Contents lists available at SciVerse ScienceDirect
Research Policy
j o ur nal homep age: www.elsev ier .com/ locate / respol
etwork dynamics in regional clusters: Evidence from Chile
lisa Giuliani ∗
ipartimento Economia & Management, University of Pisa, Via Ridolfi 10, 56124 Pisa, Italy
a r t i c l e i n f o
rticle history:eceived 2 September 2011eceived in revised form 27 March 2013ccepted 16 April 2013vailable online 29 May 2013
a b s t r a c t
A wealth of empirical literature shows that one of elements of success for regional clusters is that theyfacilitate the formation of local inter-organizational networks, which act as conduits of knowledge andinnovation. While several studies analyse the benefits and characteristics of regional cluster networks,very little is known about how such networks evolve over time and the extent to which their dynamicscan affect development processes. Using longitudinal data on a wine cluster in Chile and Stochastic
eywords:egional clustersnowledge networksetwork dynamicsine industry
hile
Actor-Oriented Models (SAOM) to measure network dynamics, this paper examines the microdynamicsunderpinning the formation of new knowledge ties among wineries. It finds that the coexistence ofcohesion effects (reciprocity and transitivity) with the weak knowledge bases of some firms in the clustercontribute to a stable informal hierarchical network structure over time. The empirical results haveimplications for theories on network dynamics in regional clusters and cluster policies.
It is generally acknowledged that firms that belong to regionallusters achieve superior innovation and economic performanceAlfred Marshall, 1920; Allen, 1983; Piore and Sabel, 1984; Aydalotnd Keeble, 1988; Pyke et al., 1990; Becattini, 1991; Krugman,991; Audretsch and Feldman, 1996; Storper, 1997; Scott, 1998;awson and Lorenz, 1999; Baptista, 2000; Cooke, 2001; Capello andaggian, 2005). However, within this well established and large lit-rature, there is a lack of consensus about what makes regionallusters special. A central tenet of contemporary studies on regionallusters is that geography per se does not guarantee firm successsee e.g. Boschma, 2005; Tallman and Phene, 2007) and that it ishe social networks that are generated across cluster organiza-ions that explain at least part of their innovativeness (Owen-Smithnd Powell, 2004; Smith-Doerr and Powell, 2005; Singh, 2005;hittington et al., 2009).Firms in regional clusters use diverse types of networks to access
nowledge from local and distant actors. Distant ties are importanto increase the variety of knowledge sources in the local contextnd to avoid the cluster formation becoming a technology trap.ocal ties, which are the focus of this paper, bring other bene-ts. First, they are typically high value in terms of the quality of
he knowledge they channel, which is often rich, fine-grained andacit – i.e. ‘capable of transmitting subtle cues’ (Bell and Zaheer,007, p. 957). Its richness derives from the geographical proximity
of managers and workers who are able to meet face to face to dis-cuss problems. Ambiguous and uncertain problems are more easilyresolved through direct observation and confrontation. Second,workers operating in similar environments are likely to encountercontext-specific problems and are more able to develop the exper-tise required to resolve them. The recombination of local skills andknowledge through social networking enables unique solutions,which, in many cases, are at the basis of firms’ product differen-tiation and innovation strategies. Thus, the embeddedness of firmsin local social networks is considered crucial for their upgradingand innovativeness (McDermott et al., 2009; Perez-Aleman, 2011).
Notwithstanding the widespread consensus on the importanceof local networks for promoting innovation in regional clusters, fewscholars have analysed their dynamics. Interest in understandinghow and why networks in regional clusters change over time isrelatively recent and is in line with a new strand of research thatinvestigates cluster evolution processes more generally (Martinand Sunley, 2006; Glückler, 2007; Giuliani and Bell, 2008; Boschmaand Frenken, 2010; Martin, 2010; Menzel and Fornhal, 2010;Boschma and Fornahl, 2011; Martin and Sunley, 2011; Staber, 2011;Ter Wal and Boschma, 2011; Balland, 2012; Balland et al., 2012; Liet al., 2012). Work on network dynamics is motivated by an inter-est in their influence on the development trajectory of clusters. Thedrivers and directions of network change are likely to condition themodes of sharing knowledge (or other valuable assets) among clus-ter firms, which, at least in part, might be a predictor of the cluster’s
future success or failure.
To explain how networks evolve over time, cluster scholarshave borrowed from established organizational sociology and net-work theory concepts and ideas. For instance, Glückler (2007, p.
22) suggests that cluster evolution is likely to be path-dependent,ainly as a result of retention mechanisms in tie formation ‘that
ause new ties to reproduce and reinforce an existing networktructure’. Among these retention mechanisms Glückler (2007)ncludes ‘preferential attachment’, which reflects the tendencyf central actors to become more central over time (Barabasind Albert, 1999), and ‘embeddedness’, which refers to the ten-ency towards network closure and clique-like network structuresGranovetter, 1985). Boschma and Frenken (2010) contribute byuggesting that in addition to geographical proximity, differentorms of inter-organizational proximity likely influence how firmsecome connected in clusters. In other words, in order to con-ect, firms need to be closely related in one or more dimensions.oschma and Frenken (2010, p. 131) posit further that if the reten-ion and proximity mechanisms of new tie formation are in place,he ‘density of network relations in geographical clusters is likelyo increase over time’, which would be undesirable because it couldrevent cluster renewal and might feed lock-in processes. Ter Walnd Boschma (2011) offer another insight into network dynamics inlusters, conjecturing that the characteristics of networks changeslong the cluster lifecycle (CLC). They suggest also that, during therowth stage of the CLC, local networks will tend towards forma-ion of a stable core–periphery structure, in which centrally locatedrms are likely to become even more central through the pro-esses of preferential attachment, and exit of firms positioned inhe periphery of the local network. Only as the CLC matures doeshe network become denser and may drive the cluster into lock-in.
Notwithstanding these attempts to develop a theory of net-ork and cluster growth, scholars (including ourselves) agree that
esearch in this area still ‘needs further development and refine-ent from a theoretical perspective’, and that ‘there is a need for
mpirical validation of the ideas suggested’ (Ter Wal and Boschma,011, pp. 929–930). The theoretical microfoundations of clusteretwork dynamics are unclear, that is, we know little about therm-level factors that drive the formation, persistence and disso-
ution of new ties and how they contribute to the overall structuralroperties of local networks. This paper intends to fill this gap by
nvestigating the factors underpinning the formation and/or per-istence of inter-organizational knowledge ties within the contextf a wine cluster in Chile.
To do so, we posit that the evolution of networks is due tohree sets of concomitant forces, which can co-exist within a clus-er. First, we argue that network dynamics in clusters are likelyo be characterized by strong cohesion effects, due to mechanismsf relational inertia such as reciprocity, and to opportunity condi-ions that lead to higher network closure among cluster firms (i.e.ransitive closure). Second, we conjecture that since the empiricalvidence on cluster networks shows systematically that they are farore fragmented and hierarchically structured than described by
onventional cluster research (e.g. Giuliani, 2007; Ter Wal, 2011a),he formation of new ties may be driven also by status effects,onsidered by network scholars to be the source of asymmetricelationships and hierarchical network structures (Gould, 2002).herefore, we explore whether preferential attachment is a statusffect that drives the formation of new ties. Third, we incorpo-ate aspects of firm-level agency (Ahuja et al., 2012) and considerrm-level variation as an important driver rather than an out-ome of network change (Baum and Mezias, 1992). We draw onvolutionary theories of firm learning and innovation (e.g. Nelsonnd Winter, 1982; Dosi, 1988; Bell and Pavitt, 1993; Dosi andelson, 2010) and explore the impact of a capability effect onetwork dynamics to take account of whether firms with simi-
ar knowledge bases are more likely to establish new ties withne another (i.e. similarity effect), and whether firms with weaknowledge bases are less likely to form new ties over time (i.e.hreshold effect). We concede that the dynamics of networks
42 (2013) 1406– 1419 1407
includes a random component which cannot be fully predicted(Snijders, 2001).
The research is set within the Valle de Colchagua cluster, whichis in one of the most thriving wine areas in Chile (Schachner, 2002,2005) and can be defined as a cluster in the growth stage of itslifecycle. Data were collected through face-to-face interviews con-ducted by the author and a survey administered to the populationsof wineries (32 firms) in the cluster in 2002, and in 2006 (a period ofcluster expansion), which was based on the same structured ques-tionnaire. Social Network Analysis (Wasserman and Faust, 1994)is employed to conduct static comparisons between knowledgenetworks over time, and cohesion, status and capability effects aretested using a class of Stochastic Actor-Oriented Models (SAOM) ofnetwork dynamics developed by Snijders (2001, 2005). The empiri-cal results show that there are two main effects guiding the networkdynamics in CV. Cohesion effects promote greater density in thecluster knowledge network by reinforcing the core of innovatingfirms. Capability effects keep firms with weak knowledge baseson the periphery of the knowledge network. This paper providesevidence that contributes to the refinement of incipient clusternetwork dynamics and cluster evolution theories.
The paper is organized as follows. The theoretical frameworkand research hypotheses are developed in Section 2. Section 3presents the study context and methodology. Section 4 presentsand discusses the empirical results. Section 5 concludes the paper.
2. Theory and hypotheses
Network studies tend to suggest that evolution of the macrostructural characteristics of a network is driven by concurrentforces operating at the micro level (Owen-Smith and Powell, 2004;Powell et al., 2005). Some are endogenously induced by the exist-ing network – for example, past relationships influence future ones(Walker et al., 1997; Gulati and Gargiulo, 1999) and firms occu-pying similar structural positions in the network are likely to beconnected in the future (Rosenkopf and Padula, 2008), while oth-ers are exogenously driven, which means that they are related tothe heterogeneity of the internal and individual characteristics ofthe actors in the network and to their agency (Ahuja et al., 2012).For instance, in a study on inter-firm alliances, similarity in firms’technological and market specializations was found to influencefuture collaboration (Gulati and Gargiulo, 1999), while diversityrather than similarity, has been shown to drive repeated formationof ties in the US biotech industry (Powell et al., 2005). Therefore, itis generally accepted that a network’s dynamics depends on a mixof exogenous and endogenous network effects (Di Maggio, 1992).
In the context of industrial clusters cohesion is a much discussedendogenous network effect. It is described in numerous clusternarratives that report inter-organizational ties as being charac-terized by reciprocity, and highlight that geographical proximityenables close knit social relations among the firms in the clus-ter (Aydalot and Keeble, 1988; Pyke et al., 1990; Saxenian, 1994).However, empirical investigation into how cohesion effects influ-ence network dynamics in regional clusters is scarce. It is plausiblethat, in the absence of any other effect, cohesion would produceincreasingly egalitarian, dense and all-encompassing networks thatdiscourage the formation of hierarchical structures (Granovetter,1973).
