The Micro-Determinants of Meso-Level Learning and Innovation: Evidence from a Chilean Wine Cluster

Abstract

Most analyses of the relationship between spatial clustering and the technological learning of firms have emphasised the influence of the former on the latter, and have focused on intra-cluster learning as the driver of innovative performance. This paper reverses those perspectives. It examines the influence of individual firms’ absorptive capacities on both the functioning of the intra-cluster knowledge system and its interconnection with extra-cluster knowledge. It applies social network analysis to identify different cognitive roles played by cluster firms and the overall structure of the knowledge system of a wine cluster in Chile. The results show that knowledge is not diffused evenly ‘in the air’, but flows within a core group of firms characterised by advanced absorptive capacities. Firms’ different cognitive roles include some—as in the case of technological gatekeepers—that contribute actively to the acquisition, creation and diffusion of knowledge. Others remain cognitively isolated from the cluster, though in some cases strongly linked to extra-cluster knowledge. Possible implications for policy are noted.

Full text

Research Policy 34 (2005) 47–68
The micro-determinants of meso-level learning and innovation:
evidence from a Chilean wine cluster
Elisa Giuliania,b,∗, Martin Bella
aSPRU, University of Sussex, Freeman Centre, Falmer, Brighton, East Sussex BN1 9QE, UK
bDEA, Faculty of Economics, University of Pisa, Italy
Received 25 February 2004; received in revised form 13 October 2004; accepted 27 October 2004
Available online 21 December 2004
Abstract
Most analyses of the relationship between spatial clustering and the technological learning of firms have emphasised the
influence of the former on the latter, and have focused on intra-cluster learning as the driver of innovative performance. This
paper reverses those perspectives. It examines the influence of individual firms’ absorptive capacities on both the functioning of
the intra-cluster knowledge system and its interconnection with extra-cluster knowledge. It applies social network analysis to
identify different cognitive roles played by cluster firms and the overall structure of the knowledge system of a wine cluster in
Chile. The results show that knowledge is not diffused evenly ‘in the air’, but flows within a core group of firms characterised by
advanced absorptive capacities. Firms’ different cognitive roles include some—as in the case of technological gatekeepers—that
contribute actively to the acquisition, creation and diffusion of knowledge. Others remain cognitively isolated from the cluster,
though in some cases strongly linked to extra-cluster knowledge. Possible implications for policy are noted.
© 2004 Elsevier B.V. All rights reserved.
Keywords: Clusters; Absorptive capacity; Knowledge communities; Technological gatekeepers
1. Introduction
Over the years, the literature on industrial clusters
has emphasised their capacity as loci for knowledge
diffusion and generation. Industrial clusters, which are
defined here as geographic agglomerations of eco-
nomic activities that operate in the same or intercon-
∗Corresponding author.
E-mail address: E.Giuliani@sussex.ac.uk (E. Giuliani).
nected sectors,1have for this reason been considered a
source of dynamic endogenous development and have
received increased attention in both the academic and
1This definition both differs from and overlaps with the numerous
expressions adopted in the literature to analyse similar economic
phenomena, such as industrial districts, localised production sys-
tems, technology districts, milieux, etc. (for different definitions,
see among others Becattini, 1989; Humphrey and Schmitz, 1996;
Markusen, 1996; Porter, 1998; Capello, 1999; Altenburg and Meyer-
Stamer, 1999; Cassiolato et al., 2003).
0048-7333/$ – see front matter © 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.respol.2004.10.008

48 E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68
policy arenas. The importance of clustering for knowl-
edge diffusion and generation was seminally stressed
by Alfred Marshall, who introduced the concept of ‘in-
dustrial atmosphere’ (Marshall, 1919) and described
the district as a place where “mysteries of the trade be-
comenomysteries;butareasitwereintheair,andchil-
dren learn many of them, unconsciously” (Marshall,
1920, p. 225).
Following this seminal contribution, later scholars
have emphasised the importance of localised knowl-
edge spillovers for innovation, due primarily to the fact
that firms in industrial clusters benefit from the avail-
abilityofapoolofskilledlabourandthat,mainlydueto
geographical and social proximity, new ideas circulate
easily from one firm to another promoting processes
of incremental and collective innovation (see among
many others: Becattini, 1989; Asheim, 1994; Saxe-
nian,1994;AudretschandFeldman,1996;Maskelland
Malmberg, 1999; Belussi and Gottardi, 2000; Baptista,
2000).
At the same time, other contributions have stressed
the importance of extra-cluster networking, since the
mere reliance on localised knowledge can result in
the ‘entropic death’ of the cluster that remains locked-
in to an increasingly obsolete technological trajectory
(Camagni, 1991; Grabher, 1993; Becattini and Rullani,
1993; Guerrieri et al., 2001; Cantwell and Iammarino,
2003).
Increasing attention has been given to the influ-
ence of clustering on industry learning and competi-
tive performance in developing countries, the context
of this study (Humphrey and Schmitz, 1996; Nadvi
and Schmitz, 1999; Rabellotti, 1999; Cassiolato et al.,
2003). Within this, emphasis has been given to the in-
ternal characteristics of clusters: the spatial agglom-
eration of firms and the derived external economies,
together with various forms of ‘joint action’. However,
the ‘openness’ of cluster knowledge systems and their
capacity to interconnect with extra-cluster sources of
knowledge seems especially important in such techno-
logically lagging regions, industries or countries (Bell
and Albu, 1999).
This paper contributes to this field of study by con-
sidering, on the one hand, the linkages that clusters es-
tablish with extra-cluster sources of knowledge; on the
other, by trying to go beyond the received Marshallian
‘knowledgeintheair’ideaofintra-clusterlearning pro-
cesses. In addressing these two issues, the paper joins
other recent work in questioning the emphasis on spa-
tial proximity that has come to characterise much of
the literature about knowledge flows and technological
learning in clustered production activities.
Withrespect to thefirst issue, extra-cluster linkages,
the paper shares with recent work a sceptical view of
the commonly presumed close relationship between
functional, relational and geographical proximity. As
emphasised by Malmberg (2003), for instance, there
is no reason why learning processes should be terri-
torially bounded, and both “local and global circuits
of interactive learning” are likely to be important (p.
157). More broadly, Amin and Cohendet (2004) have
recently highlighted the distinction between relational
and spatial proximity as different contexts for learn-
ing, with the latter no more likely than the former to
shapelearning processes: “...relational or social prox-
imity involves much more than ‘being there’ in terms
of physical co-location ... Crucially, if the sociology
of learning is not reducible to territorial ties, there is no
compelling reason to assume that ‘community’ implies
spatially contiguous communities, or that local ties are
stronger than interaction at a distance” (p. 93).
On the second issue, intra-cluster learning, we
share with recent literature a scepticism about the role
of fuzzy social relationships and ill-defined spillover
mechanisms as the basis of knowledge flows and
learning processes within territory-bounded commu-
nities. Consequently, like several others (e.g. Dicken
and Malmberg, 2001; Malmberg and Maskell, 2002;
Amin and Cohendet, 2004), we search for more struc-
turedmechanisms that shape these flowsandprocesses.
However, we pursue this search in somewhat different
directions. We do not examine meso-level structures
(e.g.cluster labour markets,as suggested by Malmberg,
2003). Instead, like Owen-Smith and Powell (2004),
we focus on characteristics of the nodes in networks
as influences on the structure of knowledge flows. But
rather than the organisational and institutional char-
acteristics of network nodes, we examine their cog-
nitive characteristics. In particular, the paper focuses
on micro-level (firm-centred) knowledge endowments
and analyse how these influence the formation of intra-
and extra-cluster knowledge networks.
The study has been based on empirical evidence
collected at firm level in a wine cluster in Chile
(Colchagua Valley). Inter-firm cognitive linkages and
relational data have been processed through social

