The micro-determinants of meso-level learning and ...

Research Policy 34 (2005) 47–68

The micro-determinants of meso-level learning and innovation:evidence from a Chilean wine cluster

Elisa Giuliania,b,∗, Martin Bella

a SPRU, University of Sussex, Freeman Centre, Falmer, Brighton, East Sussex BN1 9QE, UKb DEA, Faculty of Economics, University of Pisa, Italy

Received 25 February 2004; received in revised form 13 October 2004; accepted 27 October 2004Available online 21 December 2004

Abstract

Most analyses of the relationship between spatial clustering and the technological learning of firms have emphasised theinfluence of the former on the latter, and have focused on intra-cluster learning as the driver of innovative performance. Thispaper reverses those perspectives. It examines the influence of individual firms’ absorptive capacities on both the functioning ofthe intra-cluster knowledge system and its interconnection with extra-cluster knowledge. It applies social network analysis toidentify different cognitive roles played by cluster firms and the overall structure of the knowledge system of a wine cluster inChile. The results show that knowledge is not diffused evenly ‘in the air’, but flows within a core group of firms characterised byadvanced absorptive capacities. Firms’ different cognitive roles include some—as in the case of technological gatekeepers—thatcontribute actively to the acquisition, creation and diffusion of knowledge. Others remain cognitively isolated from the cluster,t©

K

1

hddn

d ahaveand

ousomicsys-

ns,96;eyer-

0

hough in some cases strongly linked to extra-cluster knowledge. Possible implications for policy are noted.2004 Elsevier B.V. All rights reserved.

eywords:Clusters; Absorptive capacity; Knowledge communities; Technological gatekeepers

. Introduction

Over the years, the literature on industrial clustersas emphasised their capacity as loci for knowledgeiffusion and generation. Industrial clusters, which areefined here as geographic agglomerations of eco-omic activities that operate in the same or intercon-

∗ Corresponding author.E-mail address:[email protected] (E. Giuliani).

nected sectors,1 have for this reason been consideresource of dynamic endogenous development andreceived increased attention in both the academic

1 This definition both differs from and overlaps with the numerexpressions adopted in the literature to analyse similar econphenomena, such as industrial districts, localised productiontems, technology districts, milieux, etc. (for different definitiosee among othersBecattini, 1989; Humphrey and Schmitz, 19Markusen, 1996; Porter, 1998; Capello, 1999; Altenburg and MStamer, 1999; Cassiolato et al., 2003).

048-7333/$ – see front matter © 2004 Elsevier B.V. All rights reserved.doi:10.1016/j.respol.2004.10.008

48 E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68

policy arenas. The importance of clustering for knowl-edge diffusion and generation was seminally stressedby Alfred Marshall, who introduced the concept of ‘in-dustrial atmosphere’ (Marshall, 1919) and describedthe district as a place where “mysteries of the trade be-come no mysteries; but are as it were in the air, and chil-dren learn many of them, unconsciously” (Marshall,1920, p. 225).

Following this seminal contribution, later scholarshave emphasised the importance of localised knowl-edge spillovers for innovation, due primarily to the factthat firms in industrial clusters benefit from the avail-ability of a pool of skilled labour and that, mainly due togeographical and social proximity, new ideas circulateeasily from one firm to another promoting processesof incremental and collective innovation (see amongmany others:Becattini, 1989; Asheim, 1994; Saxe-nian, 1994; Audretsch and Feldman, 1996; Maskell andMalmberg, 1999; Belussi and Gottardi, 2000; Baptista,2000).

At the same time, other contributions have stressedthe importance of extra-cluster networking, since themere reliance on localised knowledge can result inthe ‘entropic death’ of the cluster that remains locked-in to an increasingly obsolete technological trajectory(Camagni, 1991; Grabher, 1993; Becattini and Rullani,1993; Guerrieri et al., 2001; Cantwell and Iammarino,2003).

Increasing attention has been given to the influ-ence of clustering on industry learning and competi-t texto via t al.,2 in-t lom-e ies,t er,t theirc s ofk hno-la

on-s s es-t theo lian‘ ro-c r joins

other recent work in questioning the emphasis on spa-tial proximity that has come to characterise much ofthe literature about knowledge flows and technologicallearning in clustered production activities.