Work in economic geography has sparked a different view of thenetwork dynamics in clusters. These studies show that even suc-cessful ‘hot spot’ regions may be spaces where informal relations
are fragmented, and structured very hierarchically (e.g. Ter Wal,2011a; relatedly see also: Markusen, 1996). This points to the needto examine other effects that may underpin network dynamics. Wethink that two types of effects are important. One is status, which
1 olicy
nrttna
adsP1a1
2
ansfifiiam1fitrwofi
uD‘ecGescrtatttoofpp
Ht
ntI‘etiT
reputation more rapidly because information about them diffusesthrough their many direct linkages. The most prominent actorsalso are more frequently cited, which contributes to their aura. In
1 A caveat to transitive closure is that, over the long run, it might give rise to lownetwork variety, which would be detrimental to innovation, especially explorative
408 E. Giuliani / Research P
etwork scholars consider to be a powerful source of asymmetricelationships and hierarchical network structures (Gould, 2002),he other is connected to differences in firms’ characteristics relatedo the abilities to orchestrate and contribute to the local knowledgeetwork. This paper considers differences in firms’ knowledge basess pivotal in shaping network relations (Cohen and Levinthal, 1990).
In order to test the simultaneous roles of cohesion, statusnd capability effects in network dynamics, we propose an inter-isciplinary conceptual framework drawing on (a) organizationalociology and network theories (e.g. Granovetter, 1973, 1985;owell et al., 2005), (b) economic geography (e.g. Amin and Thrift,994; Aydalot and Keeble, 1988; Storper, 1997), and (c) evolution-ry theories of firm learning and innovation (e.g. Nelson and Winter,982; Dosi, 1988; Bell and Pavitt, 1993; Dosi and Nelson, 2010).
.1. Cohesion effects
Cohesion occurs when firms are connected by stable, closednd dense social structures. We consider that cohesion within aetwork can be increased by reciprocity and by transitive clo-ure. In the context of this paper, reciprocity emerges when arm that has been the recipient of technical advice from anotherrm, decides to return (reciprocate) the favour. While reciprocity
s common in human behaviour (Gouldner, 1960), its motivationsnd drivers have been studied as mechanisms promoting the for-ation of new ties in inter-corporate networks (Lincoln et al.,
992; Fehr and Gachter, 2000), with reciprocal ties between rivalrms found to be frequent (von Hippel, 1987). If the firm decideso behave opportunistically, it will not reciprocate to any adviceeceived. Opportunistic behaviour occurs when the firm does notant to dissipate its proprietary knowledge by transferring pieces
f knowledge that might increase the competitiveness of otherrms.
In the context of industry clusters, opportunistic behaviour issually considered to be minimal (Amin and Thrift, 1994). Smith-oerr and Powell (2005, p. 20) argue that, in industrial districts,
repetitive contracting, embedded in local social relationships,ncourages reciprocity. Monitoring is facilitated by social ties andonstant contact’. Likewise, Grabher (1994, p. 181) describes Eastermany’s regional industry in the 1970s as characterized by themergence of informal networks, which ‘provided diffuse infra-tructure for barter governed by the principle of reciprocity’. In suchontexts, reciprocity is guided by two underlying motivations. First,eciprocal relationships are beneficial because they stabilize rela-ionships and increase levels of trust between the parties, creatingdvantageous repercussions for the quality of the interaction. Inhis respect, Ahuja et al. (2012) consider reciprocity to be relatedo organizational inertia, which results in stable routines and rela-ional habits. Second, within a spatially bound area, instances ofpportunistic behaviour are quickly broadcast. A reputation forpportunism will result in severing of existing ties and discourageormation of new ties with other firms. Hence, over time, reci-rocity is a safer strategy for firms keen to take advantage of theool of local knowledge. This leads to the following hypothesis:
ypothesis 1 (HP 1). In regional clusters, reciprocity promoteshe formation of new knowledge linkages among firms.
Transitive closure encourages network growth and increasesetwork cohesion. It occurs when a new link is formed betweenwo actors that are already connected to a common third actor.n social psychology, underpinning transitive closure is known asbalance theory’ (Heider, 1958), which suggests that an individual
stablishes a new linkage with a third one on the basis of whetherhe individuals she/he is already connected to have positive feel-ngs about (and are themselves connected to) this third person.he idea is that the individual perceives some sort of psychological
42 (2013) 1406– 1419
pressure from her/his direct contacts (e.g. friends) and is inducedto choose her/his new contacts in a way that preserves some con-sistency and harmony (or balance) within the social group to whichshe/he belongs (Granovetter, 1973).
Studies of regional clusters generally do not refer to the con-cept of transitive closure (see Ter Wal, 2011b for an exception), butoffer persuasive stories about the existence and importance of net-work closure and are indicative of the tendency for firms to becomeembedded in dense networks (Pyke et al., 1990). These ideas are nottoo dissimilar to the economic geography’s tradition of studies on‘untraded interdependencies’ (Storper, 1997), ‘innovative milieux’(Aydalot and Keeble, 1988; Camagni, 1991) and ‘collective learn-ing’ (Capello and Faggian, 2005). For instance, Inkpen and Tsang(2005, p. 153; emphasis added) consider that characteristics of anindustrial district are “dense, non hierarchical networks of firmslocated within the district, with some of them forming cliques.’Likewise, Scott (1988, p. 31; emphasis added) defines industriallocalities as ‘agglomerations [of producers] that coalesce out ofthe dense networks of transactional interrelations that form as thesocial division of labour deepens and as particular groups of pro-ducers are brought into intense and many-sided interaction withone another’.
One of the main reasons for firms to be part of triads is thatthey are social spaces that allow relationships to be monitoredeasily, which guards against opportunistic behaviour and is likelyto promote exchanges of valuable, tacit and fine-grained knowl-edge (Uzzi, 1997). In other words, they are spaces where the local‘mysteries of the trade become no mysteries’ (Marshall, 1920, p.271) and where knowledge is used and improved.1 Opportunityconditions may also induce network closure, since firms find itconvenient to form relationships with other firms that are sociallyproximate (Ahuja et al., 2012). These conditions are magnified bythe geographical proximity of cluster firms, which provides other-wise unconnected professionals with opportunities to get to knoweach other, for example, at local social events, thus “closing” thetriplet. Professional staff in firms operating in close proximity usu-ally form less antagonistic relationships, which promotes a betterworking environment and more ‘balanced’ relationships. This leadsto the following hypothesis:
Hypothesis 2 (HP 2). In regional clusters, the search for transitiveclosure leads to the formation of new knowledge linkages amongfirms.
2.2. Status effects
While cohesion effects strengthen existing direct or indirectconnections among firms, status acts as a signal to firms withno network contacts and little knowledge about where to seekadvice. Status is defined here as the perceived quality of the actorand a prominent status signals reputation within the community(Podolny, 1993). Ibarra and Andrews (1993) and Lazega et al.(2012) among others, suggest that, under conditions of uncertaintyand ambiguity, the search for advice is socially derived. This meansthat the most centrally positioned actors in the network gain
types of innovative activity (Rowley et al., 2000). However, being part of a triad isnot per se a sign of knowledge redundancy (see e.g. the case of small world net-work structures where local closure is combined with distant ties) and individualactors cannot foresee the consequences of their connectivity choices for the overallnetwork structure when deciding about establishment of a new tie.
olicy
am‘deaa
bnk(etal(mtrtfbiosccti
Hs
2
toMeeealbd1fpbotfid
tbe
d(ao
E. Giuliani / Research P
ddition to reducing uncertainty, selection of a prominent actoray be preferred to selection based on quality judgments which
are costly to make’ (Gould, 2002, p. 1149).2 The status effectepends on the existence of a prominent firm that shapes thevolution of the knowledge network in a regional cluster over time,
phenomenon often associated with the concept of preferentialttachment (De Solla Price, 1976; Barabasi and Albert, 1999).
Preferential attachment is based on the idea that firms guidedy status when searching for technical advice, will target promi-ent firms. This behaviour is common among new entrants with nonowledge about the other firms in their competitive environmente.g. Rosenkopf and Padula, 2008), but can also be associated withxisting firms in a regional cluster. In particular, while it is truehat ‘proximity makes information about local competitors morevailable because managers are better able to scan the activities ofocal competitors compared to the activities of outside competitors’Pouder and St John, 1996, p. 1996), the extent to which this infor-
ation is accessible easily to all cluster firms is debatable. Firmshat are particularly resource-poor may be unable either to collecteliable information about the quality of other firms or to judgeheir value. In a survival or rural cluster in a developing country,or example, the observational scope of many firms may be limitedy the routine of day-by-day activities, which reduces the flexibil-
ty required to search for and accumulate valid information aboutther firms. In other cases, the number of cluster firms may makecanning the quality of all potential sources of advice overly time-onsuming. For these reasons, status can be a valid and time-savingriterion for deciding which firms to approach for advice. This leadso progressive reinforcement of the centrality of prominent firmsn the cluster according to the following hypothesis:
ypothesis 3 (HP 3). In regional clusters, firms with prominenttatus are likely to form more linkages over time.
.3. Capability effects
Endogenous network effects – cohesion and status – are impor-ant, but alone may not be sufficient to explain the significantrganizational variation among co-localized firms (Baum andezias, 1992). We would propose that the formation of new knowl-
dge linkages may be influenced by both endogenous networkffects and differences in one important dimension critical for thestablishment of knowledge linkages – the knowledge base (Cohennd Levinthal, 1990; Giuliani and Bell, 2005). Firms patterns ofearning and knowledge creation vary widely. Their knowledgeases are built through the inherently imperfect, complex and path-ependent process of cumulative learning (Nelson and Winter,982; Arthur, 1988; Dosi, 1988), which results in persistent dif-erences among firms. These differences are likely to be even morerofound in emerging/developing country firms, many of which areehind the technology frontier (Perez-Aleman, 2011). Also, studiesf firm-level learning in developing countries suggest that tradi-ionally these types of firms continue to be technological laggardsor decades, since the accumulation of capabilities through train-ng and knowledge generation efforts is long term and requiresedication and commitment (Bell and Pavitt, 1993).
Studies of regional clusters tend to ignore this aspect, and
hose that do take account of differences among firms’ knowledgeases usually focus on qualitative differences, essentially the pres-nce/absence of technological overlaps in areas of specialization
2 This does not mean that prominence is established in a vacuum and is indepen-ent of the actor’s real and observable qualities. However, as suggested by Gould2002: 1146) ‘socially influenced judgments amplify underlying differences, so thatctors who objectively rank above the mean on some abstract quality dimension arevervalued while those ranking below the mean are undervalued’.
42 (2013) 1406– 1419 1409
(Cantner and Graf, 2006; Tallman and Phene, 2007). The approachproposed here focuses not on differences in fields of knowledge,but rather on the level of sophistication of the firms’ knowledgebases: some are more advanced in relation to the quality and experi-ence of their professional technical workers, and in relation to theircommitment in knowledge-creating activities. This paper suggeststwo ways in which heterogeneity in the strength of knowledgebases influences the formation of new knowledge linkages, thatis, through similarity and threshold effects.