E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68 49
network analysis (Wasserman and Faust, 1994) and
graph theoretical indicators.
The paper is structured as follows: in Section 1,we
review some of the outstanding questions in this field,
outlinethetheoreticalframeworkand formulate the hy-
potheses for research. Section 2introduces the case
study cluster in Chile in the context of recent devel-
opments in the international wine industry. Section 3
explains the methodology applied in the research; and
Section 4presents the empirical evidence. A conclud-
ing discussion is provided in Section 5, with comment
on the possible policy implications.
2. Research questions and theoretical
framework
2.1. External openness and the concept of cluster
absorptive capacity
The degree of openness of a cluster is inevitably
tied to the degree of extra-cluster openness of its
member firms and institutions. At a meso level of
analysis, we define cluster absorptive capacity as the
capacity of a cluster to absorb, diffuse and exploit
extra-cluster knowledge (Giuliani, 2002).2The focus
of this study is specifically on firms’ abilities to access
and absorb external knowledge. However, external
knowledge lies not merely outside the firm, as with
Cohen and Levinthal (1990) absorptive capacity,
but it lies outside both the firm and the Colchagua
cluster. Consequently, we use external knowledge and
extra-cluster knowledge interchangeably.
It is argued here that firms are heterogeneous in
their capabilities and knowledge bases (Dosi, 1997)
and, therefore, they are likely to play different roles
in interfacing between extra- and intra-cluster knowl-
edge systems. At the micro level, absorptive capac-
ity is considered a function of the firm’s level of prior
knowledge (Cohen and Levinthal, 1990). It, therefore,
reflects the stock of knowledge accumulated within the
firm,embodied in skilled human resources and accrued
2This definition draws on the concept of the absorptive capacity
of firms, defined by Cohen and Levinthal (1990) as “the ability of a
firm to recognise the value of new, external information, assimilate
it, and apply it to commercial ends” (Cohen and Levinthal, 1990,p.
128).
through in-house learning efforts. Consequently, it is
defined here independently of any linkages with ex-
ternal sources of knowledge.3So, following the argu-
ment of Cohen and Levinthal, it is firms with higher
absorptive capacities in a cluster that are more likely to
establish linkages with external sources of knowledge.
Thisis explained on the basis of cognitivedistances be-
tween firms and extra-cluster knowledge, so that firms
with higher absorptive capacities are considered more
cognitively proximate to extra-cluster knowledge than
firms with lower absorptive capacities. From this, the
first hypothesis is elaborated:
Hypothesis 1. Firms with higher absorptive capacity
are more likely to establish knowledge linkages with
extra-cluster sources of knowledge.
From this, it would follow that a cluster does not
absorb external knowledge uniformly through all its
constituent firms, but selectively through only those
firms with a low cognitive distance from the techno-
logical frontier. Interestingly enough, firms with high
external openness could be potentially fruitful at lo-
cal level if they contribute to the diffusion of acquired
knowledge to other firms in the cluster, and perform as
technological gatekeepers (Allen, 1977; Rogers, 1983;
Gambardella, 1993). Accordingly, we expect that firms
perform different cognitive roles, according to their
knowledge bases, in interfacing with extra- and intra-
cluster knowledge.
2.2. Firms’ absorptive capacity and the
intra-cluster knowledge system
Several contributions in the economics of innova-
tion literature have emphasised that the propensity of
firms to establish knowledge linkages with other firms
is associated with the degree of similarity/dissimilarity
in their knowledge bases (see e.g. Rogers, 1983; Lane
andLubatkin,1998). Wedrawon this body of literature
and claim that, even within a cluster context, firms will
exchange knowledge depending on: (i) the amount of
knowledge they have accumulated over time and can,
therefore, release to others and (ii) their capacity to
3The operational measure of absorptive capacity is discussed in
Section 3.

50 E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68
decode and absorb knowledge that is potentially trans-
ferable from other cluster firms.
In particular, and in contrast to the conventional
knowledge-spillover story, the exchange of knowl-
edge follows some structured rules of behaviour which
are determined by the relative values of firms’ ab-
sorptive capacities (i.e. by the cognitive distances be-
tween them). So, while on one hand, when firms have
very similar absorptive capacities the exchange of
knowledge is more likely to occur on a mutual basis
(Coleman, 1990), as “reciprocity appears to be one of
the fundamental rules governing information trading”
(Schrader, 1991, p. 154);4on the other, it seems likely
that differences between the knowledge bases of firms
will lead them to play differing, sometimes asymmet-
ric roles within the cluster knowledge system. Hence,
firms with particularly advanced knowledge bases are
likely to be perceived by other cluster firms as ‘tech-
nological leaders’ or ‘early adopters’ of technologies
in the local area, leading to them being sought out as
sources of advice and knowledge more often than firms
with less advanced knowledge bases. This is also likely
to lead to a degree of imbalance in the knowledge in-
teractions of the leading firms since they are less likely
to seek out useful knowledge from firms with ‘lower’
knowledge bases (Schrader, 1991). Some firms may,
therefore, transfer more knowledge than they receive
from other local firms, so acting as net ‘sources’ within
the cluster knowledge system. At the same time, firms
have more incentives to ask for technical advice when
they know that they will be able to decode and apply
the received knowledge (Carter, 1989). Consequently,
while the similar levels of their knowledge bases may
lead some firms into balanced exchange, other firms
with lower but still significant capacities are likely to
absorb more knowledge than they release, so acting as
net ‘absorbers’ within the cluster knowledge system.
Finally, however, the knowledge base of some firms
may be so low that it neither offers anything of value
to other firms nor provides a capacity to acquire and
exploit knowledge that others may have. Such firms
4Although we recognise the well-known distinction between in-
formation and knowledge, we use the terms interchangeably here.
This is consistent with the contributions we discuss in this section
(i.e. Von Hippel, 1987; Carter, 1989; Schrader, 1991), in which a va-
riety of terms are used interchangeably: e.g. ‘know how’, ‘technical
information’ and ‘technical advice’.
are likely to be isolated within the cluster knowledge
system—a position that is not considered significant
within perspectives on cluster dynamism that empha-
sise the importance of homogeneous meso characteris-
tics leading to the pervasive availability of knowledge
and learning opportunities ‘in the air’.
These considerations lead to the formulation of the
following hypotheses:
Hypothesis 2(a). Links between local firms are more
likely to develop among firms with higher absorptive
capacity.
Hypothesis 2(b). Firms with differing levels of ca-
pacityarelikely to establish different kinds of cognitive
positions within the cluster knowledge system.
Underlying this argument about the differentiation
of roles is a more general set of issues about how com-
munication within a cluster knowledge system is struc-
tured. These are now explored.
2.3. Knowledge communities and the structure of
the intra-cluster knowledge system
Beside inter-firm cognitive distances, the structure
of the intra-cluster knowledge system is likely to be
influenced by the formation of local communities of
knowledgeworkers,5who share common language and
technical background, seek advice from other peers of
the same community and in so doing develop spon-
taneous (but not random) networking practices, which
boostprocessesofknowledgeexchange and generation
(Von Hippel, 1987; Haas, 1992; Wenger and Snyder,
2000; Lissoni, 2001).6
The formation of these communities is driven by
the existence of a certain degree of intra-community
homophily (Rogers, 1983; McPherson et al., 2001),
based upon the similarity of the members’ personal
5Knowledge workers are defined by the literature as individuals
with high education and training in a particular profession. These
characteristicsare normally combined witha high capability in prob-
lem solving (Drucker, 1993; Creplet et al., 2001).
6Such communities have been variously defined in the literature
e.g. as ‘communities of practice’ (Brown and Duguid, 1991; Wenger
and Snyder, 2000) or ‘epistemic communities’ (Haas, 1992). For an
insight into the differences between these two types of communities
see Creplet et al. (2001).

E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68 51
technical background, which is inevitably entangled
with the professional experience followed within the
firm where they work. At the same time, such knowl-
edgeworkersseekadvice from other community mem-
bers in search of complementary, different solutions
to their specific technical problems, or simply inter-
connect to exchange experiences and improve their
technical knowledge accordingly. It is conceivable
that such networks tend to be structured by a local
homophily-diversity trade-off. On the basis of this, we
do expect that local knowledge flows within ‘cogni-
tive subgroups’ of professionals (and, therefore, firms)
rather than randomly in the ‘air’ (Breschi and Lissoni,
2001). Hence the third hypothesis is formulated as
follows:
Hypothesis 3. The knowledge system within a clus-
ter will be structured and differentiated, reflecting the
existence of distinct ‘cognitive subgroups’.
Wetest each ofthesehypotheses inSections5.1–5.3,
aimingtoshedlightontherelationshipsbetweenfirms’
absorptive capacities and their patterns of extra- and
intra-cluster knowledge acquisition and diffusion. In
Section 5.4, we examine the connection between exter-
nal linkages and intra-cluster communication patterns,
the technological gatekeeper’ role as well as other cog-
nitive roles taken by firms in the cluster.
3. The wine cluster in Colchagua Valley
3.1. The context: the Chilean wine industry
In the past decade, the international wine industry
hasbeen characterisedbya veryrapidgrowthofexports
and by the emergence of new wine producing countries
and their entry into the global market. Besides tradi-
tional producers, such as France and Italy, ‘new world’
exporters (primarily Argentina, Chile, New Zealand,
South Africa and the US) have increased their share of
global exports and upgraded the quality of their wines
(Anderson and Norman, 2001a,b).
The case of Chile is an interesting example. De-
spite its long-standing tradition in the production of
wines (Del Pozo, 1998), it is only since the 1980s
that sustained growth in the production and export
of wine has been achieved. In the 1990s, the coun-
try’s participation in the global wine trade increased
at a rate of 27 per cent per year, and the quality
of the product was substantially improved, attain-
ing widespread positive appraisal from international
experts.
Chile presents ideal conditions for wine pro-
duction because of the country’s excellent natural
endowments that result in numerous wine regions
characterised by favourable terroir.7In addition
wineries have made considerable efforts to modernise
their technologies and adopt novel productive prac-
tices. Old methods have been replaced and firms have
deepened their commitment to experimentation and
upgrading of the production process. This rapid and
pervasive transition has been described as a ‘wine
revolution’ (Crowley, 2000).
Considerable investment at an institutional level
in Chile has supported the firm-level efforts to up-
grade and expand the Chilean industry. With respect to
technology, co-financing through competitive funding
schemes has sustained applied research in viticulture
and oenology and the interaction between wine pro-
ducers and various research institutes and universities
(e.g. Universidad Catolica, Universidad de Talca and
Universidad de Chile). With respect to marketing, the
export of wines has been supported through the ad-
vice and intermediation of specific institutions, such as
PROCHILE.
The Colchagua Valley is one of the promising
emerging wine clusters of the country (Tapia, 2001). It
is located in the VIth Region, about 180 km south-west
of Santiago and is closed off to the east by the Andes
and approximately 80km to the west by the Coastal
Range mountains. The area, is traditionally rural, with
a history of wine production dating back to the XIXth
century. It has recently increased its specialisation in
wine production and since the 1990s the cluster has
experienced a period of growth and prosperity, tied,
mainly, to the success of the wine industry (Schachner,
2002).
3.2. Key features of the Colchagua wine cluster
Being traditionally a wine producing area, the clus-
ter is populated by a myriad of predominantly micro
and small grape growers and wine producers. In many
7Terroir is a French term that refers to the combination of soil and
climatic conditions of a specific wine area.