With respect to the first issue, extra-cluster linkages,the paper shares with recent work a sceptical view ofthe commonly presumed close relationship betweenfunctional, relational and geographical proximity. Asemphasised byMalmberg (2003), for instance, thereis no reason why learning processes should be terri-torially bounded, and both “local and global circuitsof interactive learning” are likely to be important (p.157). More broadly,Amin and Cohendet (2004)haverecently highlighted the distinction between relationaland spatial proximity as different contexts for learn-ing, with the latter no more likely than the former toshape learning processes: “. . . relational or social prox-imity involves much more than ‘being there’ in termsof physical co-location. . . Crucially, if the sociologyof learning is not reducible to territorial ties, there is nocompelling reason to assume that ‘community’ impliesspatially contiguous communities, or that local ties arestronger than interaction at a distance” (p. 93).

On the second issue, intra-cluster learning, weshare with recent literature a scepticism about the roleof fuzzy social relationships and ill-defined spillovermechanisms as the basis of knowledge flows andlearning processes within territory-bounded commu-nities. Consequently, like several others (e.g.Dickenand Malmberg, 2001; Malmberg and Maskell, 2002;A c-t esses.H rentd ures( ,2 )w orksa Butr har-a og-n seso ntsa tra-a

encec ile( ndr ocial

ive performance in developing countries, the conf this study (Humphrey and Schmitz, 1996; Nadnd Schmitz, 1999; Rabellotti, 1999; Cassiolato e003). Within this, emphasis has been given to the

ernal characteristics of clusters: the spatial aggration of firms and the derived external econom

ogether with various forms of ‘joint action’. Howevhe ‘openness’ of cluster knowledge systems andapacity to interconnect with extra-cluster sourcenowledge seems especially important in such tecogically lagging regions, industries or countries (Bellnd Albu, 1999).

This paper contributes to this field of study by cidering, on the one hand, the linkages that clusterablish with extra-cluster sources of knowledge; onther, by trying to go beyond the received Marshal

knowledge in the air’ idea of intra-cluster learning pesses. In addressing these two issues, the pape

min and Cohendet, 2004), we search for more struured mechanisms that shape these flows and procowever, we pursue this search in somewhat diffeirections. We do not examine meso-level structe.g. cluster labour markets, as suggested byMalmberg003). Instead, likeOwen-Smith and Powell (2004,e focus on characteristics of the nodes in netws influences on the structure of knowledge flows.ather than the organisational and institutional ccteristics of network nodes, we examine their citive characteristics. In particular, the paper focun micro-level (firm-centred) knowledge endowmend analyse how these influence the formation of innd extra-cluster knowledge networks.

The study has been based on empirical evidollected at firm level in a wine cluster in ChColchagua Valley). Inter-firm cognitive linkages aelational data have been processed through s

E. Giuliani, M. Bell / Research Policy 34 (2005) 47–68 49

network analysis (Wasserman and Faust, 1994) andgraph theoretical indicators.

The paper is structured as follows: in Section1, wereview some of the outstanding questions in this field,outline the theoretical framework and formulate the hy-potheses for research. Section2 introduces the casestudy cluster in Chile in the context of recent devel-opments in the international wine industry. Section3explains the methodology applied in the research; andSection4 presents the empirical evidence. A conclud-ing discussion is provided in Section5, with commenton the possible policy implications.

2. Research questions and theoreticalframework

2.1. External openness and the concept of clusterabsorptive capacity

The degree of openness of a cluster is inevitablytied to the degree of extra-cluster openness of itsmember firms and institutions. At a meso level ofanalysis, we definecluster absorptive capacityas thecapacity of a cluster to absorb, diffuse and exploitextra-cluster knowledge (Giuliani, 2002).2 The focusof this study is specifically on firms’ abilities to accessand absorb external knowledge. However, externalknowledge lies not merely outside the firm, as withCohen and Levinthal (1990)absorptive capacity,b ac ande

s inta lesi wl-e pac-i iork ,r thefi rued

acityofi ilatei1

through in-house learning efforts. Consequently, it isdefined here independently of any linkages with ex-ternal sources of knowledge.3 So, following the argu-ment of Cohen and Levinthal, it is firms with higherabsorptive capacities in a cluster that are more likely toestablish linkages with external sources of knowledge.This is explained on the basis of cognitive distances be-tween firms and extra-cluster knowledge, so that firmswith higher absorptive capacities are considered morecognitively proximate to extra-cluster knowledge thanfirms with lower absorptive capacities. From this, thefirst hypothesis is elaborated:

Hypothesis 1. Firms with higher absorptive capacityare more likely to establish knowledge linkages withextra-cluster sources of knowledge.

