to Saxenian‟s
network-based framework (Saxenian, 1994) nor to the Life Cycle theory discussed
above. Indeed, Martin and Sunley (2011) use a typology of “meta-models” covering
the various forms of CAS dynamics (inspired by Cumming and Collier, 2005). These
meta-models range from deterministic (traditional) life cycles to totally random walks.
Life cycle trajectories are envisaged as special case of CAS among others4. Among
the proposed models, Martin and Sunley argue that clusters dynamics are well
depicted by the so-called “adaptive life-cycle model”5 and they try to adapt it to the

4
The typology of meta-models of CAS includes the following meta-models (types of complex
systems): life cycle, random walk, replacement, limitation, succession, adaptive cycle, evolutionary.
5
The “adaptive cycle model of the evolution of a complex system” (Martin and Sunley, 2011 p.1307) is
similar to the “modified adaptive cycle” represented in Figure 2, minus the alternative trajectories of
“failure”, “constant cluster mutation”, “cluster disappearance”, “cluster stabilization” and “cluster re-
orientation”.
 8

cluster dynamics. The resulting “modified cluster adaptive cycle” that they propose is
reproduced in Figure 2. Arguably, this “meta-model” is very similar to the knowledge-
based life cycle proposed by Menzel and Fornahl (2010), as we can easily draw a
parallel between their respective alternative trajectories: “constant cluster mutation” in
Martin and Sunley (2011) stands for “adaptation” in Menzel and Fornahl (2010),
similarly “cluster stabilization” stands for “renewal”, and “cluster re-orientation” stands
for “transformation”. However, the two cycles are not equivalent: in Menzel and
Fornahl (2010), knowledge heterogeneity between firms and other actors explains
the emergence of a cluster, and the process of knowledge homogenization drives
cluster‟s evolution. In Martin and Sunley (2011), there is no general mechanism of
evolution, since there is no general principle explaining why a cluster shifts from one
phase to another. Instead, these authors propose a descriptive list of potential drivers.
For instance, cluster re-emergence is possible thanks to “sufficient resources,
inherited capabilities and competencies” (p. 1313) left after a phase of decline, or a
constant mutation comes from “high rates of spin-offs” (p. 1313). Apart from a chance
factor, there is no explanation of why the rate of spin-offs is high or why the
remaining capabilities are enough and up-to-date. Another weakness of their model
is that, despite their definition of a CAS, they do not precisely ground clusters‟
dynamics in a network-based view of the actors, and the actors are not considered as
heterogeneous entities.

Figure 2: Martin and Sunley (2011) "modified cluster adaptive cycle" (p. 1312)

Although we point out limitations of Martin and Sunley‟s (2011) adaptive cycle, we
find very relevant their proposition to rely on CAS for conceptualizing clusters‟
 9

functioning and dynamics. Rather than trying to classify such systems, we consider
that a general definition and a list of properties can justify this point of view.

According to Holland (2012) a CAS “consists of a multitude of interacting components

called agents […]. The agents are diverse rather than standardized, and both their
behavior and their structure change as they interact” (p. 57). Furthermore CAS
display the following three main features:
1. “There is no universal competitor or global optimum in a CAS” (p. 58).
2. “Innovation is a regular feature of CAS” (p. 58).
3. “In a CAS, anticipations change the course of the system” (p. 60).

Clusters‟ network structure has already been documented by many authors, including
Porter (1998; 2000) and Saxenian (1994). All of them focus on the diversities of the
“agents” involved: firms, universities, think-tanks, etc. It might thus be argued that, as
structures, clusters are examples of CAS. But do they share CAS properties?

Applied to clusters, the first characteristic mentioned above implies (i) that networks
of agents can be found in many technological or market niches, and (ii) that
cooperation between specialized agents can always allow for improvements. The
remark about niches is particularly relevant for clusters, since the clustering
phenomenon reflects a tendency towards regional specializations in very distinctive
activities including vine production, sportswear, semiconductors, etc.

Regarding the second characteristic, evidence shows that, within clusters, agents are
specialized and that they cooperate in order to be more innovative (Porter, 1998;
2000; Saxenian, 1994). Innovations can take various forms, without following any
pre-determined sequence (Gallouj and Weinstein, 1997).

Finally with regards to the third characteristics, it can be underlined that within a
cluster, every agent can anticipate/forecast new technological or market opportunities,
although their anticipation is imperfect because of bounded rationality (Frenken, 2006;
Desmarchelier et al., 2013b). This characteristic is important, because it contradicts
the very conception of a from birth to death pre-determined cluster cycle. In addition,
unlike Menzel and Fornahl (2010), who introduce exogenous factors as the main
drivers likely to change the course of a system, in a CAS approach, clusters adapt
because of their agents‟ individual anticipations.

In conclusion, and as we will try to confirm it in the empirical part of this work, these
characteristics seem to fit well with what is known about clusters functioning. Then,
how does the conception of clusters as CAS change the way we understand their
dynamics?