52 E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68
cases, these produce for personal consumption or sell
grapes or processed bulk wine at the local level. In ad-
ditionto them, the cluster has been characterised by the
development of firms that produce wine for the domes-
ticand foreign market with an orientation towardsqual-
ity. Together with long established firms, domestic and
foreign investors have been attracted by the favourable
terroir and have, therefore, established new production
plants in the cluster.
During the 1990s, the favourable market conditions
boosted the planting of new vineyards, which doubled
in size in the second half of the 1990s (S.A.G., 2000).
Accordingly, in the same period, the local production
ofwinetripled(S.A.G., 2000).8Most of the firms in the
cluster have invested in new technologies. The cellar
is usually the first step made towards modernisation:
French steel tanks for fermentation, French or Ameri-
can barriques, Italian bottling lines are not difficult to
findin the cluster.Firms are also very dynamic in intro-
ducing new methods and techniques in pruning, irriga-
tion and canopy management. More recently, starting
from the end of the 1990s, the sustained growth trend
culminatedin an overproductioncrisis.Wineproducers
are currently affected by a global slow down of wine
consumption and by increased competition (Anderson
and Norman, 2001a,b), and this has spurred them to
intensify their efforts to improve product quality and
to enter higher value niches in international markets.
Acritical role in the recent innovationprocess of the
wine industry has been played by specialised knowl-
edge workers, such as oenologists and agronomists
employed in firms. Having university qualifications
in technical fields, these professionals have the scien-
tific understanding of the wine making process, which
allow new methods of production to be applied and
moreintenseexperimentationtobe carried out in firms.
Besides, such knowledge workers boost technologi-
cal change in firms also as external consultants. Also
known as ‘flying winemakers’, consultants represent
a vehicle of national and international transfer of both
tacitand codified knowledge,spreadingfrontier knowl-
edge on grape growing and wine making processes
across many different places.
8These data refer to the VI Region and not specifically to the
Colchagua Valley. Nonetheless, it is believed that this trend is repre-
sentative of this area.
One can distinguish between four types of firm in
the cluster:
(a) Firms that are vertically integrated at the local
level, producing bottled wines, usually for quality
markets. They undertake all phases of the produc-
tion chain, from the vineyard to the market, grow-
ing their own grapes, processing them and bottling
their own branded wines at the local level. These
can be either domestic or foreign-owned.
(b) Firms that are vertically disintegrated at the local
level: the local vineyard ‘subsidiaries’ of large na-
tional firms that own properties in different areas
of the country and perform the final steps of the
process (vinification, bottling, branding and mar-
keting) in their headquarters outside the cluster.
(c) Vertically integrated grape growers and producers
of bulk wine, usually at low quality.
(d) Non-integrated small-scale growers selling grapes
directly to one of the three groups above.
A firm’s commitment to producing higher quality
products reinforces its need to exercise control over
the process of grape-growing and viticulture, and this
has stimulated a trend towards the vertical integration
of wine producers via either of the first two firm cat-
egories noted above. Correspondingly, it has also pro-
gressively reduced the importance of subcontracting to
independent grape-growers.
4. Methodology
4.1. The sample and data collection
The study has been based on the collection of pri-
mary data at firm level. This was done via interviews
in a sample of firms in the cluster. As summarised
in Table 1 below, the sample was determined in the
following way. From the total population of wine
producers in the cluster (approximately 1009), we
first selected the total population of producers that
bottle wine and sell under their brand names—28
firms, including the subsidiaries of national wineries
that normally perform within this cluster only a part
9This is an estimate kindly provided by the Servicio Agricola y
Ganadero, Santa Cruz. The number includes all producers of wine,
including those that produce for personal consumption.

E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68 53
Table 1
The sample
Wine producers
Branded Not branded
Locally based, vertically
integrated firms Subsidiaries, vertically
disintegrated firms Vertically integrated, bulk suppliers
Population (N=100) 21 7 72
(Estimate including all other types of firm)
Sample (n=32) 21 7 4
Of which:
National 15 7 4
Foreign 6 0 0
of the value chain, typically grape-growing. Then,
following the guidance of an expert informant, we
selected four other firms that produce and sell bulk
wine, normally to the first two groups of firms. No
independent grape-growers were selected.
A number of pilot interviews in the cluster indicated
that technical people in the firms were usually the best
informantsabout the history and current characteristics
of the firms. More important, they were also key nodes
in the cognitive interconnections between firms.10 The
interviews, based on a structured questionnaire, were,
therefore, held with the chief oenologist or the cellar-
man of each of the sampled firms.
Apart from general background and contextual
information, the interviews sought information that
would permit the development of quantitative indica-
tors in three key areas: (a) the ‘absorptive capacity’ of
the firms, (b) their intra-cluster knowledge communi-
cation patterns, and (c) their acquisition of knowledge
from extra-cluster sources.
4.1.1. Firm-level absorptive capacity
In the literature this concept, a key element in the
analysis here, is described in terms of the knowledge
base of the firm. This is usually identified not only in
terms of human resources (skills, training, experience,
10 This role of the oenologist was consistent with the behaviour
stressed by Von Hippel (1987): “When required know-how is not
available in-house, an engineer typically cannot find what he needs
in publications either: much is very specialised and not published
anywhere. He must either develop it himself or learn what he needs
to know by talking to other specialists” (Von Hippel, 1987, p. 292).
etc.) but also in terms of in-house knowledge-creation
effort (usually R&D) as in Cohen and Levinthal
(1989, 1990). Correspondingly, the structured in-
terviews sought detailed information about (i) the
number of technically qualified personnel in the firm
and their level of education and training, (ii) the
experience of professional staff—in terms of time in
the industry and the number of other firms in which
they had been employed, and (iii) the intensity and
nature of the firms’ experimentation activities—an
appropriate proxy for knowledge creation efforts,
since information about expenditure on formal R&D
would have been both too narrowly defined and too
difficult to obtain systematically. This information
was transformed into a scalar value via Principal
Component Analysis as explained in Section 3.2.
4.1.2. Intra-cluster knowledge communication
patterns
In the questionnaire-based interview, these kinds of
relational data were collected through a ‘roster recall’
method: each firm was presented with a complete list
(roster) of the other firms in the cluster, and they were
asked the following questions:
Q1: Technical support received [inbound]
Ifyou are in a critical situationandneedtechnical ad-
vice, to which of the local firms mentioned in the
rosterdo youturn?[Please indicate theimportance
youattach to the informationobtainedineach case
by marking the identified firms on the following
scale: 0=none; 1=low; 2= medium; 3= high].

54 E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68
Q2: Transfer of technical knowledge (problem solving
and technical advice) [outbound]
Which of the following firms do you think have ben-
efited from technical support provided from this
firm? [Please indicate the importance you attach
tothe information provided to eachofthefirms ac-
cording to the following scale: 0= none; 1= low;
2=medium; 3=high].
These questions specifically address problem solv-
ingandtechnicalassistance because they involvesome
effort in producing improvements and change within
the economic activity of a firm. This is meant to go
beyond the mere transfer of information, whose access
canbeeasilyattainedthroughotherchannels(e.g.trade
fairs, the internet, specialised reviews, etc.). Instead,
our interest here is to investigate whether local stocks
of contextualised complex knowledge are not only
accessible but also eventually absorbed by localised
firms. So, for example, knowledge is transferred by
providing a suggestion on how to treat a new pest or
how to deal with high levels of wine acidity during
fermentation. Accordingly, the knowledge transferred
is normally the reply to a query on a complex
problem that has emerged and that the firm seeks to
solve.
The ways in which the responses to questions have
been operationalised into a set of relational variables
are indicated in Section 3.2.
4.1.3. The acquisition of knowledge from
extra-cluster sources
The interview also asked about the firms’ acqui-
sition of knowledge from sources outside the cluster,
both at national and international level. Specifically,
respondents were asked to name on a roster of possible
extra-cluster sources of knowledge (universities,
suppliers, consultants, business associations, etc.)
those which had contributed to the technical enhance-
ment of firms11. They were also asked to indicate
whether the firm had co-operated with any of those
sources for joint research and experimentation. More
specifically two different questions were formulated:
11 The list also contains open lines to permit the respondent to men-
tion extra-cluster sources which were not included in the pre-defined
roster.
Q3: Technical support received [inbound]
Question Q3: Could you mark, among the actors in-
cluded in the roster, those that have transferred
relevant technical knowledge to this firm? [Please
indicate the importance you attach to the infor-
mation obtained in each case by marking the
identified firms on the following scale: 0= none;
1=low; 2= medium; 3= high].
Q4: Joint experimentation
Question Q4: Could you mark, among the actors in-
cluded in the roster, those with whom this firm
has collaborated in research projects in the last 2
years? [Please indicate the importance you attach
to the information obtained in each case by mark-
ing the identified firms on the following scale:
0=none; 1=low; 2= medium; 3= high].
4.2. Operationalising key indicators
Testing our hypotheses required operationalisation
of the following firm-level concepts: absorptive ca-
pacity, external openness, and intra-cluster knowledge
linkages. It also required operational indicators of the
extent to which, and the most important ways in which,
the cluster knowledge system was structured into ‘cog-
nitive subgroups’. Table 2 summarises the basis of the
measures and indicators used. Further information is
provided in the Appendix.
5. Main empirical findings
5.1. Absorptive capacity and external openness
In general terms, the Colchagua Valley cluster can
be described as an ‘open’ knowledge system (Bell
and Albu, 1999) as many of its constituent firms have
established linkages with external sources of knowl-
edge. Firms tend to establish frequent knowledge link-
ages with many of the leading research and technology
transfer institutions and with universities (see Table 3).
Suppliers of materials and machinery, jointly with con-
sultants, are also important sources of knowledge and
seem to be the main drivers of technical change in
the firms. The cluster firms are also well connected
with international sources of knowledge—in particu-
lar with foreign consultant oenologists (‘flying wine-
makers’) that play an important role in the transfer of
frontier knowledge and techniques in the field. How-