From this, it would follow that a cluster does notabsorb external knowledgeuniformly through all itsconstituent firms, but selectively through only thosefirms with a low cognitive distance from the techno-logical frontier. Interestingly enough, firms with highexternal openness could be potentially fruitful at lo-cal level if they contribute to the diffusion of acquiredknowledge to other firms in the cluster, and perform astechnologicalgatekeepers(Allen, 1977; Rogers, 1983;Gambardella, 1993). Accordingly, we expect that firmsperform different cognitive roles, according to theirknowledge bases, in interfacing with extra- and intra-c

2i

va-t ty offi rmsi rityi ea rea wille nt ofk can,t y to

ed inS

ut it lies outside both the firmand the Colchaguluster. Consequently, we use external knowledgextra-cluster knowledge interchangeably.

It is argued here that firms are heterogeneouheir capabilities and knowledge bases (Dosi, 1997)nd, therefore, they are likely to play different ro

n interfacing between extra- and intra-cluster knodge systems. At the micro level, absorptive ca

ty is considered a function of the firm’s level of prnowledge (Cohen and Levinthal, 1990). It, thereforeeflects the stock of knowledge accumulated withinrm, embodied in skilled human resources and acc

2 This definition draws on the concept of the absorptive capf firms, defined byCohen and Levinthal (1990)as “the ability of arm to recognise the value of new, external information, assimt, and apply it to commercial ends” (Cohen and Levinthal, 1990, p.28).

luster knowledge.

.2. Firms’ absorptive capacity and thentra-cluster knowledge system

Several contributions in the economics of innoion literature have emphasised that the propensirms to establish knowledge linkages with other fis associated with the degree of similarity/dissimilan their knowledge bases (see e.g.Rogers, 1983; Lannd Lubatkin, 1998). We draw on this body of literatund claim that, even within a cluster context, firmsxchange knowledge depending on: (i) the amounowledge they have accumulated over time andherefore, release to others and (ii) their capacit

3 The operational measure of absorptive capacity is discussection3.


decode and absorb knowledge that is potentially trans-ferable from other cluster firms.

In particular, and in contrast to the conventionalknowledge-spillover story, the exchange of knowl-edge follows some structured rules of behaviour whichare determined by the relative values of firms’ ab-sorptive capacities (i.e. by the cognitive distances be-tween them). So, while on one hand, when firms havevery similar absorptive capacities the exchange ofknowledge is more likely to occur on a mutual basis(Coleman, 1990), as “reciprocity appears to be one ofthe fundamental rules governing information trading”(Schrader, 1991, p. 154);4 on the other, it seems likelythat differences between the knowledge bases of firmswill lead them to play differing, sometimes asymmet-ric roles within the cluster knowledge system. Hence,firms with particularly advanced knowledge bases arelikely to be perceived by other cluster firms as ‘tech-nological leaders’ or ‘early adopters’ of technologiesin the local area, leading to them being sought out assources of advice and knowledge more often than firmswith less advanced knowledge bases. This is also likelyto lead to a degree of imbalance in the knowledge in-teractions of the leading firms since they are less likelyto seek out useful knowledge from firms with ‘lower’knowledge bases (Schrader, 1991). Some firms may,therefore,transfermore knowledge than they receivefrom other local firms, so acting as net ‘sources’ withinthe cluster knowledge system. At the same time, firmshave more incentives toaskfor technical advice whent plyt ,w ayl rmsw toa ng asn em.F rmsm luet ande rms

in-f here.T ction(r icali

are likely to be isolated within the cluster knowledgesystem—a position that is not considered significantwithin perspectives on cluster dynamism that empha-sise the importance of homogeneous meso characteris-tics leading to the pervasive availability of knowledgeand learning opportunities ‘in the air’.

These considerations lead to the formulation of thefollowing hypotheses:

Hypothesis 2(a). Links between local firms are morelikely to develop among firms with higher absorptivecapacity.

Hypothesis 2(b). Firms with differing levels of ca-pacity are likely to establish different kinds of cognitivepositions within the cluster knowledge system.