Our literature review identified the very reason of clusters‟ existence: the knowledge-
seeking behavior of the firms. They seek knowledge from other firms or from other
types of agents – notably KIBS, including universities. But we also identified an
important difference between Life Cycle theory and network-based theory, regarding
 10

the way they address the knowledge dynamics within clusters: the life cycle theory
postulates a knowledge homogenization process, whereas the network-based
approach postulates a specialization process. The first CAS property (i.e. no global
optimum) fits well with the idea of specialized agents, but the “anticipation” and
“innovation” properties are not imposing any type of pre-determined process. Agents
are heterogeneous, and KIBS may allow for both homogenization and specialization
trajectories, admitting that a general/cluster-level trajectory can be found. The two
other CAS properties (i.e. innovation and anticipation) indicate that the alternative
routes in Menzel and Fornahl (2011) (i.e. adaptation, renewal and transformation) are
the rule rather than the exception in CAS dynamics. It follows that a proper
deterministic life cycle is likely to be the reflect of a degenerative cluster (ex. Route
128 in Saxenian, 1994).

Clusters have already been modelled as CAS thanks to agent-based modelling 6

(Dilaver et al., 2014; Boero et al., 2004; Squazzoni and Boero, 2002; Albino et al.,
2003). Important question to tackle within this perspective are how specialized agents
tie to each other and then how these ties evolve. In this respect, Boero et al. (2004)
propose various matching strategies. As an example – for a specific agent – the
strategy could be: “look at the first agent with different technology/techno-
organizational asset you meet” (p. 12). These theoretical efforts are welcome, but the
building of relevant models has to rely on a set of well-established stylized facts
(Borrill and Tesfatsion, 2011). Discussing about clusters‟ dynamics and their main
drivers supposes to decide in a first step which aggregate indicators/variables
(number of actors, quantity produced, number of patents, R&D expenditures, etc.)
and which underlying networks to consider. Unfortunately, to our knowledge, there is
still no such empirical study within a CAS framework. Indeed, although theoretically
appealing, the CAS approach remains hard to put into practice. We thus propose, in
the remaining of the paper, a strategy for operationalizing the CAS framework.

2. Clusters as CAS: empirical example and insights

about clusters dynamics
In order to conduct an empirical investigation, we have to make several choices: (a)
which industry to study? (b) Which cluster to focus on within this industry? (c) What
kind of networks are we looking at? After making our choices, we explore the
dynamics of the selected empirical network of agents and we draw general
conclusions about clusters dynamics.

a. Which industry to study?

6
Applied to economics, agent-based modelling is “a computational approach that aims to explain
economic systems by modeling them as societies of intelligent software agents. The individual agents
make autonomous decisions, but their actual behaviors are constrained by available resources, other
individuals‟ behaviors, and institutions” (Osinga et al., 2011).
 11

The “non-universal competitor” principle (Holland, 2012) advocates for the study of
sectors with complex technology landscapes7, because they are the most likely to
offer many niches and thus many opportunities for clusters to emerge. According to
Arthur (2009), all products display a tree-like recursiveness: “the technology is the
trunk, the main assemblies the main branches, their subassemblies the sub-branches,
and so on, with the elemental parts the furthest twigs … The depth of this hierarchy is
the number of branches from trunk to some representative twigs” (p. 38). The more
complex a product is, the more it relies on a complex technology – i.e. a technology
with a high depth. Arthur (2009), among others (Frenken, 2006; Niosi and Zhegu,
2005), argues that the aircraft industry relies on very complex technologies,
composed of many subparts.

The resulting hierarchy between producers of various airplanes‟ subparts is

represented in Figure 3, from Niosi and Zhegu (2005). Following the CAS approach,
clusters can be found in any layer of this hierarchy. The biggest clusters include
agents involved in top layers, and more particularly the “prime contractors” or
“airframe assemblers”: Bombardier in Montréal, Airbus in Toulouse, Boeing in Seattle
(Niosi and Zhegu, 2005) or Lockheed Martin in Los Angeles (Scott, 1990).
!

Figure 3: Aircraft producers' pyramid (from Niosi and Zhegu, 2005 p.8)

7
Kauffman et al (2000) define a technology landscape as a set of values attributed to all the various possible
“production recipes” (p. 8), which are represented as vertices of a “directed graph”. A production recipe
“encompasses all the deliberate organizational and technical practices which, when performed together, result
in the production of a specific good” (p. 4).
Technology landscape is a metaphor originated from biology (Kauffman, 1993) for representing the choice of
economic agents when they have to decide what to produce and how to produce it. It states that agents’ initial
choice has long term incidence on their adaptability since it constraints their innovation capabilities, this is the
reason why production recipes are embedded into a directed graph: it is not possible to switch easily from a
recipe to another.
 12

b. Which cluster to study?