E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68 55
Table 2
Key concepts and their measurement
Hypotheses and concepts Explanation/elaboration of concepts Measure adopted
Hypothesis 1
Association between firms’
(i) Absorptive capacity
and (i) Absorptive capacity has four components: (a) the
level of education of the technical personnel em-
ployed in the firm, (b) each professional’s months
of experience in the industry, (c) the number of
firms in which each professional has been previ-
ously employed, and (d) the type and intensity of
R&D undertaken by the firm
Absorptive capacity: an index derived from the ap-
plicationof Principal ComponentAnalysis to the data
about the four component indicators (see Appendix
Section A for more detail).
(ii) External openness (ii) External openness: reflects the firm’s propensity
to acquire extra-cluster knowledge External openness: the number of linkages with
extra-cluster sources of knowledge (see Appendix
Section C)
Hypothesis 2 Indicators of three key features of individual firms’
intra-cluster knowledge linkage are developed. Graph theoretical methods were adopted to measure
different dimensions of the ‘centrality’ of firms
communication patterns, and more generally their
cognitive positions in the cluster. For further details
see Appendix Section B.
Association between firms’
(i) Absorptive capacity (i) The propensity of a firm to be a local ‘source’ of
knowledge Out-degree centrality index: measures the extent
to which technical knowledge originates from a firm
to be used by other local firms. The indicator is
computed on two alternative bases
dichotomous:reflects the presence/absence of such
a linkage
valued: analyses the value given to each linkage by
the knowledge-user (a 0–3 range)
and
(ii) Intra-cluster knowledge
linkages (ii)The propensity ofa firm toabsorb knowledgefrom
intra-cluster sources In-degree centrality index: measures the extent to
whichtechnical knowledge isacquired by/transferred
to a firm from other local firms. Again the indicator is
computed on two alternative bases: dichotomous and
valued.
and
(iii) Different cognitive
positions in the cluster
knowledge system
(iii) A firm’s degree of interconnection with the
intra-cluster knowledge system Betweenness: measures the degree of cognitive
interconnectedness of a firm on the basis of its
propensity to be in-between of other firms’
knowledge linkages.
Indicators of the different roles of a firm in the local
knowledge system combine (i) and (ii) above in
order to assess the balance between a firm’s role
as source and absorber of knowledge flows
within the cluster.
In-degree/Out-degree centrality index (I/O C.I.):
measures the ratio between the knowledge received
and that transferred by each firm
If I/O C.I is >1: the firm is a net ‘absorber’ of
knowledge
If I/O C.I is <1: the firm is a net ‘source’ of
knowledge
If I/O C.I is about 1, the firm engages in the mutual
exchange of knowledge

56 E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68
Table 2 (Continued)
Hypotheses and concepts Explanation/elaboration of concepts Measure adopted
Hypotheses 3 In order to identify the extent to which knowledge
interactions in the cluster are structured within
subgroups of highly interconnected firms, three
graph-theoretic indicators are used. See, Appendix
Section B.II. for further detail.
Core and peripheral groups: core-periphery
analysis allows the identification of a cohesive
subgroup of core firms and a set of peripheral firms
that are loosely interconnected with the core.
The structuring of the cluster
knowledge system in
ways reflecting distinct
‘cognitive sub-groups’k–core (k= 4): a cohesive subgroup in which each
firm is connected to at least four other firms in the
subgroup
Source: Authors’ own specification of indicators, with sociometric indexes taken from Wasserman and Faust (1994) and Borgatti et al. (2002).
ever, as shown in Table 3, the degree of external
openness is not homogeneous across the cluster firms,
as some firms tend to establish more linkages than
others.
Given this heterogeneity, a non-parametric correla-
tion test was run between the level of firms’ absorptive
capacity and their degree of external openness (Hy-
pothesis 1). The test shows a significant correlation
between those two variables: the Kendall tau b cor-
relation coefficient is 0.45 with p<0.01. This result
confirms Hypothesis 1: firms with higher absorptive
capacity tend to interconnect more to external sources
of knowledge than other firms (see Table 4).
According to our results, the existing knowledge
base of the cluster firms appears to shape the hetero-
geneous propensity to interconnect with extra-cluster
sources of knowledge. This seems consistent with the
idea that firms with higher levels of absorptive capacity
Table 3
Linkages with extra-cluster sources of knowledge (number of firms with at least one knowledge link to the knowledge sources indicated)
Extra-cluster sources Among the firms with overall ‘openness’ to external sources that was
Above the average On the average Below the average
Research institutes 9 9 4
Ceviuc (University Catolica)
Knowledge transfer 8 6 3
Joint research projects 5 1 0
Centro Tec. Vid Y Vino (University Talca)
Knowledge transfer 7 8 3
Joint research projects 1 2 1
Fac. CC. Agronomicas (University Chile)
Knowledge transfer 6 2 0
Joint research projects 2 1 0
INIA
Knowledge transfer 2 0 1
Joint research projects 4 1 0
Business associations 8 7 2
Vinas de Chile
Knowledge transfer 6 3 0
Chilevid
Knowledge transfer 3 4 1
Corporacion Chilena de Vino
Knowledge transfer 4 4 1
Private firms: consultants and suppliers 10 10 6
Domestic knowledge transfer 10 10 6
Foreign knowledge transfer 9 4 4
Source: Authors’ own data.

E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68 57
Table 4
Correlationbetween externalopenness and absorptive capacity(non-
parametric correlation: Kendall tau b coefficient)
Average
absorptive
capacity
External openness (above the average) (n=10) 0.74
External openness (on the average) (n=10) 0.07
External openness (below the average) (n=12) −0.67
Kendall tau b correlation between external openness
and absorptive capacity 0.45**
Source: Authors’ own data.
∗∗ Coefficient is significant with p<0.01.
are cognitively closer to extra-cluster knowledge and
can, therefore, operate more easily than other firms as
interfacesor nodes of connection of the cluster with the
external environment.
5.2. Local inter-firm knowledge exchange: how
cognitive positions vary in the cluster
It is also evident that firms do not participate in
the local knowledge system in an even and homoge-
neous way. Visual inspection (see Fig. 1) suggests that
firms tend to interconnect differently to one another:
in particular one group of firms (centre of the figure)
are linked, transferring and receiving knowledge from
each other. In contrast, another group of firms (top left)
remain cognitively isolated.
In order to test whether firms that were more cog-
nitively interconnected in the cluster knowledge sys-
tem also had higher absorptive capacities (Hypothesis
2(a)), we ran a second correlation test. As indicated in
Table 5, this indicated statistically significant relation-
ships between the absorptive capacity and the different
centrality indexes.
The variation between the different correlation
statistics is also illuminating. We observe that among
these the highest correlations are between absorp-
tive capacity and Out-degree centrality, with both di-
chotomous and valued data (Kendall tau b = 0.523 and
0.532). This suggests that absorptive capacity influ-
ences the propensity of firms to transfer knowledge to
other local firms and hence to be net ‘sources’ of tech-
nical knowledge within the cluster system. For the In-
degreecentralityand betweenness indexes, the correla-
tions are weaker, but still significant. This suggests that
even at lower levels of absorptive capacity, firms might
be linked to the local knowledge system, provided that
aminimum absorptive capacity threshold is reached.
Fig. 1. The local knowledge system in the Colchagua Valley: a graphical representation source: UCINET 6 on author’s own data. Note: An
arrow from i to j indicates that i transfers knowledge to j. The diameter of the nodes is proportional to firms’ absorptive capacity.

