Papers in Evolutionary Economic Geography
Knowledge networks in the Dutch aviation industry:
the proximity paradox
Tom Broekel and Ron Boschma
Knowledge networks in the Dutch aviation industry:
the proximity paradox
Tom Broekel and Ron Boschma1
Department of Economic Geography, Faculty of Geosciences, Utrecht
University, The Netherlands
The importance of geographical proximity for interaction and knowledge sharing has been
discussed extensively in economic geography. There is increasing consensus that it is one out of
many types of proximities that might be relevant. We argue that proximity may be a crucial
driver for agents to connect and exchange knowledge, but too much proximity between these
agents on any of the dimensions might harm their innovative performance at the same time. In a
study on knowledge networks in the Dutch aviation industry, we test this so-called proximity
paradox empirically. We find evidence that the proximity paradox holds to some degree. Our
study clearly shows that cognitive, social and geographical proximity are crucial for explaining
the knowledge network of the Dutch aviation industry. But while it takes cognitive, social and
geographical proximity to exchange knowledge, we found evidence that proximity lowers firms’
innovative performance, but only in the cognitive dimension.
Keywords: geographical proximity, knowledge networks, proximity paradox, social network
analysis, aviation industry
JEL Codes: R11, R12, O18, O33
1 The authors would like to thank Matté Hartog for his help.
In Economic Geography, few issues have been studied more frequently as the question of what
role geographical proximity plays for knowledge sharing and innovation. Backed by the
argument that the exchange of tacit knowledge requires face-to-face contacts, it has long been
emphasized that knowledge sharing is highly sensitive to geographical distance (Audretsch and
Feldman, 1996; Gertler, 2003). This view on the role of geographical proximity has recently
been challenged theoretically (see e.g., Boschma, 2005; Lagendijk and Oinas, 2005; Broekel and
Binder, 2007). This critical view has been initiated by the French school of proximity dynamics
(Rallet, 1993; Kirat and Lung, 1999; Rallet and Torre, 2005). Their critical voices particularly
emphasize that geographical proximity is just one dimension among a number of other proximity
dimensions that can explain interaction between geographically proximate actors.
Boschma (2005) proposed five dimensions of proximity that impact on the likelihood of
knowledge exchange between actors and their innovative performance. His claim is that
geographical proximity is neither a necessary nor a sufficient condition for inter-organizational
learning and innovation. Boschma (2005) also argued that geographical proximity is more likely
to become effective rather indirectly through the other types of proximity. Breschi and Lissoni
(2003) and Ponds et al. (2007), among others, have confirmed this empirically for social,
institutional and cognitive proximity.
Extending these ideas, in a recent paper, Boschma and Frenken (2009) introduced what
they describe as the so-called proximity paradox. While proximity may be a crucial driver for
agents to connect and exchange knowledge, too much proximity between these agents on any of
the dimensions might harm their innovative performance. So, while a high degree of proximity
may be considered a prerequisite to make agents connected, proximity between agents does not
necessarily increase their innovative performance, and may possibly even harm it. Following
Nooteboom’s work on optimal cognitive distance (Nooteboom, 2000), Boschma and Frenken
(2009) claim it depends on the (optimal) level of proximity whether a connection between agents
will lead to higher innovative performance or not.
This issue of the proximity paradox is put central in an empirical study on the knowledge
network of the Dutch aviation industry. The Dutch aviation industry is an interesting case,
because it lost its flagship, the Fokker Company, in 1996, after which it went through a major
restructuring and reorientation process. The question then is how the knowledge network looks
like in the post-Fokker period, and what are its main drivers. Among other things, we will test
whether a shared past in the Fokker company (as a proxy for social proximity) increased the
probability of two aviation firms to connect. This study draws on own data that were collected
through semi-structured interviews of 59 profit and non-profit organizations that are active in
manufacturing activities and engineering services in the Dutch aviation sector.
Our paper has two objectives. The first objective is to assess empirically the extent to
which the different forms of proximity affect the technical knowledge network in the Dutch
aviation industry. Employing social network analysis, our study confirms the importance of
cognitive, organizational, and social proximity for the structure of the technical knowledge
network. We also found geographical proximity to be a driver of network formation, even when
controlling for the other proximities. The second objective is to determine which proximities
determine the innovative performance of aviation firms, while controlling for the usual suspects.