For every sector, it is common to find contributions focusing on successful/first class
clusters: Route 128, Silicon Valley, Detroit, Los Angeles (Saxenian, 1994; Klepper,
2010; Scott, 1990). The most important ones for the aircraft industry have already
been mentioned (Montréal, Toulouse, Seattle, Los Angeles) and won‟t be considered
for the present study. Indeed, we rather choose to focus on a niche cluster, for
highlighting the difference between the CAS and the more traditional/aggregate (or
life cycle) approach of the clustering phenomenon.
In the traditional approach, authors generally study the geographical concentration of
employment or companies by industry (Shin and Hassink, 2011; Niosi and Zhegu,
2005; Scott, 1990), or the geographical concentration of production (Shin and
Hassink, 2011) or of the innovation activity (Audretsch and Feldman, 1996a; 1996b)
for a given industry. Interestingly, there are very few accounts of the actual linkages
between these actors, although Saxenian (1994) reported – with the example of the
Route 128 – that spatial proximity does not necessarily imply strong cooperation.
One can also highlight a very limited account for actors‟ diversity within this
aggregate approach, since only “firms” are generally mentioned.
Figure 4 (left) displays the most recent account for the number of employees in
aerospace industry within all European regions. We observe that aerospace
employment is widespread, although quite concentrated in Western Europe and
Russia. This dispersion is higher than what is observed in the United States and is
generally explained by political reasons, notably the need to ensure countries
independence (Niosi and Zhegu, 2005). In an aggregate view, we would consider
studying the Southwest of France, Northern Germany or South England. However as
far as we favor the CAS approach, we rather choose a small cluster. Figure 4 (right)
displays a map of Belgium and the regional concentration of employment in
aerospace. We distinguish two relatively major poles within the southern regions -
surrounding the cities of Mons and Liège – and two smaller poles near Brussels and
Leuven. In 2006, all the actors from these 4 poles joined to create an official
association called Skywin.8 We propose to study this cluster.

9
Figure 4: Number of employees in the aerospace industry in Europe (left) and in Belgium (right) in 2011

8
http://www.skywin.be/?q=en (last access: 10 Feb. 2015)
9
www.clusterobservatory.eu (last access: 10 Feb. 2015, the numbers are for 2011)
 13

c. What kind of networks are we looking at?

As we have already pointed out, a CAS is primarily a network of heterogeneous and
specialized agents, which can be described in the following terms: (i) it emerges
within (technological or market) niches, (ii) it is strongly oriented towards innovation
and (iii) it is quick to adapt to changes in its environment thanks to its agents‟
anticipations and innovations (Holland, 2012). Skywin is presented as “a group of
companies, training centers and research units engaged in public and private
partnership and building synergies around common and innovative projects”.10 These
projects fit into six main axes that the agents anticipate to be of strategic importance
for their future development11:
1. Composite materials and processes
2. Metallic materials and processes
3. Embedded systems
4. Airport services
5. Space applications and systems
6. Modelling and simulation

These six themes reflect the niche position of Skywin within the aircraft producers‟
pyramid (Figure 3), as they mainly fit into some of the third tier activities: fuselage
and structure for the first two axes and the electronic systems for the third axe.
Interestingly, the fourth one - airport services - is not part of the aircraft production
process and it responds to a potential market in developing countries. 12 This
exemplifies clusters ability to re-orient their activity through time. The sixth axe –
“modeling and simulation” – arguably applies in every parts of the pyramid (Figure 3)
since simulation is generally involved in the conception phase of any airplane
components. Finally, the fifth axe on “space applications and systems” reveals a
specialization relevant for the space industry (not considered in the present work).

Skywin is thus a group of heterogeneous agents, and the main interactions we

should look at are those taking place within these “common and innovative projects”.
An exhaustive list is provided by the cluster website 13, which covers 46 common
projects undertaken collectively between 2006 and 2014. The following informations
are provided for every project: the agents involved - classified into two categories

10
http://www.skywin.be/?q=en/mission_and_strategy (last access: 10 Feb. 2015)
11
http://www.skywin.be/?q=en/mission_and_strategy (last access: 10 Feb. 2015)
12
As an example, it has been reported that 25.2% of flights in Mainland China have been delayed in
2012, 14.9% of them for an “unidentified reason”.
Beijing News, 23 May 2013:
http://www.bjd.com.cn/10beijingnews/201305/23/t20130523_3774403.html
13
http://www.skywin.be/sites/default/files/kcfinder/1-
vincent.marchal%40skywin.be/files/Fiches%20projets%20Skywin%20Calls%201-9.pdf (last access:
10 Feb. 2015)
 14

(“industries” and “research bodies” 14) – as well as the total budget (in millions of
Euros) and the duration of the project (start and end years).

These informations allow for the building of a bi-partite relational database linking two
sets of nodes (cf. the definition of a network): the agents and the projects in which
they are involved. This database makes it possible to draw Figure 5. In this figure,
red ellipses represent the projects and the blue ones represent the agents, whether
they are “industries” or “research bodies”. Links represent somehow the involvement
within a project or more exactly the level of expected involvement, as it may be
expressed by the budget allocated to the project15. The thicker and darker they are,
the higher the project‟s financial value (and the expected involvement) is. This Figure
does not distinguish between time periods: it summarizes all the interactions that took
part within Skywin from 2006 to 2014.

Figure 5: Skywin bi-partite innovation network (blue vertices = Agents; red vertices = projects)

d. Studying clusters dynamics through innovation networks

In Figure 5, the agents are not directly linked, that‟s why we modify the network on
the basis of the hypothesis that agents who are participating to the same project are
in fact directly linked. We also consider the time dimension and obtain as a result the
9 configurations of the network for each year from 2006 to 2014 displayed in Annex 1.
Table 1 summarizes some descriptive statistics of these configurations.