Similarity effects imply that firms prefer to establish knowledgelinkages with firms with the same level of technology or knowl-edge. This is related to firm agency: cluster firms choose whichfirms to establish linkages with on the basis of what they judgethey can gain from the interaction (Giuliani, 2007). Hence, a link-age is likely if both parties will benefit from a pool of knowledgethat is similarly sophisticated and will facilitate learning for bothfirms. If the knowledge bases are too dissimilar, and one firm ismuch more advanced than the other, then a linkage will be lesslikely. In the case of widely differing knowledge bases, the prob-lems that firms encounter will differ and they will be less likely to beengage in mutual help. Thus, knowledge base similarity influencesthe formation of future knowledge ties.
Hypothesis 4 (HP 4). In regional clusters, firms with similarknowledge bases are more likely to form knowledge linkages thanfirms with dissimilar knowledge bases.
The threshold effect is a mechanism that is seldom considered inexplanations of the formation or not of ties. It refers to new linkagesformed only when the parties have some valuable characteristicsthat are over a certain threshold level (e.g. status, power, wealth,skills, etc.). Actors with below-the-threshold characteristics are lesslikely to form linkages. Masuda and Konno (2006) consider thresh-old effect to be a determinant of the formation of elite groups,where entry depends on actors’ characteristics. In a given con-text, individuals with characteristics below a certain threshold donot establish linkages with either those whose characteristics posi-tion them above the threshold or those with similar sub-thresholdcharacteristics. For example, homeless people have similarly pre-carious conditions, but seldom interact with each other (Rokach,2004; Hersberger, 2007). It applies also to people with psycholog-ical disorders or low levels of education (McPherson et al., 2006).
One reason why the threshold effect generally is not consid-ered in studies of network dynamics is that in most of this workthe unit of analysis is resource-rich actors (e.g. inventors, scien-tists, innovative firms) (see e.g. Ter Wal, 2011a,b; Balland, 2012;Balland et al., 2012). However, there is anecdotal evidence on lackof socialization among resource-poor entrepreneurs from otherareas of research. Altenburg and Meyer-Stamer (1999) report casesof survival clusters in Latin America, often located in the shantytowns of large capital cities or in isolated rural areas. These clus-ters are described as inhabited by people who are self-employedor employed in informal work settings, who suffer from severeresource constraints: ‘most of these persons do not have substan-tial savings at their disposal. . .they typically do not master modernmanagement techniques and lack the ability to organize and con-tinuously improve production in a systematic way’ (Altenburg andMeyer-Stamer, 1999, p. 1696). The authors note that in such envi-ronments: ‘the culture of imitation makes entrepreneurs reluctantto share any kind of information; opportunistic or even predatorybehaviour may pay off, because many firm owners perceive theirbusiness as a survival activity to sustain them until a better oppor-
tunity arises’ (Altenburg and Meyer-Stamer, 1999, p. 1697). Similarfindings come from other studies showing that firms with very fewresources have diminished patterns of socialization (Visser, 1999);they are not ‘purposeful and intentional agents’ (Nohria and Eccles,
1 olicy
1t
tkbtu(mal
Hfiw
3
3
3r
ersbaI
tbitdmibpPto(o2
3
lSvpLo(ttawbtdaaC
nical professionals, in this case the agronomists and oenologistsand other technicians employed by the wineries in the cluster, toseek advice on technical problems that cannot be solved in house
3 The available data indicate that, within the cluster, the number of hectaresof vines planted for wine production almost tripled between 1997 and 2002
410 E. Giuliani / Research P
992, p. 13) able to forge new network ties to improve their posi-ioning in the local network.
In this paper, we suggest that there is a threshold effect based onhe firm’s knowledge base, which conditions the formation of newnowledge linkages. We argue that firms with weak knowledgeases are unlikely to increase their knowledge linkages over time,hey have modest knowledge resources on which to draw and arenlikely to be sought out for their knowledge by other cluster firmsGiuliani and Bell, 2005). It has been suggested also that such firms
ay lack the internal capacity to absorb the stocks of knowledgevailable in other cluster firms (Cohen and Levinthal, 1990). Thiseads to our final hypothesis:
ypothesis 5 (HP5). In regional clusters, the poorer quality therm’s knowledge base the less will be the probability that the firmill form new knowledge linkages over time.
. Method
.1. The context
.1.1. Export-led growth in Chile and the importance of naturalesources-based industries
Chile is a small country, but is one of the most thrivingconomies in Latin America. Based on exports from its natural-esources based industries (e.g. mining, agroindustry and fishing),ince 1990 Chile has enjoyed sustained economic growth, a dou-ling of per capita income and a reduction in absolute poverty,lthough income inequality remains high (Perez-Aleman, 2005;nfante and Sunke, 2009).
One sector that has achieved stunning export value growth ishe wine industry. Wine production has a long tradition datingack to the Spanish-Mexican Jesuits who came to Latin America
n the 19th century (Del Pozo, 1998); however, the wine indus-ry boom began only in 1990 in line with increased internationalemand for wine (Giuliani et al., 2011). The spectacular perfor-ance of Chile’s wine industry is evidenced in the export statistics:
n 1994 Chile accounted for only 1.73 percent of total wine exports,y 2004 its share was 4.6 percent (a 266% increase). In the sameeriod, traditional wine producing countries, such as Italy, Spain,ortugal and France, lost market share and experienced a reduc-ion in export values as a percentage of world wine exports (−17%n average). In 2007 Chile was ranked 4th for wine export volume1,157,808 tonnes) after the traditional wine producing countriesf Italy, France and Spain, and 4th for wine export value ($US,414,119,000) after France, Italy and Australia.
.1.2. The Colchagua Valley (CV) clusterExport-oriented growth in the O’Higgins region where the CV is
ocated, has been impressive. This region is about 200 km south ofantiago, the capital of Chile. Between 1990 and 2005, the exportalue of agricultural and agro-industrial activities, such as wineroduction, increased from US$3m to US$161m (Ramirez and Silvaira, 2008). The CV contributed hugely to this increase being onef the most thriving and successful wine areas in the countrySchachner, 2002, 2005). The area is traditionally rural, with a his-ory of wine production dating back to the end of the 1800s. Inhe 1870s French vines from Bordeaux were introduced in the areand since then wine production has increased. Historically, all localine production was sold to large firms near to Santiago, which
ottled the wine and traded it on the market, and, probably dueo their geographical distance from the wine markets, wine pro-
ucers in Colchagua tended to perform only the grape growingnd vinification stages, leaving the final phases of wine bottlingnd marketing to more geographically central firms. In the 1980s,V wine producers – spurred on by the favourable conditions in
42 (2013) 1406– 1419
the international wine market (Schachner, 2002) – embarked ona growth trajectory that coincided with major organizational andtechnological changes. During the 1990s, the cluster was character-ized by the proliferation of firms producing good quality wine fordomestic and foreign markets. Long established firms and domesticand foreign investors were attracted by the favourable terroir andestablished new production plants in the area. The cluster is nowdensely populated by wine producers and grape growers; otherfirms in the upstream and downstream wine production valuechain are located outside the cluster territory, close to Santiago andother major urban areas, or abroad. As a result, the vertical divi-sion of labour within the cluster is fairly shallow. The CV clusteralso includes a business association, aimed primarily at promotingthe wines and marketing them locally, but with no specific man-date to foster innovation or facilitate dissemination of technicalknowledge.
At the time of the first survey (2002), the CV wine industrywas beginning to taste success following ten years of steadilyincreasing investment.3 New modern wineries had been estab-lished and there was a general feeling that CV was set to becomeone of the leading wine areas in Chile. Despite the problems inher-ent in rural Chile (especially inadequate infrastructure), privateinvestors, mostly powerful Chilean families, were making majorefforts, sometimes jointly with public institutions (e.g. CORFO)4
to renovate and modernize the industry and catch up to thetechnological frontier. Already in 2002, some Colchagua winer-ies were as modern as the wineries in advanced countries, andmany firms were using advanced technologies, employing skilledknowledge workers (oenologists and agronomists) and undertak-ing substantial experimentation in their vineyards and cellars. Thiswas reflected in their wines which increasingly were being citedand rated in international specialized wine journals, such as WineSpectator, Decanter, and Wine Enthusiast.5 Nevertheless, a consider-able number of the firms in the cluster were technological laggardsin 2002.
By 2006 the situation had changed dramatically. The mostvisible change was the improvement to the local infrastructureincluding new paved roads, a training institute for local students tospecialize in wine production, and plans for a research laboratoryand a technology transfer office allied to the University of Talca. Thecluster was promoting a set of marketing initiatives ranging fromstrengthening the wine route to setting up new ventures connectedto the flourishing local economy (promotion of local artisans, fairs,restaurants, etc.). These changes were paralleled by the continuingcommitment of the wineries to match international wine qualitystandards. In 2005 Colchagua was awarded ‘Wine Region of theYear’ by Wine Enthusiast, and in 2007, Wine Spectator’s Top 100wines included two Colchaguan wines.
3.1.3. Why this caseThis case is a particularly appropriate context for this study.
The first reason is that earlier research on this cluster (Giulianiand Bell, 2005) showed the presence of local networks. One ofthe most significant networks at the local level is the knowledgenetwork, which has been built through the interactions of tech-
(www.sag.gob.cl).4 CORFO is Corporacion de Fomento, a Chilean government institution that pro-
motes industry development.5 For instance, the number of times Colchagua’s wines have been cited annually
by Wine Spectator increased 10-fold in the period 1994–2002.
sional composite indicator was used: formal training of humanresources; human resources’ experience in the field; and firm’sexperimentation intensity (see below). The first two refer to the
E. Giuliani / Research P
consistent with other industry accounts (e.g. von Hippel, 1987;axenian, 1994). For example, they may want advice on how toreat a pest infestation or how to deal with high acidity levels dur-ng wine fermentation. These networks become established whenhe wineries are committed to improving their products via incre-
ental innovation based on the solutions to technical problems,chieved through interaction with professionals working in otherineries.
The second reason for choosing this case is that this inter-rganizational knowledge network is the result of an informalnd spontaneous networking process, and there is no design orrescription underlying the network structure observed in thistudy. This inter-organizational network has been built through thenformal interactions among individuals (agronomists, oenologists,echnicians), who are the gatekeepers of the firm’s technical knowl-dge and who apply the knowledge acquired from other firms toheir organizational routines. This is consistent with the indus-rial cluster literature, which describes linkages among firms asften poorly formalized through contracts and based mainly onorkers’ and managers’ personal connections. Inkpen and Tsang
2005, p. 153) argue that ‘connectivity between network mem-ers in an industrial district is usually established through informal
nterpersonal relations’. Such networks operate in a similar wayo communities of practice in other contexts (Brown and Duguid,991; Wenger and Snyder, 2000). Agronomists and oenologistsransfer and receive technical advice from professionals in rivalrms in the cluster such that the network operates as a community
n which knowledge exchange is not controlled by firm owners whoight be worried about knowledge leakages (Powell and Grodal,
005).The third reason why this is an ideal case to study network
icrodynamics is that there are no significant external perturba-ions that might influence the networking process for which weeed to control. In the period considered in this study the clusteras experiencing a growth phase (Section 3.1.2), which allowed us
o observe the changes that occurred in the knowledge networkuring this specific phase of the CLC. More importantly, this periodas undisturbed by external macroeconomic or market shocks orolicy interventions aimed at altering the structure of the local
nter-organizational network.