Underlying this argument about the differentiationof roles is a more general set of issues about how com-munication within a cluster knowledge system is struc-tured. These are now explored.

2.3. Knowledge communities and the structure ofthe intra-cluster knowledge system

Beside inter-firm cognitive distances, the structureof the intra-cluster knowledge system is likely to beinfluenced by the formation of local communities ofknowledge workers,5 who share common language andtechnical background, seek advice from other peers oft on-t hichb ration( der,2

byt nityh 1b nal

ualsw hesec rob-l

aturee erai nitiess

hey know that they will be able to decode and aphe received knowledge (Carter, 1989). Consequentlyhile the similar levels of their knowledge bases m

ead some firms into balanced exchange, other fiith lower but still significant capacities are likelybsorb more knowledge than they release, so actiet ‘absorbers’ within the cluster knowledge systinally, however, the knowledge base of some fiay be so low that it neither offers anything of va

o other firms nor provides a capacity to acquirexploit knowledge that others may have. Such fi

4 Although we recognise the well-known distinction betweenormation and knowledge, we use the terms interchangeablyhis is consistent with the contributions we discuss in this sei.e.Von Hippel, 1987; Carter, 1989; Schrader, 1991), in which a va-iety of terms are used interchangeably: e.g. ‘know how’, ‘technnformation’ and ‘technical advice’.

he same community and in so doing develop spaneous (but not random) networking practices, woost processes of knowledge exchange and geneVon Hippel, 1987; Haas, 1992; Wenger and Sny000; Lissoni, 2001).6

The formation of these communities is drivenhe existence of a certain degree of intra-commuomophily (Rogers, 1983; McPherson et al., 200),ased upon the similarity of the members’ perso

5 Knowledge workers are defined by the literature as individith high education and training in a particular profession. Tharacteristics are normally combined with a high capability in p

em solving (Drucker, 1993; Creplet et al., 2001).6 Such communities have been variously defined in the liter.g. as ‘communities of practice’ (Brown and Duguid, 1991; Wengnd Snyder, 2000) or ‘epistemic communities’ (Haas, 1992). For an

nsight into the differences between these two types of commueeCreplet et al. (2001).


technical background, which is inevitably entangledwith the professional experience followed within thefirm where they work. At the same time, such knowl-edge workers seek advice from other community mem-bers in search of complementary, different solutionsto their specific technical problems, or simply inter-connect to exchange experiences and improve theirtechnical knowledge accordingly. It is conceivablethat such networks tend to be structured by a localhomophily-diversity trade-off. On the basis of this, wedo expect that local knowledge flows within ‘cogni-tive subgroups’ of professionals (and, therefore, firms)rather than randomly in the ‘air’ (Breschi and Lissoni,2001). Hence the third hypothesis is formulated asfollows:

Hypothesis 3. The knowledge system within a clus-ter will be structured and differentiated, reflecting theexistence of distinct ‘cognitive subgroups’.

We test each of these hypotheses in Sections5.1–5.3,aiming to shed light on the relationships between firms’absorptive capacities and their patterns of extra- andintra-cluster knowledge acquisition and diffusion. InSection5.4, we examine the connection between exter-nal linkages and intra-cluster communication patterns,the technological gatekeeper’ role as well as other cog-nitive roles taken by firms in the cluster.

3. The wine cluster in Colchagua Valley

3

stryh portsa triesa adi-t rld’e nd,S re ofg nes(

De-s ofw st porto oun-t ed

at a rate of 27 per cent per year, and the qualityof the product was substantially improved, attain-ing widespread positive appraisal from internationalexperts.

Chile presents ideal conditions for wine pro-duction because of the country’s excellent naturalendowments that result in numerous wine regionscharacterised by favourableterroir.7 In additionwineries have made considerable efforts to modernisetheir technologies and adopt novel productive prac-tices. Old methods have been replaced and firms havedeepened their commitment to experimentation andupgrading of the production process. This rapid andpervasive transition has been described as a ‘winerevolution’ (Crowley, 2000).

Considerable investment at an institutional levelin Chile has supported the firm-level efforts to up-grade and expand the Chilean industry. With respect totechnology, co-financing through competitive fundingschemes has sustained applied research in viticultureand oenology and the interaction between wine pro-ducers and various research institutes and universities(e.g. Universidad Catolica, Universidad de Talca andUniversidad de Chile). With respect to marketing, theexport of wines has been supported through the ad-vice and intermediation of specific institutions, such asPROCHILE.