Our study provides empirical evidence for the proximity paradox with respect to the cognitive
dimension, not the geographical and social dimension. That is, proximity is required to connect
firms, but it does not necessarily yield above average innovative performance of these firms.
The paper is structured as follows. In Section 1, the different dimensions of proximity are
discussed. We specify how they influence the likelihood that actors are linked and what that
means for their innovative performance. Section 2 provides a short description of the Dutch
aviation industry, the data and the variables we constructed. Section 3 will briefly present the
methodological tools (QAP and network autocorrelation regression) we employed. Results of the
analyses are presented and discussed in Section 5. Section 6 concludes.
Firms’ embeddedness in knowledge networks has increasingly been recognized as an important
determinant of their economic and innovative performance (see, e.g., Powell et al., 1996). Given
limited resources firms can invest into research and development, their ability to collaborate and
make use of external knowledge becomes crucial for their success. It is well known that firms
have different absorptive capacities, which matter for their usage of external knowledge (Cohen
and Levinthal, 1990). This determines not only their likelihood to engage in knowledge sharing
but also the likelihood that obtained knowledge can be successfully used and implemented.
It is widely accepted that, in addition to their absorptive capacity, other factors influence
economic actors’ decisions to become engaged into knowledge sharing activities. An argument
frequently put forward in the literature is that geographical proximity facilitates knowledge
transfer (Feldman and Florida, 1994). This view on the role of geography for knowledge
exchanges has recently been challenged with the argument that it is not mere co-location that
matters for knowledge exchanges, but membership in knowledge networks (Castells, 1996). In
this sense “geographical proximity only creates a potential for interaction, without necessarily
leading to dense local relations” (Isaksen, 2001, p. 110).
The French school of proximity dynamics (see e.g. Rallet and Torre, 1999; Rallet and
Torre, 2005) has played a prominent role in this debate. They claim that geographical proximity
is just one among a number of proximity dimensions. In this context, Boschma (2005) claimed
that geographical proximity is neither a necessary nor a sufficient condition for knowledge
sharing and innovation. He proposed five dimensions of proximity (cognitive, social,
organizational, institutional, and geographical) that may impact on the likelihood of knowledge
exchange between actors and their innovative performance. We will briefly discuss each of these
below (for an extensive treatment, see Boschma, 2005), with the exception of institutional
proximity, for which we have no variance in our study.
Cognitive proximity refers to the degree of overlap between two actors concerning their
knowledge bases. Actors need to have a sufficient absorptive capacity to identify, interpret and
exploit knowledge of other actors (Cohen and Levinthal, 1990). However, if two actors’
knowledge bases are too similar, the likelihood of an innovative recombination is lower than
when dissimilar knowledge bases are merged. According to Nooteboom (2000), there exists a
tradeoff “…. between cognitive distance, for the sake of novelty, and cognitive proximity, for the
sake of efficient absorption” (p. 152). The relationship between the cognitive distance between
two actors and their innovation performance is therefore to be expected to take an inverted u-
shape (Cohendet and Llerena, 1997). In other words, both very proximate and very distant actors
are likely to gain little from cooperating in innovation activities. The optimal level of cognitive
proximity follows from the need to keep some cognitive distance (to stimulate new ideas through
recombination) and to secure some cognitive proximity (to enable effective communication and
knowledge transfer). Moreover, high cognitive proximity generally implies that two firms have
very similar competences, which means that when they engage in knowledge exchange, they run
a serious risk of weakening their competitive advantage with respect to the network partner. It is
also for this reason that one expects that excessive cognitive proximity may be harmful to
performance (see Nooteboom et al., 2007; Boschma et al., 2009; Boschma and Frenken 2009).
Consequently, it is not so much the quantity of contacts and intensity of knowledge exchanges
that matters for firms’ success, but rather the type of knowledge exchanged, and how that
matches the existing knowledge base of the firms. In this respect, cooperation is most fruitful
when network partners have technologically related, not similar knowledge bases.
Actors may also be close or not in organizational terms. Boschma (2005) defined organizational
proximity “… as the extent to which relations are shared in an organizational arrangement, either
within or between organizations” (p. 65). It can be seen as a continuous scale going from
autonomy to control. It is very low for totally independent actors and very high for actors that are
part of the same hierarchical system. According to Boschma (2005), organizational proximity
helps to manage knowledge exchange and reduce transactions costs. However, excessive
organizational proximity may also hamper interactive learning, as it constrains flexibility.