.2. Data collection
This study is based on firm level data collected in the CV clustert two points in time: 2002 and 2006. Prior to the main field-ork, exploratory interviews were conducted to obtain in depth
nowledge on the wine industry in Chile and its contextual and his-orical background. In this preliminary phase, some 50 interviewsere conducted with agronomists and oenologists from severalhilean firms (other than those in Valle de Colchagua) and otherxperts, including several representatives of the main Chilean busi-ess associations and consortia. The questionnaire used in theain fieldwork was tested in pilot interviews with agronomists
nd oenologists working in firms outside the CV cluster. The maineldwork interviews were based on almost identical structureduestionnaires, were conducted face-to-face in August–Septemberf 2002 and 2006.6 The wineries survey did not include suppliersr clients – mainly because, with the exception of grape growers,hese actors are located outside the cluster boundaries. Interview-
es were skilled workers (e.g. oenologists, agronomists) in chargef the production process at firm level; interviews lasted 90 min onverage. The surveys in both years covered the whole population
6 The 2006 questionnaire included some slight modifications which did not affecthe key variables used in this paper.
42 (2013) 1406– 1419 1411
(32 firms) of fine wine producers in the cluster. We interviewedthe agronomists and oenologists in each firm and through themtracked local inter-organizational ties.
Table 1 shows the changes to the firms’ characteristics over fouryears – reflecting cluster development. Their increased size is par-ticularly striking: in 2006 nearly half (48%) employed more than100 people, compared to only 6 percent in 2002. In 2006 the propor-tion of firms with fewer than 20 employees was less than 10%, theaverage size of firms having doubled from 55 to 110 in the period.The number of firms established since 2000 has increased from sixto ten: two exited before 2006, and six new entrants joined thecluster. This pattern reflects a broader pattern of entry and exit inthe cluster, with six firms exiting and six entering.7 The proportionof foreign owned firms had increased by about one-third in 2006.This was not a direct result of entry and exit: all new entrants weredomestic firms that established new businesses, and one of thesix exiting firms was foreign owned. The increased foreign owner-ship is the result of acquisitions of incumbent businesses by foreignowned firms, and the involvement of one domestic incumbent in ajoint venture partnership with a foreign owned firm.
In addition to the general firm-level variables presented inTable 1, the questionnaire was designed to collect other infor-mation relevant to this study in: (i) within-cluster inter-firmknowledge linkages; and (ii) firms’ knowledge bases. Relationaldata for (i) were collected using the roster recall method(Wasserman and Faust, 1994): firms were given a list (roster) ofthe other wine producing firms in the cluster and asked aboutinnovation-related knowledge transfer. Q1 and Q2 (reported below)were directed to the agronomists and enologists employed by thewineries and focus on problem solving and technical assistance andefforts to improve or change the firm’s economic activity. Knowl-edge transfer usually takes the form of a response to a query abouta problem:
Q1: Technical support received [inbound]If you are in a critical situation and need technical advice, towhich of the local firms mentioned in the roster do you turn?[Please indicate the importance you attach to the informationobtained in each case by marking the identified firms on thefollowing scale: 0 = none; 1 = low; 2 = medium; 3 = high].Q2: Transfer of technical knowledge [outbound]Which of the following firms do you think have benefited fromtechnical support provided by your firm? [Please indicate theimportance you attach to the information provided to each ofthe firms according to the following scale: 0 = none; 1 = low;2 = medium; 3 = high].
Since the data were collected in two waves (2002 and 2006),the relational data are expressed in two matrices composed of 32rows and 32 columns, corresponding to the number of firms in thecluster in each year. In the cells in the matrix the existence of a tiebetween firm i in the row to firm j in the column is denoted 1, and 0otherwise.8 The matrix is asymmetric given that, as applies to anyadvice network, knowledge transfer between firm i and firm j maynot be bi-directional.
To measure the firm’s knowledge base (KB) a three dimen-
7 Note that entry and exit of 6 firms does not mean that the exiting firms weretaken over by the new entrants. It is coincidental that over the period studied therewas perfect turnover, resulting in no change to the overall population of firms in thecluster, which was 32 operating firms in both observed periods.
8 Only dichotomous data are used for the analysis in this paper. SAOM analysisdoes not process value data.
1412 E. Giuliani / Research Policy 42 (2013) 1406– 1419
Table 1Key firms’ characteristics in 2002 and 2006.
(a) Size (number of employees)Small (1–19) (%) 28 9Medium (20–99) (%) 66 43Large (≥100) (%) 6 48Average number of employees per firm (number) 55.5 110.5
(b) Year of establishmentPre 1970 (number) 6 −1 5During the 1980s (number) 8 −1 7During the 1990s (number) 12 −2 10During the 2000s (number) 6 −2 + 6 10
(c) Firm entry and exit: 2002–2006Exit – number of firms (number) 6 (5 domestic)Entry – number of firms (number) 6 (All domestic)
(d) OwnershipDomestic (%) 81 66Foreign (%) 19 34
S
hty2eliw
(
(
w
and 2006), based on the set of network structure indicators pre-sented in Table 2(a). Second, the research hypotheses were testedusing SAOM for network dynamics, which are considered to be the
10 Factor loadings and uniqueness are available upon request.
ource: Author’s own data.
uman resources at the time of interviews (2002 and 2006); thehird takes account of experimentation that took place up to twoears prior to the interviews (the pilot fieldwork showed that a
year time span was sufficient to indicate the intensity of firms’xperimentation activity). None of the variables is influenced byocal network ties and, thus, can be considered to capture character-stics that are exogenous to the knowledge network. The variables
ere defined as follows:
a) Human resources’ formal training refers to the cognitive back-grounds of the firm’s knowledge/skilled workers measured aslevel of education. In line with previous work on the returns toeducation, it is assumed that the higher the education degreeobtained, the greater will be the contribution to the firm’sknowledge and innovation activity. Each knowledge/skilledworker is weighted according to the education degree awarded:
Human Resource = 0.8 ∗ Degree + 0.05 ∗ Masters
+ 0.15 ∗ Doctorate
A 0.8 weighting is applied for the number of the firm’s grad-uate employees and highly specialized workers. This weight isincreased by 0.05 times for number of employees with a mastersdegree and 0.15 for number of employees with a PhD degree.9
Only degrees and higher level specialization in technical andscientific fields related to wine production (i.e. agronomics,chemistry, etc.) are considered.
b) Human resources’ experience is months of work experience ofthe qualified human resources. Number of months is indicativeof the accumulation of knowledge via ‘learning by doing’. Thevariable is the result of a weighted mean of months of workaccumulated by each knowledge skilled worker in Chile andabroad:
Months of Experience in the Sector = 0.4 ∗ n◦ months (national)
+ 0.6 ∗ n◦ months (international)
A higher weight is given to time abroad because the diversityof the professional environment might stimulate active learning
9 The weights are defined ad hoc. The indicator was calculated using other weightsithout significant differences for the analysis.
behaviour and a steeper learning curve. Again, only learningexperience related to wine industry activity is considered.
(c) Experimentation intensity is a proxy for knowledge creationefforts. In the wine industry context, indicators such as R&Dexpenditure and number of patents, are neither available normeaningful. Therefore this concept is operationalized on thebasis of the specificities of the context. Based on lengthy con-sultation with industry experts, it was decided to captureexperimentation intensity in terms of the number of productionphases during which experimentation occurred, that is, exper-imentation related to the introduction of different clones orvarieties to the vineyard terroir, management of irrigation andvine training systems, fermentation techniques and enzymeand yeast analysis, and analysis of the ageing period. Experi-mentation intensity was measured on a 0–4 scale (firms withno in-house experimentation score 0).
Although these variables measure different aspects of theknowledge base, they are highly correlated – especially HumanResource and Months of Experience in the Sector (>0.7) – makingconstruction of a composite indicator for firm’s KB appropriate.The composite indicator was extracted using Principal Compo-nent Analysis, applying the same procedure as in an earlier study(Giuliani and Bell, 2005). A single factor was extracted representingmore than 70% of the variance, and denoted firm KB.10 This measureranged from −1.278 to 2.050 in 2002 and from −1.873 to 1.152 in2006.11
3.3. Analysis
The analysis proceeded in two steps. First, static comparativeanalysis of network structure in the two periods considered (2002
11 To cross-check the validity of this measure, the questionnaire has a section onqualitative descriptions of the types of production methods and experimentationactivities carried out by each firm (objectives, length, methods of analysis, etc.) Thisinformation was used to check for correspondence between the quantitative KBindicator and the knowledge base of the CV firms as reflected by more qualitativeinsights. Two experts were consulted (an academic and a consultant) to give theirassessments of the strength of the knowledge bases of each of the firms in the clusteron the basis of the qualitative information collected. Cross-checks confirmed theusefulness of the KB indicator to capture the strength of firms’ knowledge bases.
E. Giuliani / Research Policy
Table 2Summary of key measures for the analysis of the knowledge network.
(a) Measures for comparative static analysis of networksDensity Network density (ND) is defined as the proportion of
possible linkages present in a graph. ND is calculatedas the ratio of the number of linkages present, L, to itstheoretical maximum, g(g − 1)/2, where g is thenumber of nodes in the network: ND = L/[g(g − 1)/2].ND values range from a minimum of 0 to a maximumof 1
Average distance(among reachablepairs)
The average of the geodesic distances between thenodes in the network. The distance is the length of ageodesic between them, which is measured as theshortest path
Fragmentation Proportion of nodes that cannot reach each otherMutual linkages ontotal linkages (%)
Percentage of reciprocated ties on total ties in thenetwork
Share of isolates Percentage of firms with no connections to other firmsin the cluster
GINI coefficient forfirms’ degreecentrality
Distribution of knowledge linkages measured by theGINI coefficient applied to degree centrality (DC). DC isnumber of knowledge linkages established by a firmwith other firms in the cluster, irrespective of thedirection of the linkage
(b) Measure and effects for Stochastic Actor-Oriented Models (SAOM) analysisCohesion effects
Reciprocity A positive and significant ̌ coefficient means thatreciprocation is the means chosen by an actor tomaximize its objective function through the formationof new knowledge ties. If we indicate → as the transferof knowledge, it means that if i → j at time 1, then j → iat time 2
Transitive triplets A positive and significant ̌ coefficient means that newties are formed by closing triads of firms where twoconnections existed in the previous period. For thiseffect the contribution of the tie i → j is proportional tothe total number of transitive triplets that it forms,which can be transitive triplets of the type [i → j → h;i → h] as well as [i → h → j; i→ j]
Status effectPreferential
attachmentThis is tested through the out-degree activity effect. Apositive and significant ̌ coefficient effect reflects thetendencies for actors with high out-degrees (i.e.outgoing knowledge ties) to generate extra outgoingties in the subsequent period
Capability effectsKB similarity This is tested through the ‘KB similarity’ effect. A
positive and significant ̌ coefficient means that tiestend to occur more often between firms with similarvalues in their knowledge base (KB)
KB threshold This is tested through the ‘Inv-KB’ ego effect. Thevariable KB was transformed intoInv-KBi = (Max KB − KBi). A negative and significant ˇcoefficient suggests that the weaker the firm’s KB, thelower the probability that the firm will form newknowledge ties, indicating the presence of a thresholdeffect
Control variablesCovariate ego
effectsThis is controlled through the ‘Covariate ego’ effect. Apositive parameter will imply the tendency that actorswith higher values of the covariate (size, age andnationality) increase their out-degrees more rapidly,
S
mnfS
e
ego); ski(x) are the effects (selected from among a range of struc-tural, individual covariate and dyadic covariate effects) and the ˇk
are the statistical parameters of each effect. If ˇk equals 0, the
hence transfer more knowledge over time
ource: Author’s own elaboration of SNA measures.
ost promising class of model, allowing for statistical inference ofetwork dynamics by simultaneously analysing the impact of dif-
erent types of effects on network change (Snijders, 2001, 2005;nijders et al., 2010).