The Colchagua Valley is one of the promisingemerging wine clusters of the country (Tapia, 2001). Itis located in the VIth Region, about 180 km south-westo desa stalR itha thc n inw hase ied,m r,2

3

us-t croa any

andc

.1. The context: the Chilean wine industry

In the past decade, the international wine induas been characterised by a very rapid growth of exnd by the emergence of new wine producing counnd their entry into the global market. Besides tr

ional producers, such as France and Italy, ‘new woxporters (primarily Argentina, Chile, New Zealaouth Africa and the US) have increased their shalobal exports and upgraded the quality of their wiAnderson and Norman, 2001a,b).

The case of Chile is an interesting example.pite its long-standing tradition in the productionines (Del Pozo, 1998), it is only since the 1980

hat sustained growth in the production and exf wine has been achieved. In the 1990s, the c

ry’s participation in the global wine trade increas

f Santiago and is closed off to the east by the Annd approximately 80 km to the west by the Coaange mountains. The area, is traditionally rural, whistory of wine production dating back to the XIX

entury. It has recently increased its specialisatioine production and since the 1990s the clusterxperienced a period of growth and prosperity, tainly, to the success of the wine industry (Schachne002).

.2. Key features of the Colchagua wine cluster

Being traditionally a wine producing area, the cler is populated by a myriad of predominantly mind small grape growers and wine producers. In m

7 Terroir is a French term that refers to the combination of soillimatic conditions of a specific wine area.


cases, these produce for personal consumption or sellgrapes or processed bulk wine at the local level. In ad-dition to them, the cluster has been characterised by thedevelopment of firms that produce wine for the domes-tic and foreign market with an orientation towards qual-ity. Together with long established firms, domestic andforeign investors have been attracted by the favourableterroir and have, therefore, established new productionplants in the cluster.

During the 1990s, the favourable market conditionsboosted the planting of new vineyards, which doubledin size in the second half of the 1990s (S.A.G., 2000).Accordingly, in the same period, the local productionof wine tripled (S.A.G., 2000).8 Most of the firms in thecluster have invested in new technologies. The cellaris usually the first step made towards modernisation:French steel tanks for fermentation, French or Ameri-canbarriques, Italian bottling lines are not difficult tofind in the cluster. Firms are also very dynamic in intro-ducing new methods and techniques in pruning, irriga-tion and canopy management. More recently, startingfrom the end of the 1990s, the sustained growth trendculminated in an overproduction crisis. Wine producersare currently affected by a global slow down of wineconsumption and by increased competition (Andersonand Norman, 2001a,b), and this has spurred them tointensify their efforts to improve product quality andto enter higher value niches in international markets.

A critical role in the recent innovation process of thewine industry has been played by specialised knowl-e mistse nsi cien-t icha andm rms.B logi-c Alsok enta otht wl-e ssesa

theC epre-s

One can distinguish between four types of firm inthe cluster:

(a) Firms that are vertically integrated at the locallevel, producing bottled wines, usually for qualitymarkets. They undertake all phases of the produc-tion chain, from the vineyard to the market, grow-ing their own grapes, processing them and bottlingtheir own branded wines at the local level. Thesecan be either domestic or foreign-owned.

(b) Firms that are vertically disintegrated at the locallevel: the local vineyard ‘subsidiaries’ of large na-tional firms that own properties in different areasof the country and perform the final steps of theprocess (vinification, bottling, branding and mar-keting) in their headquarters outside the cluster.

(c) Vertically integrated grape growers and producersof bulk wine, usually at low quality.

(d) Non-integrated small-scale growers selling grapesdirectly to one of the three groups above.

A firm’s commitment to producing higher qualityproducts reinforces its need to exercise control overthe process of grape-growing and viticulture, and thishas stimulated a trend towards the vertical integrationof wine producers via either of the first two firm cat-egories noted above. Correspondingly, it has also pro-gressively reduced the importance of subcontracting toindependent grape-growers.