14
These “research bodies” include universities, training centers and private research institutions. We
thus assimilate them to KIBS.
15
It should be noted that as far as there is no information on how the budget is allocated between the
different partners of a given project, the whole budget is associated with each of them.
 15

Av. Path Length18

Number of Av. Density (No Loops among reachable
Actors Degree16 allowed)17 pairs
2006 15 7.33 0.52 1.00000
2007 44 13.54 0.31 1.72740
2008 52 13.57 0.26 1.88690
2009 61 14.46 0.24 1.90169
2010 67 14.68 0.22 1.90629
2011 61 13.97 0.23 1.84570
2012 62 13.45 0.22 1.76497
2013 60 10.53 0.18 1.82214
2014 56 9.53 0.17 1.79285
Table 1: Properties of Skywin temporal networks

In a traditional – aggregate – perspective, authors focus on the number of actors

within a given geographical area. In table 1, we rather report the actors who actively
cooperate on a set of common projects. Looking at first at the number of these actors
– and abstracting from the European economic crisis during the period - we may
consider that Skywin enters into a phase of decline starting from 2010. This general
movement goes hand in hand with a decrease in the average degree and networks‟
densities, which means that the agents who take part to innovation projects are more
and more loosely connected. In other words, the cluster loses its attractiveness for
potential newcomers: knowledge-seeking agents cannot find enough useful
knowledge for exploring their technological or market landscapes, or simply the
niches within which Skywin is evolving are not promising enough. Overall, these
evolutions are consistent with the life-cycle theory.

However, several elements contradict this pessimistic and deterministic conclusion. (i)
The configurations of the network displayed in Annex 1 are particularly dynamic
throughout the considered time span. From year to year actors are leaving and
others are entering into the network, following projects life cycles. Active actors are
thus changing: looking at the network‟s configurations, we count in total 92 different
active agents from 2006 to 2014, although no more than 67 were operating at the
same time. This suggests a positive rate of turnover among these agents. (ii) There
remain plenty of opportunities for partnerships, as innovation networks‟ densities

16
The average degree of a network gives the average number of links per agent within the network.
For instance, the agents taking part to innovation projects in 2008 were, on average, linked to 13.57
agents.
17
Network density is the ratio between the actual number of links and the maximum possible in a
hypothetical situation where each agent is connected to all the others. For instance, a density equal to
0.52 means that 52 % of the connection possibilities are exploited by the agents.
18
The path length between two agents i and j is the shortest distance between them (i.e. the shortest
sequence of vertices). The average path length of a network is obtained by averaging all the path
length between all the reachable agents of this network. Isolated agents are thus not taken into
account.
 16

always evolve within a range comprised between 17 % and 31 %. 19 In this respect,

Skywin accounts for a total of 117 members.20 Considering the 92 active agents, this
highlights both a high rate of cooperation between Skywin members and the
existence of unused opportunities. (iii) The path length values in table 1 are always
much smaller than the number of agents within the network. This reveals the
presence of a small-world effect (Newman, 2003): because of a high rate of overlap
between the members of different projects, it becomes easy for any agent – even for
a newcomer – to obtain information or knowledge coming from other projects. These
short path lengths can thus be considered as a good proxy for evaluating the
advantages of taking part in this cluster. Considering the relative stability of this
measure in Table 1, we cannot sustain the declining hypothesis that results from the
sole observation of the number of active agents.

We argue that, in a knowledge-seeking perspective, agents benefit from the

existence of a small-world effect, which itself emerges from a certain degree of
overlap between the various projects. It follows that the core agents of a given cluster
are those who allow for overlaps to occur. A good way to identify them is to compute
agents‟ degree centralities, i.e. for a given agent to count the number of links
he/she has with the other agents in the network. We can also take into account the
fact that different links are not equivalent, in the sense that – at least for our current
networks – the financial values of the various projects are different. In order to take
these links‟ values into account, we compute the weighted degree of centrality
measure proposed by Opsahl et al. (2010).

Be a tuning parameter, the number of links connected to the agent and

the average weight (or value) of these links, then , the weighted degree
centrality of the agent is given by the following equation:

The more is important, the more we attribute importance to links‟ values in

computing agents‟ centralities. Top 10 weighted degree centralities for every year
and for various values are reported in Annex 2. We observe that, in virtually all
cases, the two most central actors belong to the “research bodies” category. The
University of Liège and the Catholic University of Leuven are particularly central.
Industrial firms are also well represented in these rankings and, even though we
observe the recurrence of national leaders in the aerospace industry (e.g. Sonaca),
they are much more “volatile” than universities in the sense that their relative
positions are less stable and that there is a much higher rate of turnover among firms
within the top 10.

19
We omit the density of the network configuration in 2006 (date of the creation of the cluster) because its
relatively high value is explained by the fact that the cluster included only one project with several actors.
20
http://www.skywin.be/?q=en/members
 17

These results give interesting insights about the drivers of clusters‟ attractiveness –
measured in our case by a small-world effect within clusters‟ innovation networks. We
show that this attractiveness relies on the presence of a stable core of highly
connected knowledge intensive business services (universities or research bodies in
general). Arguably, what determines if a cluster is declining is not the age of the
cluster as a whole, nor the number of the (active) agents it includes, but the quality
and connectivity of the knowledge intensive business services in its core part. When
looking at the number of innovation-active agents, we could say that Skywin is
entering into a phase of decline, but a closer look to its innovation networks‟
properties reveals an attractive cluster and thus show potential for new phases of
growth.