SAOM can take account of three classes of effects (see Ripleyt al., 2011 for a full description): first, endogenous or structural
42 (2013) 1406– 1419 1413
effects, which are derived purely from sociological theories anddepend on the network itself (e.g. reciprocity, network clo-sure effects, degree-related effects); second, individual covariateeffects,12 which account for the characteristics of the actors in thenetwork (e.g. ego-effects expressing the tendency of actors withhigher values for a given characteristic to have higher out-degrees,and alter-effects expressing the tendency of actors with higher val-ues of a given characteristics to have higher in-degrees; etc.); and,third, dyadic covariate effects,13 based on the existence of some kindof proximity or distance between pairs of actors in the network andexpressing the extent to which a tie between two actors is morelikely if the dyadic covariate is larger.
To assess which effects are likely to drive network change, theSAOM relies on a set of fundamental assumptions:
1. the model is about directed relationships (i.e. a tie goes fromactor i to actor j) and the underlying time parameter t is contin-uous;
2. the changing network is the outcome of a Markov process, whichmeans that the current state of the network determines proba-bilistically its further evolution, whereas the earlier past playsno role;
3. the actors control their outgoing ties, which means that changesto ties are made by the actors that initiate the tie, on the basis oftheir and others’ attributes and their positions in the network,which is why these models are described as ‘actor-oriented’;
4. that at any given moment, one probabilistically selected actor(called ego) may have the opportunity to change one outgoingtie, and that no more than one tie change can be made at anymoment. This means that two actors cannot decide jointly toform reciprocal ties at the same moment between time t andt + 1. Hence, we do not know the order in which ties are createdor terminated between time t and time t + 1.
We assume also that the actor-based network change pro-cess is decomposed into two sub-processes, both of which arestochastic:
5. the change opportunity process, which models the frequency ofactors’ tie changes (change rate);
6. the change determination process, which models the probabilityof tie changes, which depend on the three effects described above(i.e. structural, individual covariate and dyadic covariate).
An important aspect of SAOMs is that they are actor-based sim-ulation models used for statistical inference. This means that themodel parameters have to be estimated from observed data using astatistical procedure involving methods of moments implementedby computer simulations of the network change process. The firstobserved network is used as the starting point for the simulations.The first step in the model is choosing the ego which is allowedto make a change, that is, to initiate or withdraw a tie, or to donothing. The probabilities of this choice depend on the objectivefunction, which expresses how likely it is that an actor will changeits network, and is a linear combination of a set of effects:
fi(ˇ, x) =∑
k
ˇkski(x)
where fi(ˇ, x) is the value of the objective function for actor i (the
12 Individual covariates refer to individual (i.e. actor-bounded) variables.13 Dyadic covariates are pair-wise variables. They are normally expressed in a
squared data matrix, where each cell indicates the proximity of each pairs of actorswith respect to a given dimension (e.g. geographical distance).
1 olicy 42 (2013) 1406– 1419
cidnom(n
bb
mtt
f
(
(
(
mslcvbts
4
4
ypdiwikcshnittf
Table 3Changes in the knowledge network: a comparative overview between 2002 and2006.
Indicators 2002 2006
Density 0.09 0.23Average distance (among reachable pairs) 2.16 1.76Fragmentation 0.44 0.24Mutual linkages on total linkages (%) 43% 75%Isolates on total firms (%) 19% 13%GINI Coefficient for firms’ degree centrality 0.45 0.45
Source: Author’s own data.
Fig. 1. The structure of the local knowledge network in 2002.Source: Author’s own elaboration based on Netdraw.
Fig. 2. The structure of the local knowledge network in 2006.Source: Author’s own elaboration based on Netdraw.
414 E. Giuliani / Research P
orresponding effect plays no role in the network dynamics, if ˇk
s positive then there will be a higher probability of moving in airection where the corresponding effect will be higher; if ˇk isegative the reverse applies. Estimates of the parameters in thebjective function are approximately normally distributed, whicheans that the parameters can be tested by referring the t-ratio
parameter estimate divided by the standard error) to a standardormal distribution (Snijders et al., 2010).
The probability that an actor i makes a change and choosesetween some set C of possible new states of the network is giveny:
exp(fi(ˇ, x))∑x′ ∈ C exp(fi(ˇ, x′))
This formula is used in multinomial logistic regressions andeans that the probability of an actor making a change is propor-
ional to the exponential transformation of the objective function ofhe new network resulting from this change (Snijders et al., 2010).
According to our research hypotheses, in this paper we use theollowing effects (ski(x)), which are also described in Table 2(b):
1) Cohesion effects:• reciprocity effects: the tendency to reciprocate ties over time;• transitive triplets: the tendency towards clustering (e.g.
friends of friends become friends).2) Status effect:
• preferential attachment: the tendency of central actors tobecome more central over time.
3) Capability effects:• similarity in KB effect: the tendency for actors with similar
KB values to form ties with one another;• KB threshold effect: the tendency of actors with lower KB
values to form fewer ties over time.
In the estimation we control also for firm-level variables thatight influence the formation of new ties, such as firm size, mea-
ured as number of full time employees, based on the idea thatarger firms may have a higher propensity to form more ties. Weontrol also firm’s nationality (a dummy variable that takes thealue 1 if the company is foreign owned) since domestic firms maye more likely to engage in local interactions, and for firm age onhe premise that older firms would have had more time to embedocially in the cluster.
. Results
.1. Network characteristics and changes over time
This section presents the results of the static comparative anal-sis of the knowledge networks in CV, in 2002 and 2006. Table 3resents the key structural indicators and shows that the overallensity of the network increased greatly, from 0.09 in 2002 to 0.23
n 2006. This large increase in the total number of links in the net-ork could be interpreted as the higher inclusion of cluster firms
n the knowledge network, reflecting more egalitarian diffusion ofnowledge among cluster firms. The comparative values of otherohesiveness indicators reflect this: average distance decreased byome 40 percent (from 2.16 in 2002 to 1.76), and fragmentationalved (from 0.44 to 0.24), indicating a significant reduction in theumber of disconnected firms in the network. Greater cohesiveness
s reflected also in mutual ties, which account for 75 percent of totalies in the 2006 network, nearly double the 2002 value (43%). Also,he share of isolated firms in total cluster firms slightly diminishedrom 19 percent in 2002 to 13 percent in 2006.
Despite increases in density and reciprocation, the distributionof knowledge linkages does not vary over time. The GINI coefficientof degree centrality, which measures the degree of concentration ofknowledge ties in the network, is the same (0.45) over the period,suggesting that the network’s structural features have not variedsignificantly over time.
To explore this further, the structures of the cluster network in2006 and 2002 are depicted in Figs. 1 and 2 and compared in Table 4,which shows that in 2002 the network had a core–periphery
E. Giuliani / Research Policy 42 (2013) 1406– 1419 1415
Table 4Core–periphery structures in 2002 and 2006.
Source: Author’s own data.Note: Results of stochastic approximation. Estimated parameter based on 987 itera-tions. The convergence of the models was good in all cases (t-ratios were all inferiorto 0.10 for all coefficients in all models) and no severe problems of multicollinearitywere encountered.
** p < 0.01.*** p < 0.001.
Table 7Exploring the threshold-effect of knowledge base in 2002 and 2006.
Average KB 2002 Average KB 2006
(a) Isolated firmsIsolates −0.88 −1.22Rest of the firms 0.58 0.31t-Test (p-value) (0.000) (0.001)
ource: Author’s own data.a The percentages are calculated on the population of firms present in 2002 (32).
t thus includes incumbents of 2006 but not new entrants.
tructure,14 which became even more marked in 2006 (finalt, indicating the extent to which the network matches a pureore–periphery structure, increased from 0.43 in 2002 to 0.86 in006). Also, the density of core-to-core relations increased (from.34 in 2002 to 0.86 in 2006), while peripheral firms persisted ineing poorly connected to the core and especially to other periph-ral firms (periphery-to-periphery density is 0.03 in 2002 and 0.05n 2006). Thus, the structural features that were present in 2002
cohesive core and a loose periphery – have persisted over time.able 5 shows also that, over time, 60 percent of the firms that wereeripheral in 2002 were also peripheral in 2006, and 30 per centad exited the cluster and the industry. Only 10 percent of periph-ral firms had joined the core by 2006. Similarly, the majority of the002 core firms maintained their positions over time. This explainshy, despite increased network density, network linkages contin-ed to be distributed in the same uneven way. In summary, overime network density increased, but the overall core–peripherytructure and linkage distribution did not change.
.2. Results of SAOM analysis
This section reports the empirical results of the SAOM esti-ations and tests the research hypotheses. Table 6 reports the
stimation results. The rate parameter and density effects areeported by default in this type of estimation. The rate parame-er is positive and significant in all models, indicating a significanthange in the formation of new ties, while the negative and signifi-ant coefficient of density indicates that firms tend not to establishnowledge linkages with just any other firm in the cluster (Snijders
t al., 2007).
We test Hypotheses 1 and 2 about the importance of reciprocitynd transitive closure for the formation of new ties. As expected,
14 Core/Periphery Models are based on the notion of a two-class node partition,amely, a cohesive sub-graph (the core) in which nodes are connected to each other
n some maximal sense and a class of nodes that are more loosely connected tohe cohesive subgroup, but lack maximal cohesion with the core. The analysis setshe density of the core to periphery ties in an ideal structure matrix. The densityepresents the number of ties within the group on total ties possible (Borgatti andverett, 1999).
t-Test (p-value) (0.000) (0.000)
Source: Author’s own data.
reciprocity is a very strong and significant driver of the forma-tion of new knowledge ties ( ̌ = 2.67 and s.e. 0.44), which providesstrong support for Hypothesis 1. The network shows a tendencyfor transitive closure, although this effect is not as strong as thereciprocity effect, evidenced by the smaller coefficient size ( ̌ = 0.25and s.e. 0.06).15 This result supports Hypothesis 2. Hence, there isan endogenous cohesion effect which increases the overall densityof the knowledge network.