58 E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68
Table 5
Correlation between absorptive capacity and centrality indexes
Indexes of different communication patterns
Out-degree C.
(dichotomous) Out-degree C.
(valued) In-degree C.
(dichotomous) In-degree C.
(valued) Betweenness
Absorptive capacity 0.523** 0.532** 0.399** 0.323*0.291*
Source: Authors’ own data (non-parametric correlation: Kendall tau b coefficient), see Appendix Section B for definitions of the indexes.
∗Correlation is significant with p<0.05.
∗∗ Correlation is significant with p<0.01.
This association between firms’ absorptive capac-
ities and their propensity to take different cognitive
positions in the cluster knowledge system (Hypothe-
sis 2(b)) is shown more directly in Table 6. This in-
troduces a set of classifications of the firms’ cognitive
positions in the cluster—knowledge transferers, mu-
tual exchangers, absorbers or isolates—as described in
the first column in the Table.
The Table shows clearly that average absorptive ca-
pacity varies considerably across the different cogni-
tive positions. Particularly interesting is the difference
observable between the first three groups: sources, mu-
tual exchangers and absorbers and the last one, which
is characterised by isolated firms. This result supports
the idea that a threshold for inter-firm knowledge ex-
change exists, so that when firms’ absorptive capacity
is very low, the cognitive distance with other firms’
knowledge bases becomes too high (i.e. infinite) and
the firms tend to be isolated. Correspondingly, those
firms that are sufficiently above the minimum thresh-
old have a higher probability of being interconnected
with other local firms. Given the way the questions
Table 6
Firms’ absorptive capacities and cluster cognitive positions
Cognitive positions in the cluster Absorptive
capacity
measure
Sources (n=5) 1.00
firms with an In/Out degree centrality ratio >1
Absorbers (n=5) 0.65
firms with an In/Out degree centrality ratio <1
Mutual exchangers (n=8) 0.07
firms with an In/Out degree centrality ratio= 1
Isolates (n=14) −0.88
firms with In and Out centralities
approximating to 0
Source: Authors’ own data.
were centred on problem-solving and performance im-
provement,these linkedfirms, in contrast totheisolated
firms, are likely to improve the quality of their produc-
tion by virtue of such linkages. On the basis of these
results we accept Hypothesis 2(b).
5.3. Structure of the intra-cluster knowledge
system
In order to analyse how knowledge flows were
structured within the cluster knowledge system, we
adoptedgraph theoretical measuresforidentifying cog-
nitive subgroups within the cluster, by which we mean
subgroups of firms that have established more re-
lations with members internal to the subgroup than
with non-members (Alba, 1973). In particular, we ap-
plied core/periphery models to our data (Borgatti and
Everett,1999).These allow the identification of central
andperipheralpoleswithinclusterknowledgesystems.
In the case of the Colchagua cluster we observe the
formation of a clear core-peripheral knowledge struc-
ture where (a) firms in the core tend to be highly inter-
connected among themselves, whereas (b) peripheral
firmstend to establishlooselinkages with the corefirms
and virtually no interconnections with other peripheral
firms. More specifically, we show in Table 7, the den-
sity of the four types of relations namely: core-to-core
(top left), core-to-periphery (top right), periphery to
core (bottom left) and periphery to periphery (bottom
right).12 Density is highest for core-to-core relations
(0.571), which means that core firms tend to transfer
knowledge more often within the core. As expected,
they are also identified as sources of knowledge by pe-
ripheral firms (core-to-periphery density is 0.155), but
thisrelation is muchlooserthanthe previousone.At the
12 For core/periphery analysis we adopted a directional dataset.

E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68 59
Table 7
Density of linkages within and between core and peripheral groups
The density of linkages
(knowledge transfer from
row to column)
Average
absorptive
capacity
Core Periphery
Core (nC=14) 0.571 0.155 0.58
Periphery (nP=18) 0.083 0.026 −0.45
Source: UCINET 6 applied to author’s own data. Note: The density
of a network is the total number of ties divided by the total number
of possible ties.
sametimecorefirmsdonotmentionperipheralfirmsas
sources of technical advice (periphery-to-core density
is very low 0.083) and even less do peripheral firms
transfer or receive knowledge from other peripherals
(periphery-to- periphery density is 0.026).
The association between these core/periphery po-
sitions and the absorptive capacity of the constituent
firms is interesting. Core firms have an average absorp-
tivecapacity of 0.58,while the same data for peripheral
firms is 0.45. This seems consistent with the idea that
core firms, having higher absorptive capacities, boost
local processes of incremental learning and stimulate
some peripheral firms to ask for technical advice, al-
though these relations are not nearly as intense as those
within the core group itself.
To improve understanding of this core sub-group,
we undertook a further step in the analysis, the identifi-
cationof 4-cores within the cluster.This identifies firms
that have established at least four knowledge linkages
with other firms of the sub-group. This is carried out by
taking into account the whole set of knowledge links,
independentlyoftheirestimatedimportance.Asshown
in Fig. 2, we find a complete network of interrelated
firms, which correspond to the core identified above.
These results support Hypothesis 3. In particular,
the local knowledge system has the structural charac-
teristics of a core/periphery set where knowledge in-
teractions are clearly concentrated within a subset of
core firms. Furthermore, consistent with previous sec-
tions, this core group is formed by firms that have, on
average, higher absorptive capacities than the firms in
the periphery. The data are consistent with the exis-
tence of a single community of fairly well connected,
skilled knowledge workers who tend to exchange more
knowledge within the community (i.e. within the core)
than outside it. In contrast, with their relatively weak
knowledgebase,peripheralfirms arenotpart of thecore
knowledge community in the cluster. In other words,
weobservethatfirms’ absorptivecapacities andthepar-
ticipation of their professional personnel in knowledge
communities are interwoven elements which shape the
structure of the local knowledge system.
5.4. Linking intra- and extra-cluster knowledge
systems: the role of technological gatekeepers
In this section, we bring together the data about (a)
the external openness of firms and (b) the ‘cognitive
position’ of firms within the local knowledge system.
Combining these parameters we identified five main
learning patterns within the cluster, corresponding to
thefollowingfivetypes of ‘cognitiverole’, as indicated
Fig. 2. The core group: an analysis of 4-cores Source: UCINET 6 on author’s own data. Note: The linkages are undirected as we adopted a
symmetrised version of the original dataset. The diameter of the nodes is proportional to firms’ absorptive capacity.

60 E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68
Table 8
Differing firm-centred cognitive roles in cluster learning
Source: Authors’ own data.
in the cells in Table 8 that are highlighted inside the
heavy broken lines. More detailed characteristics of
the firms playing these five types of role are indicated
in Table 8.
(a) Technological gatekeepers (TGs): firms that have
a central position in the network in terms of
knowledge transfer to other local firms and that
are also strongly connected with external sources
of knowledge.
(b) Active mutual exchangers (AMEs): firms that form
a central part of the local knowledge system with
balanced source/absorber positions within the
cluster. They also have relatively strong external
links. Although they are less strongly connected
to external sources than the TG firms, they behave
in a similar way to ‘technological gatekeepers’
by bridging between external sources and local
absorbers of knowledge.
(c) Weak mutual exchangers (WMEs) consist of firms
that are similar to AMEs in that they are well
linked to external knowledge sources and play a
relativelybalanced source and absorber role within
the cluster. However, compared with AMEs, they
are less well connected to other firms in the cluster.
(d) External Stars (ES):13 firms that have established
strong linkages with external sources, but have
limited links with the intra-cluster knowledge sys-
tem. These weak intra-cluster links are primarily
inward and absorption–centred.
(e) Isolated firms (IFs): are poorly linked at both the
local and extra-cluster levels.
Tables 8 and 9 indicate that firms playing three of
the five cognitive roles (TGs, AMEs and WMEs) are
connected actively into the cluster knowledge system
andcontribute positivelyto its learning processes. They
constitute the core of the absorptive capacity of the
cluster. The other two groups’ cognitive links with the
cluster system are much more marginal.
All the firms contributing to the cluster’s absorp-
tive capacity engage in a combination of three key
activities: acquiring knowledge from outside the
cluster, generating new knowledge through their own
intra-cluster experimentation, and contributing to
intra-cluster diffusion. The strength of this combina-
tion of positive roles differs between the groups and is
13 This term is taken from Allen (1977) where it describes individ-
uals having similar positions within firms i.e. with strong links to
external sources of knowledge plus weak links to the internal knowl-
edge system.

E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68 61
Table 9
Heterogeneity in the cluster knowledge system
Characteristics of firms’ differing cognitive roles in the cluster knowledge system
ES (n=3) TG(n=5) AME (n= 4) WME (n=3) IF(n=10) Others (n= 7) Total (n=32)
1. Absorptive capacity
Composite index 1.71.00 0.16 0.04 −0.70 – 0
Within which
Intensity of experimentation 3.03.01.75 1.66 0.7– 1.59
2. Openness
External openness 10.33 11.08.58.93.1– 7.12
3. Intra-cluster cognitive positions
Average Out-degree centrality
Dicot 3.07.26.03.00.36 – 2.90
Valued 5.312.811.57.30.73 – 5.37
Average In-degree centrality
Dicot 4.34.05.03.00.5– 2.90
Valued 8.36.610.25 7.61.1– 5.37
Ratio In/Out degree centrality
Dicot 1.44 0.58 0.81.01.5– 0.85
Valued 1.56 0.51 0.91.01.6– 0.86
Betweenness 20.043.226.75 8.00.2– 13.90
Source: Authors’ own data.
most evident in the case of the TGs and AMEs. They
have relatively high Out-degree centrality indexes, and
this is particularly so with respect to the valued data
indexes, reflecting the qualitative importance attached
to the knowledge they transfer to other firms. There are
nevertheless substantial differences between these two
groups. The TG firms play a striking role as net sources
ofknowledge fortheclustersystem (with In/Out degree
indexes of only about 0.5), while the AMEs act only
marginally as net sources (with In/Out degree indexes
around 0.8–0.9). Behind this difference lie differing
firm-level capacities. Compared to the TGs, the AMEs
undertake more modest local experimentation and
have more limited intra-firm knowledge resources. In
contrast,theTGs are not just well connected to external
knowledge sources. They are also significant creators
of knowledge in their own right, demonstrating a high
intensity of local experimentation; and this is backed
by relatively strong intra-firm knowledge resources.
These are typically ‘advanced’ firms that operate very
close to the technological frontier14 and whose pro-
14 The wine industry has gone through substantial scientific and
technological changes over recent decades. Both old world produc-
ers (e.g. France) and new world producers (e.g. California, Australia,
South Africa, etc.) have contributed to define a widely accepted fron-
tier of technology in this industry (see Paul, 2002).
duction is oriented towards the exportation of premium
wines.
These TG firms, therefore, tend to be local deposi-
tories of technical novelties, which they apply and con-
textualise in their economic practice. Their technolog-
ically advanced position is normally acknowledged by
the rest of the firms in the cluster15 and this spurs the
latter to ask for advice. This explains their asymmet-
ric position as substantial net sources of knowledge in
the cluster. It is pertinent to note that these TG firms
in the majority of the cases have vertically integrated
operations located within the cluster and are, therefore,
well embedded in the local area.16 Their willingness to
engage in unreciprocated knowledge transfer to other
local firms, may reflect the positive externalities asso-
ciated with this. In a wine area, such as Colchagua,
which is currently investing in achieving international
acknowledgement for the production of high quality
wines, the improvement of every producer in the area
is likely to generate positive marketing-related exter-
15 A question specifically addressed this issue. Respondents were
asked to provide three names of firms that they considered advanced
in the cluster, with respect to their degree of technical modernisation
and the quality of wine produced.
16 This does not mean that such firms are all locally-owned. Indeed
in three of the five cases the ownership is wholly or partially foreign.