An alternative way to define organizational proximity is the degree to which organizations
have similar routines and incentive mechanisms (Metcalfe, 1994). In innovation studies,
researchers tend to make a distinction between profit and non-profit organizations. To put it
simply, profit organizations have an interest to keep their knowledge away from competitors,
while non-profit organizations like universities have a public mission and, therefore, are more
open to exchange knowledge with others. Because of these different routines, a profit and a non-
profit organization have a low degree of organizational proximity, which lowers their probability
to connect and collaborate. This is in line with the problematic relationship between universities
and private firms, which has been documented extensively. According to Broekel and Binder
(2007), actors’ search biases make it more likely that non-profit organizations will interact with
other non-profit organizations. This might also be true for profit-oriented organizations,
especially when the firms are not direct competitors.
Social proximity refers to the social embeddedness of actors in terms of friendship, kinship, and
experience at the micro-level (Boschma, 2005). This has to be seen as being distinct from
institutional proximity, which refers to institutions (like ethnic and religious values) at the
macro-level. Of particular interest is the role of trust, which is likely to be positively influenced
by social proximity. Trust has frequently been argued to foster knowledge exchange (Maskell
and Malmberg, 1999). In particular with respect to secrecy and the dangers of free riding, trust-
based relations are often depicted as superior to anonymous or newly established relations.
Hence, social proximity should be a strong predictor of the existence of a link between two
actors. The existence of “old boys networks”, for instance, is likely to influence knowledge
sharing activities. The same can be argued for actors with a shared history, like being at the same
school, university or company, which generates a sense of belonging to the same community.
However, too much social proximity may also be harmful for innovative performance, because
of an overload of loyalty and commitment in social relationships (Boschma, 2005).
None of the previously introduced proximities require co-location. It can be argued, though, that
geography plays a role by facilitating the other types of proximities. Broekel and Binder (2007)
have argued that geographical proximity may also directly impact on the likelihood that actors
will exchange knowledge. Through various mechanisms, geography influences individuals’
motivation and search heuristics and can bias them towards spatially close knowledge sources.
Hence, geographical proximity influences the other types of proximity but also increases the
likelihood of two actors to engage in knowledge exchange more directly. However, while
geographical proximity may offer certain advantages to knowledge sharing activities, there is
increasing evidence that a dominance of local linkages may also reduce the innovative
performance of a firm (Boschma, 2005; Broekel and Meder, 2008).
In sum, there are good reasons for why these different types of proximities should impact on
actors’ knowledge networks. Although empirical research is still primarily focused on the role of
geographical proximity (see e.g. Jaffe, 1989; Audretsch and Feldman, 1996), there is a growing
number of empirical approaches that aim to disentangle the different types of proximity and their
effects on firms’ behavior (Fleming et al., 2007; Ter Wal, 2009). Breschi and Lissoni (2003) for
example, using patent citations show that geographical proximity looses its predictive power for
two actors being linked when controlling for social proximity. Ponds et al (2007) find that
geographical proximity is of smaller relevance for research collaborations between academic
organizations, as opposed to collaborations between academic and non-academic organizations.
However, they suggest that geographical proximity is still helpful to overcome institutional
barriers between different types of organizations. Cantner and Meder (2007) find that cognitive
and social proximity are relevant for cooperation activities. They demonstrate that the
technological overlap between two actors (cognitive proximity) positively influences the
likelihood of these actors to engage in cooperation. This is also true for having cooperation
experience with the same cooperation partner. In this latter respect, past cooperation experiences
may have led to a decrease of the social distance between the two partners, which positively
impacts on future cooperation. Mowery et al. (1998) found similar results and, moreover,
suggested an inverted U-shape relationship between the probability to cooperate and the
technological (cognitive) similarity of two actors. However, while most of these studies take into
account at least two types of proximity, they hitherto have ignored or overlooked the relevance
of the other types.
The proximity paradox
Another critique on existing studies is that in such a proximity framework, it remains unclear
whether in the end, the choice of network partner matters for firms’ economic performance. In a
recent paper, Boschma and Frenken (2009) introduced what they describe as the so-called
proximity paradox. While proximity may be a driver for agents to connect and exchange
knowledge, too much proximity between these agents on any of the dimensions might not
necessarily increase their innovative performance, and may possibly even harm it.