The model also includes the status effect of preferential attach-ment. The ̌ coefficient for out-degree activity effect, whichmeasures the existence of a preferential attachment phenomenon,is positive, but small and not significant ( ̌ = 0.01 and s.e. 0.02),which does not support Hypothesis 3.
Finally, the model tests the role of capability effects in the for-mation of new knowledge ties. A KB similarity effect is used totest Hypothesis 4. Contrary to expectations, the coefficient is pos-itive, but barely significant ( ̌ = 0.81 and s.e. 0.45), which does notfully support Hypothesis 4. This result is commented on later inthe paper. The ̌ coefficient for the KB threshold effect is negativeand significant ( ̌ = −0.44 and s.e. 0.17), suggesting that firms withless solid knowledge bases are less likely over time to form newknowledge linkages, which supports Hypothesis 5.
The model includes control variables for firm size, nationality
and age, none of which turns out to be significant.
Table 7 shows that in both 2002 and 2006, isolated or periph-eral firms in the knowledge network, tend to have lower level
15 Alternative measures of transitivity and network closure were used to test thishypothesis (e.g. transitive ties), all gave significant results but the transitive tripleteffect was strongest.
1 olicy
kioilistfiaa
4
eratestfiettisno2tptnwaseh
4
oto(paptrdnc
mtkiton1v
416 E. Giuliani / Research P
nowledge bases on average than other firms in the cluster. Thiss evidence that firms with particularly weak knowledge bases arenly poorly connected to the cluster knowledge network and, moremportantly, that firms with weak knowledge bases do not forminkages with similar alters, demonstrated by the low density ofntra-periphery ties (see Table 4). This result explains the weakupport for Hypothesis 4: while it is plausible that similarity mat-ers when firms’ knowledge bases are above a certain threshold,rms with similarly-weak knowledge bases do not establish link-ges with each other, which reduces the power of similarity effects an explanatory variable.
.3. The role of entry and exit
The results reported above do not take explicit account of firmntry and refer only briefly to firms’ exit from the cluster. Thisaises the question of whether the characteristics of the networkre influenced significantly by relative entry and exit patterns. Notehat there are considerable differences in the characteristics of newntrants and firms that exit. New entrants are larger and havetronger knowledge bases than firms that exit the cluster; also,he average number of knowledge workers per employee in newrms is more than double the average in incumbent firms, and newntrants’ average experimentation intensity is seven times higherhan the level in exiting firms and marginally higher on averagehan in incumbent firms. Thus, new entrants display a very strongnnovative profile. They have some influence on the increased den-ity of the overall network, but the large change in this aspect ofetwork structure results mainly from the fact that the densityf incumbents more than doubled from a relatively high base in002. The limited connectedness of new entrants is reflected byhe fact that only one joined the core group, the other five sim-ly replaced firms that had exited the cluster from a position inhe network periphery (Table 5). Hence, most new entrants didot immediately take up positions in the cluster knowledge net-ork, that might have been expected based on their knowledge
nd emerging innovative profiles. We would speculate that cohe-ion effects were probably underway, but that in the case of newntrants compared to incumbent firms which have accumulatedistorical relations in the cluster, they require more time.
.4. Discussion of results
We found that the knowledge network is structurally stablever time, but is quite dynamic at the micro-level. Our finding ofhe persistence of a core–periphery structure is in line with previ-us studies showing the stability of network structures over timee.g. Walker et al., 1997; Uzzi et al., 2002), and with accounts ofath-dependence in the cluster development processes (Martinnd Sunley, 2006; Glückler, 2007). Of interest in relation to theresent findings is that this stability is achieved in spite of (orhanks to) firms establishing many new linkages over time, aseflected by the increased network density, and the micro-levelynamics is not disruptive – it neither rejects the pre-existingetwork structure nor changes the development trajectory of theluster.
The stability of the network in our case is explained by twoicro-level dynamics. As expected, cohesion effects turned out
o be very significant: both reciprocity and transitive closure areey drivers of the formation of many new knowledge ties. Thiss consistent with narratives of regional clusters that describehem as contexts of dense and cohesive networks, and with much
f the organizational sociology literature on inter-organizationaletworks (e.g. Pyke et al., 1990; Lincoln et al., 1992; Saxenian,994; Inkpen and Tsang, 2005). Qualitative insights from the sur-ey interviews confirmed that reciprocity is beneficial in stabilizing
42 (2013) 1406– 1419
relationships over time, helping to make interactions more fluidand spontaneous. In terms of the benefits of transitive closure, oneoenologist interviewed suggested that ‘three is an ideal number tosolve a problem: you have three brains to count on, who interact andshare different expertise, and you reach a solution quickly. Discuss-ions with more than three people are also fruitful but they are oftenlengthy and less effective’. Interviewees also confirmed that geo-graphic proximity acts as a significant trigger for triadic closure:‘there are many occasions within clusters in which professionals withwhom I have a tie meet each other and start interacting’.
However, cohesion effects can be assumed to be responsiblefor the increased density of linkages among core firms, whereasthe threshold effect keeps firms with weak knowledge bases atthe periphery of the cluster knowledge network. It is intriguingthat firms characterized by weak knowledge bases are untouchedby local socialization dynamics and the strength of the cohesioneffects in the cluster. These are resource-poor firms that do notbenefit from the opportunities enabled by physically proximity tohighly interactive firms (i.e. the firms in the core), display littleagency, and show limited internal push for or interest in forgingnew linkages. We interpret this result by referring to Cohen andLevinthal’s (1990) idea of absorptive capacity, which holds that thecapacity of firms to form linkages with external actors depends ontheir knowledge bases, since the ability of a firm to recognize thevalue of new, external information, to assimilate it, and apply it forcommercial ends, requires prior accumulated knowledge (Bell andPavitt, 1993). Hence, firms with weak knowledge bases are unableto absorb and exploit local knowledge flows and, due to their mod-est knowledge resources, are unlikely to be sought out by othercluster firms. However, there may be other interpretations. Thelack of agency of these firms is unusual when examined from theperspective of advanced countries’ research contexts and, in thissense, our result may be somewhat context-specific. However, inmany emerging economies, firm behaviour (and business culturein general) suffers the legacy of perennial macro-economic fluctu-ations and high uncertainty, which often results in short-sightedinvestments and corporate inertia (Arza, 2005). To an extent, thiscontradicts earlier studies on the power of collective learning forthe world’s poorest producers. For instance, Perez-Aleman (2011)shows the relevance of local collective learning for stimulatingSmall and Medium Enterprises’ upgrading in the Nueva Guineamilk production cluster in Nicaragua. Similarly, McDermott et al.(2009) show how even the weakest wine producers in Mendoza,Argentina, managed to upgrade their production thanks to theirconnections. However, the upgrading of weak producers in thesecases was due largely to specific pressures: investment by a multi-national corporation in the Nicaraguan case, and the existenceof policies to stimulate connectivity with Public–Private Institu-tions (PPIs) – in the Argentinean case. Neither of these specificpressures applied to the case analysed here, which may explainwhy the weakest firms remained at the margins of the local net-work.
What merits further discussion is the developmental poten-tial that this hierarchical structure could have for the cluster. Weobserved that this structure does not preclude catching up by thestrongest firms in the cluster. On the contrary, the core–peripherystructure enables the circulation of high quality, tacit and fine-grained knowledge among the densely connected core firms, whosepotential to upgrade their knowledge is high. The persistence ofa core–periphery structure, therefore, acts as an insurance forthe strongest firms in the cluster because it minimizes the riskthat transferred knowledge will become ‘downgraded’ by firms
with weaker knowledge bases which consistently are relegated tomarginal positions in the network. This explains why firms thathave caught up to the technological frontier and become interna-tional leading wine producers are able to coexist in a cluster that
olicy
itHtfio
iatOtegditustociawtn
5
5
rSaMBrrtnatctiowWcphcsp
tdwn2prt
E. Giuliani / Research P
ncludes lagging firms, although these latter may eventually exithe industry, mimicking the behaviour of earlier peripheral firms.ence, the wine producers in this cluster seem to have given rise
o a structure that is functional for the objectives of the strongestrms, which will likely become leading firms in the local economyver the long run.
Finally, we find that hierarchy is associated with the heterogene-ty in the firms’ knowledge bases rather than to status differencesmong firms. This suggests the need to examine possible reasons forhe lack of a status effect and its implications for network dynamics.ne explanation might be that, within regional clusters, uncer-
ainty about firm quality is mitigated by firms operating in the samenvironment, which makes it more likely that information can beleaned first hand and negates the need to rely on status wheneciding about links (Pouder and St John, 1996). However, one
nterviewee had a different interpretation: ‘not all of us have accesso such information not because it is secret, but because to be able tonderstand the real quality of something or someone you need to haveome accumulated experience on that particular quality aspect’. Onhis basis, there may be some cluster firms that are bound to relyn status rather than effective qualities, especially in resource-poorontexts. Although counter-intuitive, this would explain the find-ng of lack of a status effect: firms that are better able to accessnd scan information firsthand, by definition, do not rely on status,hile those firms that might rely on status to orient themselves,
hat is, firms with fewer resources and weak knowledge bases, areot very likely to form linkages because of the threshold effect.
. Conclusions
.1. Contributions
One of the current most debated and least studied issues inegional and innovation studies is network dynamics (Martin andunley, 2006; Glückler, 2007; Giuliani and Bell, 2008; Boschmand Frenken, 2010; Martin, 2010; Boschma and Fornahl, 2011;artin and Sunley, 2011; Ter Wal and Boschma, 2011; Staber, 2011;
alland, 2012; Balland et al., 2012; Li et al., 2012). The presentesearch contributes to this debate. First, we add to the body ofesearch on the dynamics of networks in regional clusters. We showhat cohesion effects are an essential component of a theory ofetwork dynamics in regional clusters. Cohesion effects arise as
result of the strong opportunity conditions within clusters – par-icularly physical proximity – which facilitate triadic closure and aertain degree of inertia and consolidation of existing ties throughhe mechanisms of reciprocity. We show also that cohesion coex-sts with other microdynamics, which are responsible for muchf the fragmentation and hierarchy that characterize most realorld networks, including cluster networks (Giuliani, 2007; Teral, 2011a). Our findings suggest that, at least in emerging country
lusters, lack of agency of some firms may be contributing to theath dependent trajectory of the cluster and its network. Also, theeterogeneity of firms in terms of their accumulated resources andapabilities, contributes to the formation of hierarchical networktructures, where firms with more capabilities occupy privilegedositions within the local network.