62 E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68
nalities for the whole area, and these may outweigh the
possible cost to these firms associated with imbalanced
knowledgetransfer relationships withcompetingfirms.
Among the relatively active participants in the clus-
ter knowledge system, the WME group has some sim-
ilar attributes to the AME firms (moderate levels of
openness and experimentation). However, with a very
low level of absorptive capacity, the WME firms have
much more modest knowledge resources to draw on. It
isperhapsnot surprising, therefore, that they seem to be
soughtout only infrequentlytocontributeknowledgeto
other firms, as reflected in the low Out-degree central-
ityindexes(3.0and 7.3 for the dichotomous and valued
data respectively). At the same time they have, simi-
larly, low levels of In-degree centrality. Consequently,
although they have a balanced exchange of knowledge
with other firms (an In/Out degree ratio of 1.0), those
interactions take place at a much lower level than in the
case of the AME firms.
The two marginal groups of firms differ widely. The
IFs have extremely low scores on almost all the indi-
cators. They have very low absorptive capacity; they
undertake almost no experimentation; and they acquire
almost no knowledge from extra-cluster sources. Not
surprisingly, they are rarely sought out as knowledge
sources by other firms. But it is striking that they also
rarely seek out knowledge from other cluster firms
(demonstrating by far the lowest In-degree centrality
indexes in the sample: 0.5 and 1.1). These firms, ac-
counting for nearly one-third of the sample, are barely
connected to the cluster knowledge system at all.17
The External Stars are marginally connected to the
knowledge system in a different way. They have by
far the highest index of firm-level absorptive capacity
amongall the firms in the cluster(1.7),andthis includes
a high intensity of experimentation carried out in the
cluster.This suggests they facealowcognitive distance
from extra-cluster sources of knowledge, enabling
them to draw heavily on those sources for their own
learning and innovation (as reflected in the high open-
nessindexof10.3). At the same time, though, theyplay
a highly imbalanced cognitive role inside the cluster.
On the one hand, they pass on very little of their knowl-
17 In addition, there are four other isolated firms, not included in
any of the selected groups of firms. These have, similarly, low levels
of connection to other firms in the cluster, though stronger links with
extra cluster sources of knowledge.
edge to other cluster firms (reflected in very low Out-
degree indexes: 3.0 and 5.3). On the other, as reflected
in their high In-degree centrality values, they seek out
advicefrom the ‘advanced’ firms inside the cluster,par-
ticularly the technological gatekeepers, although they
tend not to reciprocate the transfer of knowledge.
Inotherwords,althoughtheseexternalstarfirmsare
perhaps best positioned of all the cluster firms to make
positive contributions to the cluster knowledge system,
theyrarely do so. Totheextentthat they engage withthe
intra-cluster knowledge system, this is primarily about
extracting and absorbing cluster knowledge, not about
contributing to it. This pattern appears not to reflect
the ownership status of the firm,18 but may reflect the
production structure of the firms. They are mainly ver-
tically disintegrated subsidiaries of large national wine
producers that base their main operations elsewhere,
and are, therefore, not well embedded in the local area.
This might partially explain why their behaviour con-
trasts so sharply with that of the other advanced firms
that do operate as local technological gatekeepers.
Insummary,then, because of thelimited role played
by these two groups of firms, the overall technological
dynamism of the cluster as a whole seems to be
driven by less than half of the sample firms—the
TGs, AMEs and, to a lesser extent, the WMEs (12 out
of 32). Although this is a relatively dynamic cluster
that is moving ‘upwards’ in international markets, its
technological dynamism is not driven by a widespread
community of technologically dynamic firms oper-
ating, similarly, and pervasively across the cluster.
Instead, it is driven by a relatively small group of firms
that is organised within a core interacting knowledge
community, surrounded by greater number of largely
passive firms that occasionally absorb elements of
knowledge from the core group or, in a small number
of cases, directly from external sources. Moreover, that
18 Thesmallnumber ofobservationsprecludes meaningful analysis.
However, it may be pertinent to note that, while two of the three stars
were subsidiaries of large national (not foreign) firms, the remaining
five of the seven nationally-owned subsidiaries were as follows: one
TG, one WME, one IF and two in the ‘Other’ categories. Similarly,
clear conclusions cannot be drawn about the six foreign firms in
the cluster. Half of them were technological gatekeepers and the
others fell into more isolated categories. Ongoing research by the
first author in other wine clusters suggests that the cognitive roles of
foreign companies depends more on the duration of the location in
the cluster than on their foreignness.

E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68 63
Fig. 3. Approximate location of wine cluster firms with differing cognitive roles, Colchagua Valley, Chile.
core group is itself organised in a hierarchical knowl-
edge structure that has a few very capable, knowledge-
creating, and external knowledge-acquiring firms at
the ‘top’, acting as the main sources of new knowledge
for the cluster. Below that, firms with progressively
lower levels of these qualities shift from being net
contributors into the cluster knowledge system to
being net absorbers of that circulating knowledge.
It is interesting to note also that this hierarchy does
not seem to be influenced in any way by geography.
Fig. 3 indicates the physical location of firms in the
cluster, distinguished by their cognitive roles. All the
firms are distributed along the valley running from San
Fernandoin theEastto MarchihueandLolol in theWest
and their spatial propinquity and cognitive roles seem
unrelated. Indeed, in the subcluster of firms around
Perallillotechnological gatekeepers are closely located
with all the other types of firms. Similarly, in the sub-
clusters around Santa Cruz and Nancagua, the closest
neighbours to the technological gatekeepers are cogni-
tively isolated at the local level.
6. Conclusions
The results of the analysis in this paper call
into question the extent to which clustering per se
influences the learning behaviour of cluster firms.
In Colchagua at least, the spatially clustered wine
producers demonstrated a wide range of different