4

4

pri-m wsi sedi thef epfi thatb 28fi iest art

la yG wine,i

dge workers, such as oenologists and agronomployed in firms. Having university qualificatio

n technical fields, these professionals have the sific understanding of the wine making process, whllow new methods of production to be appliedore intense experimentation to be carried out in fiesides, such knowledge workers boost technoal change in firms also as external consultants.nown as ‘flying winemakers’, consultants represvehicle of national and international transfer of b

acit and codified knowledge, spreading frontier knodge on grape growing and wine making procecross many different places.

8 These data refer to the VI Region and not specifically toolchagua Valley. Nonetheless, it is believed that this trend is rentative of this area.

. Methodology

.1. The sample and data collection

The study has been based on the collection ofary data at firm level. This was done via intervie

n a sample of firms in the cluster. As summarin Table 1below, the sample was determined inollowing way. From the total population of winroducers in the cluster (approximately 1009), werst selected the total population of producersottle wine and sell under their brand names—rms, including the subsidiaries of national winerhat normally perform within this cluster only a p

9 This is an estimate kindly provided by the Servicio Agricoanadero, Santa Cruz. The number includes all producers of

ncluding those that produce for personal consumption.


Table 1The sample

Wine producers

Branded Not branded

Locally based, verticallyintegrated firms

Subsidiaries, verticallydisintegrated firms

Vertically integrated, bulk suppliers

Population (N= 100) 21 7 72(Estimate including all other types of firm)

Sample (n= 32) 21 7 4

Of which:National 15 7 4Foreign 6 0 0

of the value chain, typically grape-growing. Then,following the guidance of an expert informant, weselected four other firms that produce and sell bulkwine, normally to the first two groups of firms. Noindependent grape-growers were selected.

A number of pilot interviews in the cluster indicatedthat technical people in the firms were usually the bestinformants about the history and current characteristicsof the firms. More important, they were also key nodesin the cognitive interconnections between firms.10 Theinterviews, based on a structured questionnaire, were,therefore, held with the chief oenologist or the cellar-man of each of the sampled firms.

Apart from general background and contextualinformation, the interviews sought information thatwould permit the development of quantitative indica-tors in three key areas: (a) the ‘absorptive capacity’ ofthe firms, (b) their intra-cluster knowledge communi-cation patterns, and (c) their acquisition of knowledgefrom extra-cluster sources.

4.1.1. Firm-level absorptive capacityIn the literature this concept, a key element in the

analysis here, is described in terms of the knowledgebase of the firm. This is usually identified not only interms of human resources (skills, training, experience,

10 This role of the oenologist was consistent with the behaviourstressed byVon Hippel (1987): “When required know-how is nota eedsi sheda eedst

etc.) but also in terms of in-house knowledge-creationeffort (usually R&D) as in Cohen and Levinthal(1989, 1990). Correspondingly, the structured in-terviews sought detailed information about (i) thenumber of technically qualified personnel in the firmand their level of education and training, (ii) theexperience of professional staff—in terms of time inthe industry and the number of other firms in whichthey had been employed, and (iii) the intensity andnature of the firms’ experimentation activities—anappropriate proxy for knowledge creation efforts,since information about expenditure on formal R&Dwould have been both too narrowly defined and toodifficult to obtain systematically. This informationwas transformed into a scalar value via PrincipalComponent Analysis as explained in Section3.2.

4.1.2. Intra-cluster knowledge communicationpatterns

In the questionnaire-based interview, these kinds ofrelational data were collected through a ‘roster recall’method: each firm was presented with a complete list(roster) of the other firms in the cluster, and they wereasked the following questions:

Q1: Technical support received [inbound]If you are in a critical situation and need technical ad-

vice, to which of the local firms mentioned in theroster do you turn? [Please indicate the importance

aseg

.

vailable in-house, an engineer typically cannot find what he nn publications either: much is very specialised and not publinywhere. He must either develop it himself or learn what he n

o know by talking to other specialists” (Von Hippel, 1987, p. 292).

you attach to the information obtained in each cby marking the identified firms on the followinscale: 0 = none; 1 = low; 2 = medium; 3 = high]


Q2: Transfer of technical knowledge (problem solvingand technical advice) [outbound]

Which of the following firms do you think have ben-efited from technical support provided from thisfirm? [Please indicate the importance you attachto the information provided to each of the firms ac-cording to the following scale: 0 = none; 1 = low;2 = medium; 3 = high].