Conclusion
Clusters‟ dynamics are generally understood as the evolution of an aggregate
indicator, like the number of firms operating in a given geographical area. Two
competing theories aim at explaining the dynamics of this indicator: the life cycle
theory and the network-based approach. Both consider knowledge processing as the
main driver but in an opposite way. For the tenants of the life cycle, clusters evolve
through a process of knowledge homogenization among their members, whereas the
network-based approach considers that knowledge becomes more and more
specialized. We argue that KIBS play a major role in both of these directions, and we
thus advocate for an alternative/ synthesizing approach.

Such synthesis should combine the aggregate point of view of the life cycle theory
with the actor-centered network-based approach, while avoiding their deterministic
predictions. Complex adaptive systems are a promising candidate for such a purpose.
In order to consider their implications for clusters‟ dynamics, we conducted within this
framework an empirical analysis on a given industrial cluster (the aeronautics cluster
in Belgium).

We discovered that this cluster‟s innovation networks exhibit a small-world effect.

This implies that any agent who takes part into an innovation project of this cluster
can easily benefit from knowledge and information generated in another ongoing
project. We argue that this effect is an interesting proxy of a cluster‟s attractiveness
and an appropriate aggregate variable for studying clusters‟ dynamics as it shows
cluster‟s potential for further growth. We also demonstrate that KIBS are the main
responsible for the emergence of this small-world effect in innovation networks.

References
Abernathy W. J. and Utterback, J. M. (1978), “Patterns of industrial innovation”,
Technology Review, Vol. 80 (7) pp. 40-47
 18

Albino V., Carbonara N., Giannoccaro I. (2003), “Coordination mechanisms based on

cooperation and competition within industrial districts: an agent-based
computational approach”, The Journal of Artificial Societies and Social Simulation,
vol.6 http://jasss.soc.surrey.ac.uk/6/4/3.html

Arthur W.B. (2009), The Nature of Technology. What Is It and How It Evolves, Free
Press

Audretsch D.B., Feldman M.P. (1996a), “Innovative clusters and the industry life
cycle”, Review of Industrial Organization, vol. 11 pp. 253-273

Audretsch D.B., Feldman M.P. (1996b), “R&D spillovers and the geography of
innovation and production”, The American Economic Review, vol. 86 pp. 630-640

Barras R. (1986), “Towards a theory of innovation in services”, Research Policy, vol.

15 pp. 161-173

Boero R., Castellani M., Squazzoni F. (2004), “Micro behavioral attitudes and macro
technological adaptation in industrial districts: an agent-based prototype”, The
Journal of Artificial Societies and Social Simulation, vol. 7
http://jasss.soc.surrey.ac.uk/7/2/1.html

Borrill P., Tesfatsion L. (2011), “Agent-based modeling: the right mathematics for the
social sciences?”, in Davis J.B. and Hands D.W. (Eds) The Elgar Companion to
recent Economic Mathodology, Edward Elgar

Boschma R., Fornahl D. (2011), “Cluster evolution and a roadmap for future
research”, Regional Studies, vol. 45 pp. 1295-1298

Callon M. (1991), “Réseaux technico-économiques et irréversibilité”, in Boyer., et al.

(ed.), Figures de l’irréversibilité en économie, Paris, Edition de l'EHESS, pp.195-
230.

Cumming G.S., Collier J. (2005), “Change and Identity in Complex Systems”,

Ecology and Society, vol. 10, pp.

Den Hertog P. (2000), “Knowledge-intensive business services as co-producers of

innovation”, International Journal of Innovation Management, vol. 4 pp. 491-528

Den Hertog P. (2002), “Co-producers of innovation: on the role of knowledge-

intensive business services in innovation”, in Gadrey J. and Gallouj F. (Eds.)
Productivity, Innovation and Knowledge in Services, Edward Elgar

Desmarchelier B., Djellal F., Gallouj F. (2013a), “Knowledge-intensive business

services and long term growth”, Structural Change and Economic Dynamics, vol.
25 pp. 188-205
 19

Desmarchelier B., Djellal F., Gallouj F. (2013b), “Environmental policies and eco-
innovation by service firms: an agent-based model”, Technological Forecasting &
Social Change, vol. 80 pp. 1395-1408

Dilaver O., Bleda M., Uyarra E. (2014), “Entrepreneurship and the emergence of
industrial clusters”, Complexity, vol. 19 pp. 14-29

Frenken K. (2006), Innovation, Evolution and Complexity theory, Edward Elgar

Frenken K., Cefis E., Stam E. (2015), “Industrial dynamics and clusters: a survey”,
Regional Studies, vol. 49 pp. 10-27

Gadrey J. Gallouj F. (1998), “The provider-customer interface in business and

professional services”, The Service Industries Journal, vol. 18 (2), April, pp. 1-15

Gallouj F. (1998), “Innovating in reverse: services and the reverse product cycle”,
European Journal of Innovation Management, vol. 1 pp. 123-138

Gallouj F. (2002), “Knowledge-intensive business services: processing knowledge

and producing innovation”, in Gadrey J. and Gallouj F. (Eds.) Productivity,
Innovation and Knowledge in Services, Edward Elgar

Gallouj F., Weinstein O. (1997), “Innovation in services”, Research Policy, vol. 26 pp.
537-556