64 E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68
communication and learning behaviours. For some,
learning links with other organisations ran strongly
outside the valley, almost exclusively so in some
cases; and a substantial number of other firms were
almost totally isolated from any learning processes
at all whether within the cluster or outside. Among
the firms that did demonstrate cluster-centred learning
relationships, the cognitive positions and roles varied
widely. This heterogeneity, associated with the dif-
ferences in firms’ absorptive capacities, suggests that
a cluster is a complex economic and cognitive space
where firms establish knowledge linkages not simply
because of their spatial proximity but in ways that are
shaped by their own particular knowledge bases.
Consequently, the results shed light on the relation-
shipbetween meso andmicrowithinthe cluster.Instead
of the common emphasis given to ways in which the
meso-level characteristics of clusters influence micro
behaviour, this study highlights the importance of the
opposite direction of influence. It was the capacities of
individual firms to absorb, diffuse and creatively ex-
ploit knowledge that shaped the learning dynamics of
the cluster as a whole.
This direction of relationship was exemplified
particularly clearly in the case of one meso charac-
teristic that has been suggested as important for the
longer run growth of a cluster: its openness to external
knowledge, or more specifically its capacity to acquire
external knowledge and absorb it into its production
activities. In the Colchagua cluster, this meso-level
absorptive capacity was determined by the knowledge
bases of the firms. This was not simply a matter of the
cluster capacity being an aggregation of the individual
firms’ capacities, since the channels of knowledge ac-
quisition and diffusion between the firms were also key
components of the overall cluster absorptive capacity.
However, the density and structure of those channels
into and within the cluster, and hence their impact
on the extent of learning in the cluster, were strongly
shaped by the knowledge bases of the individual firms.
These conclusions align with the distinction be-
tween relational and spatial proximity highlighted by
Amin and Cohendet (2004). However, in several re-
spectstheydiffer from other recent contributions in this
field. In particular, although spatially bounded knowl-
edge interactions within the cluster were important in
thelearning process, instead of being“unstructuredand
unplanned”, arising “by chance” (Malmberg, 2003, pp.
157–158), they were highly structured. Also, although,
like Owen-Smith and Powell (2004), we have high-
lighted the importance of networks nodes in shaping
the structure of interactions, we have shown the signif-
icance of their cognitive rather than organisational and
institutional characteristics. This leads into, but leaves
open,questionsabout why firms with these characteris-
ticsbehavedin the ways we describe. Inparticular,why
didfirms with very similar cognitive characteristics be-
have as differently as the Technological Gatekeepers
and the External Stars? Also, what factors drive the in-
teractions among the members of the knowledge com-
munity? These questions are not pursued here but are
being examined in further research by the first author.
Our conclusions also prompt speculative questions
about the long term evolution of cluster cognitive sys-
tems of this type. Our cross sectional data do not throw
light directly on that dynamic, but the indirect illumi-
nation they provide does prompt reflection about the
circumstances that might underpin it. Recall key fea-
tures of the current situation: one group of firms with
the strongest knowledge bases, the most intensive in-
house experimentation and the strongest links to exter-
nal knowledge sources were the External Stars which
contributed very little to the intra-cluster learning sys-
tem. At the same time, a relatively small number of
other firms played strong positive roles in acquiring
or developing new knowledge and diffusing it more
widely in the cluster; and finally nearly one-third of
the sample firms were disconnected from the system.
From that base, several directions of evolution can be
envisaged. Two seem particularly interesting.
One would be towards a much more pervasive
and less polarised knowledge and learning system. A
greater number of firms would act as net contributors
into the knowledge system (in particular, with External
Stars and Active Mutual Exchangers behaving more
like Technological Gatekeepers), and the internally
isolated firms would either connect into the system
as knowledge acquirers or exit the industry. The other
direction would be towards a system in which extra-
cluster sources of knowledge became the dominant
drivers of learning and innovation in an increasingly
competitive market, with firms among the Techno-
logical Gatekeepers and Active Mutual Exchangers
reducing their willingness to act as net ‘sources’ of
knowledge (i.e. behaving more like the External Stars).
The cluster knowledge system would then turn ‘inside-

E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68 65
out’, exhibiting a different kind of polarised structure.
The most technologically advanced and dynamic firms
would concentrate their knowledge links with actors
outside the spatial cluster, contributing little or nothing
to the intra-cluster system. Increasingly isolated
other cluster firms, without their links to sources of
new knowledge, would fail to compete in growing
markets.
Both these directions of change would be likely
to enhance the overall growth and competitive
performance of wine production in the Colchagua
Valley. But they would result in two very different
meanings of ‘cluster’. One would conform to the
conventional expectations of cluster analysts: a coher-
ent cluster-centred knowledge system would act as a
positive influence on the innovative activities of the
spatially associated firms. The other would virtually
eliminate the role of spatial clustering as an influence
on the learning and innovative activities of firms that
would remain geographically, but not cognitively,
associated.
In that context, if the patterns reported here are
widespread, interesting questions arise about policy. It
seems fairly clear that, since learning and innovation
are driven primarily by the knowledge bases (absorp-
tive capacities) of individual firms, measures designed
merely to foster spatial agglomeration may have lim-
ited influence—a view consistent with that of Breschi
and Malerba (2001). Similarly, measures designed to
foster intra-cluster communication and collaboration
might not do much to change firms’ cognitive roles
if those also are shaped primarily by their knowledge
bases as well as strong underlying motivations. In
contrast, measures focused on strengthening firms’
knowledge bases might be expected to lead to stronger
extra-cluster links, greater new knowledge creation
and more intensive intra-cluster diffusion.
Acknowledgements
The authors are grateful to Christian Diaz Bravo
(S.A.G., Chile) and Marcelo Lorca Navarro (I.N.I.A.,
Chile) for their invaluable help during the fieldwork.
Thanks go also to Mario Cimoli, Jorge Katz and Gio-
vanni Stumpo for their support during the first author’s
stay at CEPAL and to Simona Iammarino, Aldo Ge-
una and three anonymous referees for comments on
an earlier draft. Financial support provided for the first
author’s doctoral research by the EU Marie Curie Pro-
gramme, the UK Economic Social Research Council
and the Italian Ministry for Education, University and
Research is gratefully acknowledged.
Appendix A. Absorptive capacity
Absorptive capacity has been measured by applying
a Principal Component Analysis to the following four
correlated variables:
A.1. Variable 1: Human resources
This variable represents the cognitive background
of each firms’ knowledge skilled workers on the of
their degree of education. According to previous stud-
ies regarding returns to education, we assume that
the higher the degree of education the higher is their
contribution to the economic returns of the firm. On
this assumption we weight each knowledge skilled
worker differently according to the degree attained so
that:
Humanresource =0.8×degree +0.05 ×master
+0.15 ×doctorate
Aweight of 0.8 has been applied to the number of grad-
uateemployees in the firm whichincludealsothose that
received higher levels of specialisation. In such cases
the value adds up a further 0.05 times the number of
employees with masters and 0.15 for those that have a
Ph.D.
Only degrees and higher levels of specialisation in
technical and scientific fields related to the activity of
wine production (i.e. agronomics, chemistry, etc.) are
taken into account.
A.2. Variable 2: Months of experience in the wine
sector
This variable has been included as it represents the
cognitive background of each of the abovementioned
resources in temporal terms. Time is in fact at least
indicative of the fact that accumulation of knowledge
has occurred via ‘learning by doing’ (Arrow, 1962).
More in detail, the variable is the result of a weighted

66 E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68
mean of the months of work of each knowledge skilled
worker in the country and abroad:
Months of experience in the sector= 0.4 ×no.
of months (national)+0.6×no. of months (interna-
tional).
To the time spent professionally abroad we at-
tributed a higher weight because the diversity of the
professional environment might stimulate an active
learning behaviour and a steeper learning curve. The
learning experiences considered are those realised in
the wine industry only.
A.3. Variable 3: Number of firms in which each
knowledge skilled worker has been employed
Thisvariable includestheprofessional experience in
other firms operating in the wine industry. Also in this
case we weighted differently national and international
experiences, giving to the latter a higher weight.
Number of Firms=0.4×no. of firms (na-
tional)+0.6×no. of firms (international)
A.4. Variable 4: Experimentation
Inthis case, the levelofexperimentationat firm level
has been calculated according to the following scale:
(0) for no experimentation;
(1) when some form of experimentation is normally
carried out but only in one of the activities of the
productive chain (either in viticulture or vinifica-
tion);
(2) when is led in at least two activities of the produc-
tive chain (normally in both viticulture and vinifi-
cation);
(3) when at least two activities of the productive chain
are marked and the firm has been engaged in one
jointresearch project withauniversityoraresearch
lab in the last 2 years;
(4) when at least two activities of the productive chain
are marked and the firm has been engaged in more
than one joint research project with a university or
a research lab in the last 2 years.
Principal Component Analysis extracted one com-
ponent, which we adopted as a measure of absorptive
capacity.
Appendix B. Sociometric measures
B.I. Degree centrality depends on the links that one
node has with the other nodes of the network. It is a
simple measure because it counts the direct ties with
other nodes. It can be calculated both for undirected
and directed graphs. In this study, we computed both
In-degree and Out-degree centrality. In-degree counts
the number of ties incident to the node; Out-degree
centrality the number of ties incident from the node.
CD(ni)=d(ni)
where d(ni) is the sum of the nodes adjacent to that
node.
B.II. Actor betweenness centrality is a measure
of centrality that considers the position of nodes in-
between the geodesic (i.e. shortest path) that link any
other node of the network.
Let gjk be the proportion of all geodesics linking
node j and node k which pass through node i, the be-
tweenness of node i is the sum of all gjk where i, j and
k are distinct.
CB(ni)=j<kgjk (ni)
gjk
This index has a minimum of zero when nifalls on no
geodesics and a maximum which is (g−1) (g−2)/2
(g= total nodes in the network) which is the number of
pair of nodes not including ni.
B.III. Core/periphery models are based on the no-
tion of a two-class partition of nodes, namely, a cohe-
sive subgraph (the core) in which nodes are connected
to each other in some maximal sense and a class of
nodes which are more loosely connected to the cohe-
sive subgroup but lack any maximal cohesion with the
core. The analysis sets the density of the core to pe-
riphery ties in an ideal structure matrix. The density
represents the number of ties within the group on total
ties possible (Borgatti and Everett, 1999).
B.IV. k-core is a subgraph in which each node is
adjacent to at least a minimum number k of the other
nodes in the subgraph.
Appendix C. External openness
External openness has been measured considering
the knowledge linkages of firms with extra-cluster