Following Nooteboom’s work on optimal cognitive distance (Nooteboom, 2000), Boschma
and Frenken (2009) suggest it might depend on the (optimal) level of proximity whether a
connection between agents will lead to a higher innovative performance or not. While the
cognitive dimension has drawn most attention in the literature, optimal levels of proximity may
exist for the other forms of proximity as well (Boschma, 2005). With respect to geographical
proximity, it has been suggested that a mixture of local and non-local linkages to be best for
firms, and a combination of local buzz and global pipelines to be best for the long-term evolution
of clusters. With respect to social proximity, the optimal social distance might consist of a
balance between embedded relationships within cliques and strategic ‘structural hole’
relationships among cliques (Fleming et al. 2007). Uzzi (1996) found evidence of a mixture of
low proximity (defined as arm’s length ties) and high proximity (depicted as embedded ties) to
be best for firms. An optimal level of organizational proximity may be accomplished by loosely
coupled networks with weak ties between autonomous agents, which combine advantages of
organizational flexibility and coordination (Grabher and Stark 1997). However, the different
forms of proximity may also interact in this respect. Excessive proximity in one dimension that is
compensated by some degree of distance on another dimension may still enhance the innovative
performance of a firm. For instance, a firm may have primarily relationships with other local
firms (meaning too much geographical proximity), but when these provide access to a range of
different but related knowledge bases, this might still positively affect firm performance.
Our paper aims to add to this literature in two ways. First, we test the influences of four
types of proximity (i.e. cognitive, social, organizational, and geographical) on the likelihood of
two firms to exchange knowledge in the Dutch aviation industry. Following the literature on
knowledge networks (Giuliani and Bell, 2005; Boschma and Ter Wal, 2007; Sammarra and
Biggiero, 2008), we focus on the exchange of technological knowledge, which is regarded as
most relevant for firms’ innovation activities in this sector. Second, we examine whether
cognitive, social, organizational and geographical proximity matter for firms’ innovation
performance, controlling for factors like absorptive capacity. We also test whether there is a
curvilinear relationship between the proximity dimensions and innovative performance. Doing
so, we determine whether the proximity paradox holds in the Dutch aviation industry.
The aviation industry is known for being highly agglomerated and clustered in a few locations
worldwide, like Seattle and Toulouse (Hickie, 2006). Large multinational firms like Airbus and
Boeing dominate this industry. Their headquarters and main facilities function as attractors for
other businesses, e.g., specialized suppliers, sub-contractors, and service companies. This caused
the emergence of the typical hub-and-spoke type structure in this industry (Gray et al. 1996).
There is no large company in the aviation sector that has its headquarters or a major
production facility in the Netherlands. In the past, this was different. Since its establishment in
1919, Fokker B.V. dominated the industry in the Netherlands for almost eighty years. Fokker
used to be a crucial player in the aviation related knowledge networks in the Netherlands (van
Burg et al., 2008). Its core business was the production of 50-100 seater planes, most famously
the F 27, as well as the more recent Fo 50 and Fo 100. At its peak in mid 1991, about 13,000
people worked for Fokker, before employment declined steadily to 7,141 in early 1996
(Ligterink, 2001). In 1996, Fokker had to declare bankruptcy for three core units. This meant the
biggest single job cut in the modern history of the Netherlands by putting 5,600 people out of
work (Reuters, 1996). While about 950 jobs have been saved through the founding of a new
firm, “Fokker Aviation”, which was eventually taken over by Stork B.V., the Dutch aviation
industry lost its sole aircraft producer and one of its technological flagships.
More than a decade later, the Dutch aerospace industry consists of about 80 firms, most of
which are SMEs (NAG, 2008). These firms contribute to 0.9 percent (275 million Euros) of the
EU-25 value added in manufacturing of air- and spacecrafts in 2003 (EUROSTAT, 2002). Total
employment is about 5,000 employees. These numbers exclude maintenance and overhaul of
aircrafts. Including these, the number of employees increases to 15,000, which generate a
turnover of 2.2 billion Euro (NAG, 2008). The industry surely has regained strength by
successfully filling niches in the aviation market (Heerkens, 1999). Nevertheless, the Dutch
aviation sector can be regarded as marginal in comparison to countries like Germany and France.