Holland J.H. (2012), Signals and Boundaries. Building Blocks of Complex Adaptive
Systems, The MIT Press

Hakansson H. (1989), Corporate technological behavior, cooperation and networks,

London, Routledge

Hakansson H., Johanson J. (1993), „The network as a governance structure: interfirm

cooperation beyond market and hierarchie‟s, in Grabher G. (ed), The embedded
firm: on the socio-economics of industrial networks, London, Routledge, p. 35-51

Kauffman S. (1993), The Origins of Order. Self-Organization and Selection in

Evolution, Oxford University Press
Kauffman S., Lobo J., Macready W.G. (2000), “Optimal search on a technology
landscape”, Journal of Economic Behavior & Organization, vol. 43 pp. 141–166
Klepper S. (1996), “Entry, exit, and innovation over the product life cycle”, The
American Economic Review, vol. 86 pp. 562-583

Klepper S. (1997), “Industry Life Cycles”, Industrial and Corporate Change, vol 6(1)
pp. 145-181

Klepper S. (2010), “The origin and growth of industry clusters: the making of Silicon
Valley and Detroit”, Journal of Urban Economics, vol. 67 pp. 15-32
 20

Lancaster K.J. (1966), “A new approach to consumer theory”, Journal of Political

Economy, vol. 74 pp. 132-157

Lau A.K.W., Lo W. (2015), “Regional innovation system, absorptive capacity and

innovation performance: an empirical study”, Technological Forecasting & Social
Change, vol. 92 pp. 99-114

Martin R., Sunley P. (2011), “Conceptualizing cluster evolution: beyond the life cycle
model?”, Regional Studies, vol. 45 pp. 1299-1318

Menzel M., Fornahl D. (2010), “Cluster life cycles – dimensions and rationales of
cluster evolution”, Industrial and Corporate Change, vol. 19 pp. 205-238

Miles I., Kastrinos N., Flanagan K., Bilderbeek R., den Hertog P., Huntink W.,
Bouman M. (1995), “Knowledge-intensive business services. Users, carriers and
sources of innovation”, Report to DG13 SPRINT-EIMS

Muller E., Zenker A. (2001), “Business services as actors of knowledge

transformation: the role of KIBS in regional and national innovation systems”,
Research Policy, vol. 30 pp. 1501-1516

Newman M.E.J. (2003), “The structure and function of complex networks”, SIAM
Review, vol. 45 pp.167-256

Niosi J., Zhegu M. (2005), “Aerospace clusters: local or global knowledge spillovers?”,
Industry and Innovation, vol. 12 pp. 1-25
Opsahl T., Agneessens F., Skvoretz J. (2010), “Node centrality in weighted networks:
generalizing degree and shortest paths”, Social Networks, vol. 32 pp. 245-251
Osinga S., Hofstede G.J., Verwaart T. (2011), Emergent Results of Artificial
Economics, Springer

Porter M.E. (1998), “Clusters and the new economics of competition”, Harvard
Business Review, November-December pp. 77-90

Porter M.E. (2000), “Location, competition, and economic development: local clusters
in a global economy”, Economic Development Quarterly, vol. 14 pp. 15-34

Saviotti P.P., Metcalfe J.S. (1984), “A theoretical approach to the construction of

technological output indicators”, Research Policy, vol. 13 pp. 141-151

Saxenian A. (1994), Regional Advantage. Culture and Competition in Silicon Valley

and Route 128, Harvard University Press

Scott A.J. (1990), “The technopoles of Southern California”, Environment and

Planning A, vol. 22 pp. 1575-1605
Shin D.H., Hassink R. (2011), “Cluster life cycles: the case of the shipbuilding
industry cluster in South Korea”, Regional Studies, vol. 45 pp. 1387-1402
 21

Squazoni F., Boero R. (2002), “Economic performance, inter-firms relations and local
institutional engineering in a computational prototype of industrial district”, The
Journal of Artificial Societies and Social Simulation, vol. 5
http://jasss.soc.surrey.ac.uk/5/1/1.html

Utterback J.M., Abernathy W.J. (1975), “A dynamic model of process and product
innovation”, The International Journal of Management Science, vol. 3 pp. 639-656
 22

Annex 1: Skywin‟s innovation networks from 2006 to 2014

These networks have been obtained by formulating the hypothesis that agents‟
participating to the same project are linked together. All nodes thus represent firms or
research centers. Isolated agents are those who are taking part in projects in which
they are the only actor involved. In a given graph, darker and thicker links represent
partnerships in projects with relatively higher financial values.
i. Figure 6 : 2006