Second, our study questions the wisdom of including preferen-ial attachment microdynamics in theoretical models of networkynamics in growing clusters. The concept has achieved popularityith cluster scholars, who use it to explain the growth of clusteretworks (Giuliani, 2007; Glückler, 2007; Boschma and Frenken,
010; Ter Wal and Boschma, 2011). Yet our discussion shows thatreferential attachment drives partner choice based on status andeputational considerations and, therefore, is appropriate in con-exts where information about actors’ characteristics is unavailable,
42 (2013) 1406– 1419 1417
not to be trusted, or costly to obtain first hand. This condition isunlikely to apply to the growth phase of a cluster when firms’ vis-ibility increases and uncertainty reduces. Even new start ups thatenter the cluster appear unlikely to orient themselves solely onthe basis of status considerations: industry associations, chambersof commerce, industry events, and other social interaction spacesavailable to cluster firms can provide new entrants with informa-tion on the quality of actors and orient new entrants’ connectivitychoices thereafter. It seems that, so far, scholars have confusedpreferential attachment with the strengthening of one or a fewprominent actors that have become outstanding not through thenumber of their ties, but based on their agency, their resources andskills, and their capacity to be leading firms. Hence, our researchinvites further research to understand the emergence of dominantactors, their microdynamics, and their impact on network change.
Finally, our findings have implications for research on innova-tion in the context of developing and emerging countries. Much ofthe literature suggests that emerging/developing economies sufferfrom severe market failure and institutional weakness (Hoskissonet al., 2000; Cuervo-Cazurra and Dau, 2009). In this context, firmsthat want to enter the international competition need to cul-tivate and join different types of inter-organizational networks,for example, business groups or interpersonal networks, such asthe Chinese guanxi. These networks provide access to resources,reduce information asymmetries among firms, enable higher bar-gaining power vis-a-vis market counterparts, increase lobbyingpower with governments, and allow firms to upgrade their capa-bilities (Guillén, 2000; Khanna and Palepu, 2000; Stark and Vedres,2006; McDermott et al., 2009). They act as safety nets against uncer-tainty and an unfavourable business climate. This paper contributesby showing that firms may be incapable of becoming membersof the relevant networks. Despite the growing power of emerg-ing economies in the current global competitive scenario, thereare huge parts of these economies where backwardness and iso-lation prevail. The extent to which isolated or marginal firms willbe able to connect to valuable networks and to close the gap withthe most powerful and successful firms in their countries will affectthe competitiveness of these emerging countries with the advancedeconomies. This study shows that even firms that could becomepart of a local network based on cohesion effects, face a dividethat persists over time. Understanding how this divide could bereduced is a challenge for research on the future competitivenessof emerging and developing economies.
This raises some interesting policy questions. It seems fairlyclear that measures designed to foster the networking of firms mayhave limited influence if they try to connect technological leaderswith laggard firms. The former will not be keen to invest time ininteracting with and passing knowledge to weak firms; and the lat-ter are unlikely to be able to absorb and learn from the strongestfirms. In this sense, intermediary organizations, such as the PPIssuggested by McDermott et al. (2009), would better address theneeds of weak firms and support their upgrading process. How-ever, entry/exit dynamics reveals that new entrants are far moredynamic than firms that leave the cluster and, therefore, it is possi-ble that the weaker firms positioned in the periphery may also exitand be replaced by new and dynamic start-ups. Overall, in a grow-ing cluster, new employment and market opportunities emergeand naturally replace those lost due to exiting firms and, therefore,policies that help the weakest firms to survive may be counterpro-ductive in that context.
5.2. Limitations and further research
This paper has some important limitations which provideopportunities for further research, but also suggest some cautionin interpreting the findings. The study is based on a single case,
1 olicy
wrbarfltkwwbntsdapsfcepd
A
CCaoRigCPef
R
A
A
A
A
A
A
A
A
B
B
B
B
418 E. Giuliani / Research P
hich means that the results cannot be generalized. However, theesearch design of this study could be replicated and it woulde interesting to see whether research on other sectors, in bothdvanced and especially developing countries, produces similaresults. In particular, it would be interesting to conduct researchocused on the network dynamics of clusters at other stages in theife cycle, and on both the creation and dissolution of ties. Also,he present study focuses on only one type of local network – thenowledge network. Future research could explore multiple net-ork dynamics. Another interesting research direction for thoseith access to more than two waves of network panel data, might
e to examine the issue of time: are older ties more persistent thanewer ties? If so, what is the impact of new entrants on cohesion-ype or other types of microdynamics? Another limitation of thistudy is that it uses binary data; future research could explore theynamics of networks and look at the value of ties: how do strongnd weak ties differ in terms of persistence/dissolution, and theotential for stimulating new ties via reciprocity or other cohe-ion effects? More broadly future research in this area might shedurther light on the relative importance of agency and firm-levelharacteristics vis-a-vis (and in interaction with) pure structuralffects. Might individual firms (or other organizations) be moreowerful than endogenous network effects in shaping networkynamics? Under what conditions might this effect emerge?
cknowledgements
I would like to thank Cristian Diaz Bravo, Marcelo Lorca Navarro,ristian Goich and the other agronomists/enologists from Valle deolchagua and Mario Castillo and colleagues at CORFO. Thanks golso to five anonymous reviewers for comments on earlier versionsf this paper and to Gabriela Cares, Roberta Rabellotti, Martin Bell,affaella Rotunno and Graciela Moguillansky for their support and
nsights. Funding provided by the EU Marie Curie Fellowship Pro-ram (HPMT-GH-00-00158-01), the UK Economic Social Researchouncil (ESRC) (PTA-030200201739 and PTA-026270644) and therogetto Alfieri – Fondazione CRT (Italy) are gratefully acknowl-dged. Last but not least I am grateful to Alice and Pablo Baldaccior their time.
eferences
huja, G., Soda, G., Zaheer, A., 2012. Introduction to the Special Issue: the genesisand dynamics of organizational networks. Organization Science 23, 434–448.
llen, R.C., 1983. Collective invention. Journal of Economic Behaviour and Organi-zation 4, 1–24.
ltenburg, T., Meyer-Stamer, J., 1999. How to promote clusters: policy experiencesfrom Latin America. World Development 27, 1693–1713.
min, A., Thrift, N., 1994. Globalization, Institutions, and Regional Development inEurope. Oxford University Press, New York.
rthur, B.W., 1988. Competing technologies: an overview. In: Dosi, G., Freeman, C.,Nelson, R., Silverberg, G., Soete, L. (Eds.), Technical Change and Economic Theory.Pinter, London, pp. 590–608.
rza, V., 2005. The Impact of Business Confidence and Macroeconomic Uncertaintyon Firms’ Investment Behaviour in Argentina during the 1990s. SPRU-Scienceand Technology Policy Research, University of Sussex, Brighton (unpublishedDPhil Thesis).
udretsch, D., Feldman, M.P., 1996. R&D spillovers and the geography of innovationand production. American Economic Review 86, 630–640.
ydalot, P., Keeble, D., 1988. High Technology Industry and Innovative Environ-ments: The European Experience. Routledge, London and New York.
alland, P.A., 2012. Proximity and the evolution of collaboration networks: evidencefrom research and development projects with the global navigation satellitesystem (GNSS) industry. Regional Studies 46, 741–756.
alland, P.A., de Vaan, M., Boschma, R., 2012. The dynamics of interfirm networksalong the industry life cycle: the case of the global video games industry
1987–2007. Journal of Economic Geography, 1–25.
aptista, R., 2000. Do innovations diffuse faster within geographical clusters? Inter-national Journal of Industrial Organization 18, 515–535.
arabasi, A.-L., Albert, R., 1999. Emergence of scaling in real networks. Science 286,509–512.
42 (2013) 1406– 1419
Baum, J.A.C., Mezias, S.J., 1992. Localized competition and organizational failure inthe Manhattan hotel industry, 1898–1990. Administrative Science Quarterly 37,580–604.
Becattini, G., 1991. The industrial district as a creative milieu. In: Benko, G., Dunford,M. (Eds.), Industrial Change and Regional Development. Belhaven Press, London,pp. 102–114.
Bell, G.G., Zaheer, A., 2007. Geography, networks and knowledge flow. OrganizationScience 18, 972–995.
Bell, M., Pavitt, K., 1993. Technological accumulation and industrial growth. Indus-trial and Corporate Change 2, 157–211.
Boschma, R., 2005. Proximity and innovation: a critical assessment. Regional Studies39 (1), 61–74.
Boschma, R., Frenken, K., 2010. The spatial evolution of innovation networks. Aproximity perspective. In: Boschma, R., Martin, R. (Eds.), The Handbook of Evo-lutionary Economic Geography. Cheltenham, Edward Elgar, pp. 120–135.
Boschma, R., Fornahl, D., 2011. Cluster evolution and a roadmap for future research.Regional Studies 45, 1295–1298.
Brown, J.S., Duguid, P., 1991. Organizational learning and communities-of-practice:toward a unified view of working, learning, and innovation. Organization Science2, 40–57.
Camagni, R., 1991. Innovation Networks: Spatial Perspectives. Bellhaven Press, Lon-don and New York.
Cantner, U., Graf, H., 2006. The network of innovators in Jena: an application ofsocial network analysis. Research Policy 35, 463–480.
Capello, R., Faggian, A., 2005. Collective learning and relational capital in localinnovation processes. Regional Studies 39 (1), 75–87.
Cohen, W.M., Levinthal, D.A., 1990. Absorptive capacity: a new perspective on learn-ing and innovation. Administrative Science Quarterly 35, 128–153.
Cooke, P., 2001. Regional innovation systems, clusters and the knowledge economy.Industrial and Corporate Change 10, 945–974.
Cuervo-Cazurra, A., Dau, L.A., 2009. Pro-market reforms and firm profitabil-ity in developing countries. Academy of Management Journal 52 (6), 1348–1368.
De Solla Price, D., 1976. A general theory of bibliometric and other advan-tage processes. Journal of American Society for Information Science 27,292–306.
Del Pozo, J., 1998. Historia Del Vino Cileno. Desde 1850 Hasta Hoy. Editorial Univer-sitaria, Santiago de Chile.
Di Maggio, P., 1992. Nadel’s paradox revisited: relational and cultural aspects oforganizational structure. In: Nohria, N., Eccles, R.G. (Eds.), Networks and Orga-nizations Structure, Form, and Action. Harvard Business School Press, Boston,MA, pp. 118–142.
Dosi, G., 1988. Sources, procedures and microeconomic effects of innovation. Journalof Economic Literature 26, 1120–1171.
Dosi, G., Nelson, R., 2010. In: Hall, B., Rosenberg, N. (Eds.), Technical Change andIndustrial Dynamics as Evolutionary Processes. The Economics of Innovation,Amsterdam/New York, North Holland, pp. 51–127.
Fehr, E., Gachter, S., 2000. Fairness and retaliation: the economics of reciprocity.The Journal of Economic Perspectives 14 (3), 159–181.
Giuliani, E., 2007. The selective nature of knowledge networks in clus-ters: evidence from the wine industry. Journal of Economic Geography 7,139–168.
Giuliani, E., Bell, M., 2005. The micro-determinants of meso-level learningand innovation: evidence from a Chilean wine cluster. Research Policy 34,47–68.
Giuliani, E., Bell, M., 2008. Industrial clusters and the evolution of their knowledgenetworks: revisiting a Chilean case. SPRU Electronic Working Paper Series # 171,Falmer, UK.