A study of the knowledge network in the Dutch aviation sector is interesting for many
reasons. First of all, aviation is known to be a highly knowledge intensive industry, in which
access to external knowledge might be crucial (Niosi and Zhegu, 2005). Secondly, it is
interesting to investigate how this knowledge network is shaped, in the absence of a dominant
player. Thirdly, it is intriguing to find out whether the Fokker Company, despite its bankruptcy a
long time ago, still affects the nature of this knowledge network. Many of the current
entrepreneurs and top-managers in the Dutch aviation industry may have been former employees
of Fokker. Given the strong exposure to the very distinct Fokker identity (Kriechel, 2003), their
knowledge searching and sharing activities are still likely to be shaped by these experiences and
biased towards their former co-workers (see, e.g., Broekel and Binder, 2007). This enables us to
construct a social proximity indicator, which might affect the knowledge network in the industry.
Our empirical study is based on own data collection. In late 2008-early 2009, we interviewed 59
organizations that belong to the aviation industry in the Netherlands. Most of these organizations
are members of the Netherlands Aerospace Group (NAG), which is the most important trade
organization. Their members account for about 95 percent of the total turnover generated by
Dutch firms in the aviation industry (NAG, 2008). In 2008, the organization had 83 members.
We interviewed only those members that were active in manufacturing and/or engineering, since
for these activities, innovation and the exchange of technological knowledge is likely to be of
utmost importance.2 This applies to 40 firms, of which we interviewed 37. The three firms that
were not interviewed do not show any eye-catching features. In the course of the interviews, five
additional firms were named as being relevant, which were not member of the NAG, but clearly
active in the aviation industry. All of these have been interviewed as well. This increased our
total population to 45, of which 42 have been interviewed by extensive semi-structured
interviews on the spot. Accordingly, our response rate is 93 percent.
The list of the NAG also includes non-profit organizations, which we interviewed in a
different way. These organizations, as well as additional non-profit organizations named during
the firm interviews as relevant knowledge sources, were asked to indicate the intensity of
interaction with the other relevant non-profit organizations. This applied to 17 organizations
2 We interviewed also 2 firms that were not active in manufacturing or engineering. These firms confirmed the low
importance of technological exchange for their firms’ competiveness. Moreover, none of the interviewed firms
mentioned any maintenance-oriented firm as a relevant knowledge source.
which increased our sample to 59 organizations.3 The intensity level ranged from 1 to 3, with 1
indicating no interaction, and 2 and 3 medium and very high intensity.
Figure 1 shows the technological knowledge network of the Dutch aviation industry, based
on the data of these 59 profit and non-profit organizations. The 59 vertices account for 146 edges
and the network has a density of 0.085. This density is half as big as the density observed by
Sammarra and Biggiero (2009) for the technological knowledge network in the Italian aerospace
cluster of Rome. This cluster consists, however, of only 33 firms. Nevertheless, knowledge
network relations are rather sparse in the Netherlands, which is also indicated by the large
number of isolates. Non-profit organizations, most noticeably the Technical University of Delft
and the successor of the Fokker Company (Stork Aerospace B.V.), represent important contacts
for firms and thus, take up prominent positions in the network.
- Figure 1 here -
As explained in Section 2, the first part of our analysis aims to assess the impact of the various
forms of proximity on the structure of the technological knowledge network of the Dutch
aviation industry. More in particular, we estimate the importance of the different types of
proximities on the likelihood that two actors are linked. Our dependent variable LINKTEC is
dichotomous and indicates whether actor i or j mentions the other as a relevant source of
technological knowledge. We assume that knowledge exchanges are always reciprocal in nature.
This implies that if i is naming j as relevant contact, we also assume that i is a relevant contact of
j. In other words, we assume an undirected network. As independent variables, we have included
the 4 dimensions of proximity, as discussed in Section 2, and we have measured these as follows.
3 While most of the interviewed firms do not show strong connections to the space industry, this is not true for the
non-profit organizations. However, our main focus is on firms.
To assess the effect of geographical proximity, we calculate the logarithm of the geographical
distance in kilometers between two actors, which results in a continuous positive variable.4
While other studies use travel time (Ejermo and Karlsson, 2006), the spatial scale of the network
of Dutch aviation industry is rather small. Few contacts are not located in the direct neighboring
countries of the Netherlands (including Great Britain), for why the use of travel distances is
unlikely to change the results. The logarithm ensures that outliers in form of trans-continental
relations do not disturb the estimations. We also estimate a squared term to control for non-linear
effects (GEO2). In order to reduce multicollinearity problems, we subtract the mean in advance.