ii. Figure 7: 2007

 23

iii. Figure 8: 2008

iv. Figure 9: 2009

 24

v. Figure 10: 2010

vi. Figure 11: 2011

 25

vii. Figure 12: 2012

viii. Figure 13: 2013

 26

ix. Figure 14: 2014

 27

Annex 2: Top 10 weighted degree centralities per year

2006 2007 2008 2009 2010 2011 2012 2013 2014
Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i)
Aquapole Research 6.80948 CetIndustryc Research 15.4418 ULg Research 18.7524 ULg Research 25.3146 ULg Research 27.7763 ULg Research 25.8309 ULg Research 28.9931 ULg Research 26.9847 ULg Research 25.3146
CSL Research 6.80948 UCL Research 14.7705 ULB Research 17.8305 Cenaero Research 20.4387 UCL Research 22.4516 UCL Research 20.8232 Cenaero Research 21.9441 UCL Research 18.8493 UCL Research 20.5866
Spacebel Industry 6.30957 ULB Research 14.7705 UCL Research 17.2989 UCL Research 19.8468 Cenaero Research 20.4387 Cenaero Research 20.3615 UCL Research 20.8232 Cenaero Research 16.0135 Cenaero Research 13.4368
CetIndustryc Research 6.30957 ULg Research 14.7705 UMH Research 17.0241 ULB Research 19.4721 ULB Research 19.4721 Techspace Aero
Industry 19.1457 Techspace Aero
Industry 19.1457 Techspace Aero
Industry 14.6372 GDTECH Industry 13.2441
CreactIndustryon
Industry 6.30957 UMH Research 14.7705 Techspace Aero
Industry 16.1466 Techspace Aero
Industry 19.2091 Techspace Aero
Industry 19.2091 GDTECH Industry 17.7651 GDTECH Industry 17.7651 GDTECH Industry 11.3942 Open EngIndustryneerIndustryng
Industry 11.1082
Ulyces Industry 6.30957 Sonaca Industry 14.213 Cenaero Research 15.4421 UMH Research 18.2283 CetIndustryc Research 18.2331 CetIndustryc Research 16.7783 ULB Research 14.2981 LMS SamtechIndustry 10.2474 Techspace Aero
Industry 9.8875
Amos Industry 6.30957 Techspace Aero
Industry 12.9557 CetIndustryc Research 15.4418 Sonaca Industry 15.9253 UMH Research 18.2283 ULB Research 14.2981 Open EngIndustryneerIndustryng
Industry 13.4635 Sobelcomp Industry 9.96496 Sobelcomp Industry 9.51092
Walphot Industry 6.30957 Samtech Industry 12.9557 Sonaca Industry 14.6256 CetIndustryc Research 15.4418 Sonaca Industry 15.9253 Sonaca Industry 13.124 Sonaca Industry 13.124 CoexpaIndustryr
Industry 9.24402 CoexpaIndustryr
Industry 8.79849
IndustryonIndustrycSoftware
Industry 6.30957 Cenaero Research 12.9557 Samtech Industry 12.9557 GDTECH Industry 15.1836 GDTECH Industry 15.1836 CIndustryssoIndustryd
Industry 12.6856 CetIndustryc Research 12.6856 UMons Research 9.05126 UMons Research 8.58552
WSLux Industry 6.30957 ThalesAS Industry 11.2266 GDTECH Industry 11.3936 Samtech Industry 14.6959 Samtech Industry 14.6959 MultIndustrytel
Research 12.6856 CIndustryssoIndustryd
Industry 12.6856 M3 Systems Industry 8.48205 Numflo Industry 8.48344
APC Research 6.30957 CIndustryssoIndustryd
Industry 11.2266 ThalesAS Industry 11.2266 Amos Industry 11.3936 CIndustryssoIndustryd
Industry 14.2038 UMH Research 12.1846 MultIndustrytel
Research 12.6856 Thales BelgIndustryum
Industry 8.00911 CSL Research 8.24125

Table 1: Weighted Degree Centralities (α=0.2)

2006 2007 2008 2009 2010 2011 2012 2013 2014
Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i)
Aquapole Research 3.31662 UCL Research 6.32456 ULg Research 7.68115 ULg Research 9.69536 ULg Research 10.247 ULg Research 9.59166 ULg Research 10.247 ULg Research 9.27362 ULg Research 9.69536
CSL Research 3.31662 ULB Research 6.32456 ULB Research 7.4162 UCL Research 8.12404 UCL Research 8.77496 UCL Research 8.12404 UCL Research 8.12404 UCL Research 7.14143 UCL Research 8.544
Spacebel Industry 3.16228 ULg Research 6.32456 UCL Research 7.2111 Cenaero Research 8.06226 Cenaero Research 8.06226 Techspace Aero
Industry 7.74597 Cenaero Research 8 Techspace Aero
Industry 6.55744 GDTECH Industry 6.63325
CetIndustryc Research 3.16228 UMH Research 6.32456 UMH Research 6.9282 Techspace Aero
Industry 7.81025 Techspace Aero
Industry 7.81025 Cenaero Research 7.68115 Techspace Aero
Industry 7.74597 Cenaero Research 6.245 Cenaero Research 6
CreactIndustryon
Industry 3.16228 CetIndustryc Research 6 Techspace Aero
Industry 6.7082 ULB Research 7.74597 ULB Research 7.74597 GDTECH Industry 7.34847 GDTECH Industry 7.34847 GDTECH Industry 5.2915 Open EngIndustryneerIndustryng
Industry 5.38516
Ulyces Industry 3.16228 Sonaca Industry 5.74456 Cenaero Research 6.32456 UMH Research 7.48331 UMH Research 7.48331 CetIndustryc Research 6.08276 ULB Research 5.83095 LMS SamtechIndustry 4.79583 Techspace Aero
Industry 5.2915
Amos Industry 3.16228 Techspace Aero
Industry 5.47723 CetIndustryc Research 6 Sonaca Industry 6.48074 CetIndustryc Research 6.85565 ULB Research 5.83095 Sonaca Industry 5.65685 CoexpaIndustryr
Industry 4.47214 CoexpaIndustryr
Industry 4.89898
Walphot Industry 3.16228 Samtech Industry 5.47723 Sonaca Industry 5.83095 GDTECH Industry 6.40312 Sonaca Industry 6.48074 Sonaca Industry 5.65685 Open EngIndustryneerIndustryng
Industry 5.65685 Sobelcomp Industry 4.47214 CSL Research 4.69042
IndustryonIndustrycSoftware
Industry 3.16228 Cenaero Research 5.47723 Samtech Industry 5.47723 Samtech Industry 6.245 GDTECH Industry 6.40312 UMH Research 5.38516 UMH Research 5.38516 UMons Research 4.24264 Numflo Industry 4.47214
WSLux Industry 3.16228 ThalesAS Industry 5.09902 ThalesAS Industry 5.09902 CetIndustryc Research 6 Samtech Industry 6.245 CIndustryssoIndustryd
Industry 5.19615 CetIndustryc Research 5.19615 M3 Systems Industry 4 Samtech Industry 4.47214
APC Research 3.16228 CIndustryssoIndustryd
Industry 5.09902 CIndustryssoIndustryd
Industry 5.09902 ThalesAS Industry 5.09902 CIndustryssoIndustryd
Industry 6.08276 MultIndustrytel
Research 5.19615 CIndustryssoIndustryd
Industry 5.19615 Sonaca Industry 4 Sobelcomp Industry 4.3589