E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68 67
sourcesofknowledge.In the analysis, we have grouped
the linkages into 10 sources and channels of extra-
cluster knowledge. The importance of each source for
thetransfer of technical knowledgeintothefirmis mea-
suredon a 0–3scale,where 0 standsfor‘noimportance’
and 3 for ‘maximum importance’.
References
Alba,R.D., 1973. Agraph theoretic definitionof asociometric clique.
Journal of Mathematical Sociology 3, 113–126.
Allen, T.J., 1977. Managing the Flow of Technology: Technology
Transfer and the Dissemination of Technological Information
within the R&D Organization. MIT Press, Cambridge, MA.
Altenburg, T., Meyer-Stamer, J., 1999. How to promote clusters:
policy experiences from Latin America. World Development 27
(9), 1693–1713.
Amin, A., Cohendet, P., 2004. Architectures of Knowledge. Firms,
Capabilitiesand Communities. Oxford University Press, Oxford.
Anderson, K., Norman, D., 2001a. Global wine production, con-
sumptionand trade: 1961–1999. A statisticalcompendium. CIES
Wine Statistical Compendium, Centre for International Eco-
nomic Studies, University of Adelaide, Australia.
Anderson, K., Norman, D., 2001b. Globalisation and the worlds’
wine markets. CIES Discussion Paper 0143, Centre for Interna-
tional Economic Studies, University of Adelaide, Australia.
Arrow, K.J., 1962. The economic implications of learning by doing.
Review of Economic Studies 29, 155–173.
Asheim, B.T., 1994. Industrial districts, inter-firm cooperation and
endogenous technological development: the experience of de-
veloped countries. In: UNCTAD, Technological Dynamism in
Industrial Districts: An Alternative Approach to Industrializa-
tion in Developing Countries?, United Nations, New Yorkand
Geneva, pp. 91–142.
Audretsch,D., Feldman, M.P., 1996. R&D spilloversand the geogra-
phy of innovation and production. American Economic Review
86 (3), 630–640.
Baptista, R., 2000. Do innovation diffuse faster within geographi-
cal clusters? International Journal of Industrial Organization 18,
515–535.
Becattini, G., 1989. Sectors and/or districts: some remarks on the
conceptual foundations of industrial economics. In: Goodman,
E., Bamford, J. (Eds.), Small Firms and Industrial Districts in
Italy. Routledge, pp. 123–135.
Becattini, G., Rullani, E., 1993. Sistema locale e mercato globale.
Economia e Politica Industriale 80, 25–49.
Bell, M., Albu, M., 1999. Knowledge systems and technological
dynamism in industrial clusters in developing countries. World
Development 27 (9), 1715–1734.
Belussi, F., Gottardi, G., 2000. Models of localised technological
change. In: Belussi, F., Gottardi, G. (Eds.), Evolutionary Patterns
of Local Industrial Systems. Towards a Cognitive Approach to
the Industrial District. Ashgate Publishing Ltd., Aldershot.
Borgatti, S.P., Everett, M.G., 1999. Models of core/periphery struc-
tures. Social Networks 21, 375–395.
Borgatti, S.P., Everett, M.G., Freeman, L.C., 2002. Ucinet 6 for Win-
dows. Analytic Technologies, Harvard.
Breschi,S., Lissoni, F.,2001.Knowledgespillovers and local innova-
tion systems: a critical survey. Industrial and Corporate Change
10 (4), 975–1005.
Breschi, S., Malerba, F., 2001. The geography of innovation and eco-
nomic clustering: some introductory notes. Industrial and Cor-
porate Change 10 (4), 817–833.
Brown, J.S., Duguid, P., 1991. Organizational learning and
communities-of-practice: Toward a unified view of working,
learning and innovation. Organization Science 2 (1), 40–
57.
Camagni, R., 1991. Local milieu, uncertainty and innovation net-
works: towards a new dynamic theory of economic space. In:
Camagni, R. (Ed.), Innovation Networks: Spatial Perspectives.
Belhaven Press, London, pp. 121–142.
Capello, R., 1999. Spatial transfer of knowledge in high technology
milieux: learning versus collective learning processes. Regional
Studies 33 (4), 353–365.
Cantwell, J., Iammarino, S., 2003. Multinational Corporations and
European Regional Systems of Innovation. Routledge, London,
New York.
Carter, A.P., 1989. Know how trading as economic exchange. Re-
search Policy 18, 155–163.
Cassiolato, J., Lastres, H., Maciel, M., 2003. Systems of Innovation
and Development: Evidence from Brazil. Edward Elgar, Chel-
tenham.
Cohen, W.M., Levinthal, D.A., 1989. Innovation and learning: the
two faces of R&D. The Economic Journal 99, 569–596.
Cohen, W.M., Levinthal, D.A., 1990. Absorptive capacity: a new
perspective on learning and innovation. Administrative Science
Quarterly 35, 128–153.
Coleman,J., 1990. Foundationsof SocialTheory.HarvardUniversity
Press, Cambridge, MA.
Creplet, F., Dupouet, O., Kern, F., Mehmanpazir, B., Munier, F.,
2001. Consultants and experts in management consulting firms.
Research Policy 30 (9), 1517–1535.
Crowley, W.K., 2000. Chile’s wine industry: historical character and
changing geography. In: Yearbook, Conference of Latin Ameri-
canist Geographers, pp. 87–101.
Del Pozo, J., 1998. Historia del Vino Cileno. Desde 1850 hasta hoy.
Editorial Universitaria, Santiago de Chile.
Dicken, P., Malmberg, A., 2001. Firms in territories: a relational
perspective. Economic Geography 77 (4), 345–363.
Dosi, G., 1997. Opportunities, incentives and the collective patterns
of technical change. The Economic Journal 107, 1530–1547.
Drucker, P., 1993. Post Capitalist Society. Butterworth, Oxford.
Gambardella, A., 1993. Innovazioni tecnologiche e accumulazione
delle conoscenze: quale modello per le piccole e medie imprese
negli anni ’90? Piccola Impresa/Small Business 2, 73–89.
Giuliani, E., 2002. Cluster absorptive capability: an evolutionary ap-
proach for industrial clusters in developing countries. Paper Pre-
sentedat the DRUID Summer Conference, Copenhagen/Elsinore
6–8 June.
Grabher, G., 1993. The weakness of strong ties: the lock-in of re-
gional development in the Ruhr area. In: Grabher, G. (Ed.), The
Embedded Firm. Routledge, London, pp. 1–32.

68 E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68
Guerrieri, P., Iammarino, S., Pietrobelli, C., 2001. The Global Chal-
lenge to Industrial Districts. Edward Elgar, Cheltenham.
Haas, P.M., 1992. Introduction: epistemic communities and inter-
national policy coordination. International Organization 46 (1),
1–35.
Humphrey, J., Schmitz, H., 1996. The triple C approach to
local industrial policy. World Development 24 (12), 1859–
1877.
Lane, P.J., Lubatkin, M., 1998. Relative absorptive capacity and
interorganizational learning. Strategic Management Journal 19,
461–477.
Lissoni, F., 2001. Knowledge codification and the geography of in-
novation:the case of Brescia mechanical cluster.Research Policy
30, 1479–1500.
Malmberg, A., 2003. Beyond the cluster—local milieus and global
economic connections. In: Peck, J., Yeung, H.W.C. (Eds.),
Remaking the Global Economy. Sage, London, pp. 145–
159.
Malmberg, A., Maskell, P., 2002. The elusive concept of localization
economies: towards a knowledge-based theory of spatial cluster-
ing. Environment and Planning A 34 (3), 429–449.
Markusen, A., 1996. Sticky places in slippery space: a typology of
industrial districts. Economic Geography 72 (2), 294–314.
Marshall, A., 1919. Industry and Trade. Macmillan, London.
Marshall, A., 1920. Principles of Economics. Macmillan, London.
Maskell, P., Malmberg, A., 1999. Localised learning and indus-
trial competitiveness. Cambridge Journal of Economics 23 (2),
167–186.
McPherson, M., Smith-Lovin, L., Cook, J.M., 2001. Birds of a
Feather: Homophily in Social Networks. Annual Review of So-
ciology 27, 415–444.
Nadvi, K., Schmitz, H., 1999. Clustering and industrialization: in-
troduction. World Development 27 (9), 1503–1514.
Owen-Smith, J., Powell, W.W., 2004. Knowledge networks as chan-
nelsand conduits:The effectsof spilloversin the Boston Biotech-
nology Community. Organization Science 15 (1), 5–21.
Paul, H.W., 2002. Science, Vine and Wine in Modern France. Cam-
bridge University Press, Cambridge, MA.
Porter, M., 1998. On Competition. Harvard Business School Press,
Boston, MA.
Rabellotti, R., 1999. Recovery of a Mexican cluster: devaluation
bonanza or collective efficiency? World Development 27 (9),
1571–1585.
Rogers, E.M., 1983. Diffusion of Innovations, third ed. The Free
Press, New York.
S.A.G., 2000. Catastro Viticola Nacional. Servicio Agricola y
Ganadero. Departamento de Proteccion Agricola, Santiago de
Chile.
Saxenian,A.L., 1994.Regional Advantage.Culture andCompetition
in Silicon Valley and Route, vol. 128. Harvard University Press,
Cambridge, MA.
Schachner, M., 2002. The Next Napa? Chile’s Colchagua Valley is
Aiming for Big Time. Wine Enthusiast, March, Elmsford, New
York.
Schrader, S., 1991. Informal technology transfer between firms:
cooperation through informal trading. Research Policy 10,
153–171.
Tapia, P., 2001. Los Vinos de Colchagua. PMC Pinnacle Worldwide
y Ed. Planetavino, Santiago de Chile.
Von Hippel, E., 1987. Cooperation between rivals: informal know-
how trading. Research Policy 16, 291–302.
Wasserman, S., Faust, K., 1994. Social Network Analysis. Methods
and Applications. Cambridge University Press, Cambridge, MA.
Wenger, E.C., Snyder, W.M., 2000. Communities of prac-
tice: the organizational frontier. Harvard Business Review,
January–February, 139–145.