We refer to cognitive proximity as the technological similarity of two actors’ knowledge bases.
We have constructed two measures. For the first measure, we rely on the technology classes that
are assigned by the Netherlands Aerospace Agency (NAG). The NAG defines 15 technologies of
which 13 are relevant for the firms considered in this study. The technological fields and the
according number of organizations are listed in Table 1. In case the interviewed organization is
not a member of the NAG, the profile was created on the basis of the organization’s webpage.
The variable TECNAG is defined dichotomously, with a value of one if both organizations are
active in the same technology, and zero otherwise.
- Table 1 here -
The second variable is based on 3-digit NACE codes assigned to each organization. The
assignment of the NACE codes has been done as follows. First, if an organization named another
organization being a relevant knowledge source, it was asked to provide information about the
content of the knowledge exchange, i.e. which technologies this exchange concerned, and to
what 3-digit NACE codes this corresponded. Second, we asked each organization to mention the
three most important sectors (3-digit NACE codes) from which they recruit their key personnel.
Lastly, we searched for information about the organizations on the Internet. This included the
4 We add 0.001 km to all distances to ensure a positive logarithm.
organizations’ own websites as well as the company information webpage:
www.mintportal.bvdep.com, which also classifies firms according to the NACE scheme. This
last option was particularly relevant for organizations that have not been interviewed, but which
were named as relevant knowledge sources. Based on these sources of information, each
organization has been characterized by a number of 3-digit NACE codes.
In order to define a similarity indicator for two organizations, we first had to define the
similarity between two technologies (NACE codes). In a first step, following Breschi and Lissoni
(2003), the similarity between technologies is estimated on the basis of their co-occurrence at the
level of the organization. If technology A has frequently been assigned to organizations that are
also characterized by technology B, the technologies are perceived being related. In addition to
this direct relation, we also consider the indirect relation between two technologies. Accordingly,
if technology A is frequently assigned to the same organizations as technology C, and the same
is true for technologies B and C, A and B are also comparatively similar. In practice, we estimate
the Cosine index, as given in Ejermo (2003):
with t as the number of technologies and g,k,z as indices of technologies under consideration. In
this equation, wzk is the number with which technologies z and k coincide at the organization
level. In total, 72 different technologies appear in our data set.
When organizations have multiple technologies, we do not know the technologies’ relative
importance, i.e. no information is available on the share of turnover or employees attributed to
each technology. We therefore estimate similarity in two ways. In the first, we search for the
most similar pair of technologies in the firms’ technology vectors. More precisely, we compare
two organizations’ (i,j) vectors of technologies (Ti and Tj). Next, we identify for each technology
z (z ! ti) of organization i, the maximal rizg within organization j’s technologies. The same is done
for the technologies of organization j. The rizg are added up and divided by the sum of the
number of technologies assigned to both organization i and j. The latter ensures that the resulting
similarity index sij is not biased positively towards diversified organizations (those with many
technologies assigned to). The estimation shows the following.
Because the values of the Cosine index rzg are between 0 and 1, the similarity index ranges
from 0 and 1 as well, with 1 indicating perfect technological similarity. In extreme cases, all of
organization i’s technologies are compared to one technology of organization j. The rationale
behind this is that lacking information on the relative importance of a technology for an
organization we assume that access to just one particular technology makes two organizations
similar. This is because their knowledge bases overlap, which enables efficient communication.
In the estimations we also consider a quadratic term of the similarity indicators to check for
non-linear effects. Because these variables are likely to cause multicollinearity with the
similarity indicators, we subtract the mean of the variable before the squaring.
Hence SIM2 will be large for small and large values of similarity.
We have constructed a dichotomous variable to account for the (likely) existence of social
relations between organizations. The variable (FOK) amounts to one if former employees of
Fokker B.V. are members of the top management of both firms and zero otherwise. As pointed
out in Section 2.1, having a shared past in Fokker may reflect a community feeling that might
still affect the structure of the network, even after its collapse. The importance of Fokker for the
Dutch aviation industry has been immense in the past (see van Burg et al., 2008). Hence, “old-
boys” networks may still be in place and give exclusive knowledge sharing opportunities. The
data for this variable were collected in the interviews by means of the following question: “Do
you or anyone else of the firm's top-management have a personal relation to the former Fokker
N.V., i.e. has been a former employee of Fokker N.V.?”