Table 2: Weighted Degree Centralities (α=0.5)

2006 2007 2008 2009 2010 2011 2012 2013 2014
Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i) Agents Type Cwd(i)
Aquapole Research 1.61539 UCL Research 2.7081 ULg Research 3.14626 ULg Research 3.71328 ULg Research 3.7802 ULg Research 3.56163 ULg Research 3.62155 ULg Research 3.18699 ULg Research 3.71328
CSL Research 1.61539 ULB Research 2.7081 ULB Research 3.08461 UCL Research 3.32547 UCL Research 3.4296 UCL Research 3.16954 UCL Research 3.16954 Techspace Aero
Industry 2.93772 UCL Research 3.54599
Spacebel Industry 1.58489 ULg Research 2.7081 UCL Research 3.00598 Cenaero Research 3.18024 Cenaero Research 3.18024 Techspace Aero
Industry 3.13386 Techspace Aero
Industry 3.13386 UCL Research 2.70567 GDTECH Industry 3.32224
CetIndustryc Research 1.58489 UMH Research 2.7081 UMH Research 2.81953 Techspace Aero
Industry 3.17558 Techspace Aero
Industry 3.17558 GDTECH Industry 3.03966 GDTECH Industry 3.03966 GDTECH Industry 2.45739 Techspace Aero
Industry 2.83186
CreactIndustryon
Industry 1.58489 CetIndustryc Research 2.33133 Techspace Aero
Industry 2.78696 ULB Research 3.08134 ULB Research 3.08134 Cenaero Research 2.89763 Cenaero Research 2.91651 Cenaero Research 2.43545 CoexpaIndustryr
Industry 2.72774
Ulyces Industry 1.58489 Sonaca Industry 2.32181 Cenaero Research 2.59032 UMH Research 3.07215 UMH Research 3.07215 Sonaca Industry 2.43828 Sonaca Industry 2.43828 LMS SamtechIndustry 2.24447 Cenaero Research 2.6792
Amos Industry 1.58489 ThalesAS Industry 2.31594 CetIndustryc Research 2.33133 GDTECH Industry 2.70028 GDTECH Industry 2.70028 UMH Research 2.38005 UMH Research 2.38005 CoexpaIndustryr
Industry 2.16356 CSL Research 2.6695
Walphot Industry 1.58489 CIndustryssoIndustryd
Industry 2.31594 Sonaca Industry 2.32469 Samtech Industry 2.6538 Samtech Industry 2.6538 ULB Research 2.37793 ULB Research 2.37793 Sonaca Industry 2.06913 Open EngIndustryneerIndustryng
Industry 2.61069
IndustryonIndustrycSoftware
Industry 1.58489 MultIndustrytel
Research 2.31594 ThalesAS Industry 2.31594 Sonaca Industry 2.63731 Sonaca Industry 2.63731 Samtech Industry 2.2444 Open EngIndustryneerIndustryng
Industry 2.3768 SABCA Industry 2.06913 Amos Industry 2.50622
WSLux Industry 1.58489 Techspace Aero
Industry 2.31558 CIndustryssoIndustryd
Industry 2.31594 CetIndustryc Research 2.33133 CIndustryssoIndustryd
Industry 2.60494 Open EngIndustryneerIndustryng
Industry 2.2444 Samtech Industry 2.2444 Sobelcomp Industry 2.00703 Samtech Industry 2.47352
APC Research 1.58489 Samtech Industry 2.31558 MultIndustrytel
Research 2.31594 ThalesAS Industry 2.31594 MultIndustrytel
Research 2.60494 CetIndustryc Research 2.20523 CetIndustryc Research 2.12841 UMons Research 1.98867 Spacebel Industry 2.40225

Table 3: Weighted Degree Centralities (α=0.8)