Giuliani, E., Morrison, A., Rabellotti, R. (Eds.), 2011. Innovation and Catching Up: TheChanging Geography of Wine Production. Edward Elgar, Cheltenham.
Glückler, J., 2007. Economic geography and the evolution of networks. Journal ofEconomic Geography 7, 619–634.
Gould, R.V., 2002. The origins of status hierarchies: a formal theory and empiricaltest. The American Journal of Sociology 107, 1143–1178.
Gouldner, A.W., 1960. The norm of reciprocity: a preliminary statement. AmericanSociological Review 25, 161–178.
Grabher, G., 1994. The disembedded regional economy: the transformation of EastGerman industrial complexes. In: Amin, A., Thrift, N. (Eds.), Globalization, Insti-tutions, and Regional Development in Europe. Oxford University Press, NewYork, pp. 177–195.
Granovetter, M., 1973. The strength of weak ties. American Journal of Sociology 78,1360–1380.
Granovetter, M., 1985. Economic action and social structure: the problem of embed-dedness. American Journal of Sociology 91, 481–510.
Guillén, M.F., 2000. Business groups in emerging economies: a resource-based view.Academy of Management Journal 43 (3), 362–380.
Gulati, R., Gargiulo, M., 1999. Where do interorganizational networks come from?The American Journal of Sociology 104, 1439–1493.
Heider, F., 1958. The Psychology of Interpersonal Relations. Wiley, New York.
Hersberger, J.A., 2007. Extreme disruptions of egocentric social networks: informa-
tion consequences and affective outcomes. In: Paper Presented at the SunbeltXXVII Conference, 3 May, Corfu, Greece.
Hoskisson, R.E., Eden, L., Lau, C.M., Wright, M., 2000. Strategy in emergingeconomies. Academy of Management Journal 43 (3), 249–267.
barra, H., Andrews, S.B., 1993. Power, social influence and sense making: effectsof network centrality and proximity on employee perceptions. AdministrativeScience Quarterly 38, 277–303.
nfante, R., Sunke, O., 2009. Chile: towards inclusive development. CEPAL Review97, 133–152.
nkpen, A.C., Tsang, E., 2005. Networks, social capital, and learning. Academy ofManagement Review 30, 146–165.
hanna, T., Palepu, K., 2000. The future of business groups in emerging markets:long-run evidence from Chile. Academy of Management Journal 43, 268–285.
rugman, P., 1991. Geography and Trade. MIT Press, Cambridge, MA.awson, C., Lorenz, E., 1999. Collective learning, tacit knowledge and regional inno-
vative capacity. Regional Studies 33, 305–317.azega, E., Mounier, L., Snijders, T.A.B., Tubaro, P., 2012. Norms, status and the
dynamics of advice networks: a case study. Social Networks 34, 323–332.i, P., Bathelt, H., Wang, J., 2012. Network dynamics and cluster evolution: chang-
ing trajectories of the aluminium extrusion industry in Dali, China. Journal ofEconomic Geography 12 (1), 127–155.
incoln, J.R., Gerlach, M.L., Takahashi, P., 1992. Keiretsu networks in the Japaneseeconomy: a dyad analysis of intercorporate ties. American Sociological Review57, 561–585.
arkusen, A., 1996. Sticky places in slippery space: a typology of industrial districts.Economic Geography 72, 293–313.
arshall, A., 1920. Principles of Economics. MacMillan, London.artin, R., Sunley, P., 2006. Path dependence and regional economic evolution.
Journal of Economic Geography 7, 603–618.artin, R., 2010. Rethinking regional path dependence: beyond lock-in to evolution.
Economic Geography 86, 1–27.artin, R., Sunley, P., 2011. Conceptualizing cluster evolution: beyond the life cycle
model? Regional Studies 45, 1299–1318.asuda, N., Konno, N., 2006. VIP-club phenomenon: emergence of elites and mas-
terminds in social networks. Social Networks 28, 297–309.cDermott, G., Corredoira, R., Kruse, G., 2009. Public–private institutions as cat-
alysts of upgrading in emerging market societies. Academy of ManagementJournal 52, 1270–1296.
cPherson, M., Smith-Lovin, L., Brashears, M.E., 2006. Social isolation in America:changes in core discussion networks over two decades. American SociologicalReview 71, 353–375.
enzel, M.-P., Fornhal, D., 2010. Cluster life cycles – dimensions and rationales ofcluster evolution. Industrial and Corporate Change 19, 205–238.
elson, R.R., Winter, S.G., 1982. An Evolutionary Theory of Economic Change. Belk-nap Press of Harvard University Press, Cambridge, MA.
ohria, N., Eccles, R.G. (Eds.), 1992. Networks and Organizations: Structure, Formand Action. Harvard Business School Press, Cambridge, MA.
wen-Smith, J., Powell, W.W., 2004. Knowledge networks as channels and conduits:the effects of spillovers in the Boston biotechnology community. OrganizationScience 15, 5–21.
erez-Aleman, P., 2005. Cluster formation, institutions and learning: the emer-gence of clusters and development in Chile. Industrial and Corporate Change14, 651–677.
erez-Aleman, P., 2011. Collective learning in global diffusion: spreading qualitystandards in a developing country cluster. Organization Science 22, 173–189.
iore, M.J., Sabel, C.F., 1984. The Second Industrial Divide: Possibility for Prosperity.Basic Books Inc., New York.
odolny, J.M., 1993. A status-based model of market competition. The AmericanJournal of Sociology 98, 829–872.
ouder, R., St John, C.H., 1996. Hot spots and blind spots: geographicalclusters of firms and innovation. Academy of Management Review 21,1192–1225.
owell, W.W., Grodal, S., 2005. Networks of innovators. In: Fagerberg, J., Mowery,D.C., Nelson, R.R. (Eds.), The Oxford Handbook of Innovation. Oxford UniversityPress, Oxford, pp. 56–85.
owell, W.W., Koput, K.W., White, D.R., Owen-Smith, J., 2005. Network dynamicsand field evolution: the growth of interorganizational collaboration in the lifesciences. The American Journal of Sociology 110, 1132–1205.
yke, F., Becattini, G., Sengenberger, W., 1990. Industrial Districts and Inter-firm
Co-operation in Italy. International Institute for Labour Studies, Geneva.
amirez, J.C., Silva Lira, I., 2008. Globalization and regional development: the eco-nomic performance of Chile’s regions, 1990–2002. CEPAL Review 95, 103–123.
ipley, R., Snijders, T.A.B., Preciado Lopez, P., 2011. Manual for RSiena. University ofOxford, Department of Statistics; Nuffield College (1 May).
42 (2013) 1406– 1419 1419
Rokach, A., 2004. Loneliness then and now: reflections on social and emotionalalienation in everyday life. Current Psychology 23, 24–40.
Rosenkopf, L., Padula, G., 2008. Investigating the microstructure of network evo-lution: alliance formation in the mobile communication industry. OrganizationScience 19, 669–687.
Rowley, T., Behrens, D., Krackhardt, D., 2000. Redundant governance structures: ananalysis of structural and relational embeddedness in the steel and semicon-ductor industries. Strategic Management Journal 21, 369–386.
Saxenian, A., 1994. Regional Advantage Culture and Competition in Silicon Valleyand Route 128. Harvard University Press, Cambridge, MA.
Schachner, M., 2002. The next Napa? Chile’s Colchagua valley is aiming for big time.Wine Enthusiast (March).
Schachner, M., 2005. Wine Region of the Year – Colchagua Valley Chile’s leadingregion for red wines finds its groove. Wine Enthusiast (March).
Scott, A.J., 1988. New Industrial Spaces. Flexible Production Organization andRegional Development in North America and Western Europe. Pion Limited,London.
Scott, A.J., 1998. Regions and the World Economy the Coming Shape of GlobalProduction, Competition, and Political Order. Oxford University Press, New York.
Singh, J., 2005. Collaborative networks as determinants of knowledge diffusionpatterns. Management Science 51, 756–770.
Smith-Doerr, L., Powell, W.W., 2005. Networks and economic life. In: Smelser,N.J., Swedberg, R. (Eds.), The Handbook of Economic Sociology. , 2nd ed.Russell Sage Foundation/Princeton University Press, Princeton, NJ, pp. 379–402.
Snijders, T.A.B., 2001. The statistical evaluation of social network dynamics. Socio-logical Methodology 31 (1), 361–395.
Snijders, T.A.B., 2005. Models for longitudinal network data. In: Carrington, P.,Scott, J., Wasserman, S. (Eds.), Models and Methods in Social Network Analysis.Cambridge University Press, New York (Chapter 11).
Snijders, T.A.B., Steglich, C.E.G., Schweinberger, M., Huisman, M., 2007. Manualfor SIENA Version 3.1. University of Groningen: ICS/Department of Sociology;University of Oxford: Department of Statistics.
Snijders, T.A.B., Steglich, C.E.G., van de Bunt, G.G., 2010. Introduction to actor-basedmodels for network dynamics. Social Networks 32, 44–60.
Staber, U., 2011. Partners forever? An empirical study of relational ties in two small-firm clusters. Urban Studies 48, 235–252.
Stark, D., Vedres, B., 2006. Social times of network spaces: network sequencesand foreign investment in Hungary. American Journal of Sociology 111 (5),1367–1411.
Storper, M., 1997. The Regional World Territorial Development in a Global Economy.Guilford Press, New York.
Tallman, S., Phene, A., 2007. Leveraging knowledge across geographic boundaries.Organization Science 18, 252–260.
Ter Wal, A.L.J., 2011a. Cluster emergence and network evolution: a longitudinalanalysis of the inventor network in Sophia Antipolis. Regional Studies 45 (avail-able online 26.08.11).
Ter Wal, A.L.J., 2011b. The dynamics of the inventor network in German biotechnol-ogy: geographical proximity versus triadic closure. In: Papers in EvolutionaryEconomic Geography (PEEG), 1102, Utrecht University.
Ter Wal, A.L.J., Boschma, R., 2011. Co-evolution of firms, industries and networks inspace. Regional Studies 45, 919–933.
Uzzi, B., 1997. Social structure and competition in interfirm networks: the paradoxof embeddedness. Administrative Science Quarterly 42, 35–67.
Uzzi, B., Spiro, J., Delis, D., 2002. Emergence: the origin and evolution of careernetworks in the Broadway musical industry, 1877 to 1995. Northwestern Uni-versity, Evanston, IL (Working Paper).
Visser, E.-J., 1999. A comparison of clustered and dispersed firms in the small-scaleclothing industry of Lima. World Development 27, 1553–1570.
von Hippel, E., 1987. Cooperation between rivals: informal know-how trading.Research Policy 16, 291–302.
Walker, G., Kogut, B., Shan, W., 1997. Social capital, structural holes and the forma-tion of industry network. Organization Science 8, 109–125.
Wasserman, S., Faust, K., 1994. Social Network Analysis. Methods and Applications.Cambridge University Press, Cambridge.
Wenger, E., Snyder, W.M., 2000. Communities of practice: the organizational fron-tier. Harvard Business Review, 139–145.