These questions specifically addressproblem solv-ingandtechnical assistancebecause they involve someeffort in producing improvements and change withinthe economic activity of a firm. This is meant to gobeyond the mere transfer of information, whose accesscan be easily attained through other channels (e.g. tradefairs, the internet, specialised reviews, etc.). Instead,our interest here is to investigate whether local stocksof contextualised complex knowledge are not onlyaccessible but also eventually absorbed by localisedfirms. So, for example, knowledge is transferred byproviding a suggestion on how to treat a new pest orhow to deal with high levels of wine acidity duringfermentation. Accordingly, the knowledge transferredis normally the reply to a query on a complexproblem that has emerged and that the firm seeks tosolve.

The ways in which the responses to questions havebeen operationalised into a set of relational variablesare indicated in Section3.2.

4e

ui-s ter,b lly,r ssiblee ies,s etc.)t nce-m tew oses ores ed:

men-t finedr

Q3: Technical support received [inbound]Question Q3:Could you mark, among the actors in-

cluded in the roster, those that have transferredrelevant technical knowledge to this firm? [Pleaseindicate the importance you attach to the infor-mation obtained in each case by marking theidentified firms on the following scale: 0 = none;1 = low; 2 = medium; 3 = high].

Q4: Joint experimentationQuestion Q4: Could you mark, among the actors in-

cluded in the roster, those with whom this firmhas collaborated in research projects in the last 2years? [Please indicate the importance you attachto the information obtained in each case by mark-ing the identified firms on the following scale:0 = none; 1 = low; 2 = medium; 3 = high].

4.2. Operationalising key indicators

Testing our hypotheses required operationalisationof the following firm-level concepts: absorptive ca-pacity, external openness, and intra-cluster knowledgelinkages. It also required operational indicators of theextent to which, and the most important ways in which,the cluster knowledge system was structured into ‘cog-nitive subgroups’.Table 2summarises the basis of themeasures and indicators used. Further information isprovided in the Appendix.

5

5

canba vee owl-e link-a logytS on-s ands e int tedw cu-l e-m r off w-

.1.3. The acquisition of knowledge fromxtra-cluster sources

The interview also asked about the firms’ acqition of knowledge from sources outside the clusoth at national and international level. Specificaespondents were asked to name on a roster of poxtra-cluster sources of knowledge (universituppliers, consultants, business associations,hose which had contributed to the technical enhaent of firms11. They were also asked to indicahether the firm had co-operated with any of thources for joint research and experimentation. Mpecifically two different questions were formulat

11 The list also contains open lines to permit the respondent toion extra-cluster sources which were not included in the pre-deoster.

. Main empirical findings

.1. Absorptive capacity and external openness

In general terms, the Colchagua Valley clustere described as an ‘open’ knowledge system (Bellnd Albu, 1999) as many of its constituent firms hastablished linkages with external sources of kndge. Firms tend to establish frequent knowledgeges with many of the leading research and techno

ransfer institutions and with universities (seeTable 3).uppliers of materials and machinery, jointly with cultants, are also important sources of knowledgeeem to be the main drivers of technical changhe firms. The cluster firms are also well connecith international sources of knowledge—in parti

ar with foreign consultant oenologists (‘flying winakers’) that play an important role in the transfe

rontier knowledge and techniques in the field. Ho


Table 2Key concepts and their measurement

Hypotheses and concepts Explanation/elaboration of concepts Measure adopted

Hypothesis 1Association between firms’(i) Absorptive capacityand

(i) Absorptive capacityhas four components: (a) thelevel of education of the technical personnel em-ployed in the firm, (b) each professional’s monthsof experience in the industry, (c) the number offirms in which each professional has been previ-ously employed, and (d) the type and intensity ofR&D undertaken by the firm

Absorptive capacity: an index derived from the ap-plication of Principal Component Analysis to the dataabout the four component indicators (see AppendixSection A for more detail).

(ii) External openness (ii) External openness:reflects the firm’s propensityto acquire extra-cluster knowledge

External openness: the number of linkages withextra-cluster sources of knowledge (see AppendixSection C)

Hypothesis 2 Indicators of three key features of individual firms’intra-cluster knowledge linkage are developed.

Graph theoretical methods were adopted to measuredifferent dimensions of the ‘centrality’ of firmscommunication patterns, and more generally theircognitive positions in the cluster. For further detailssee Appendix Section B.

Association between firms’(i) Absorptive capacity (i) The propensity of a firm to be a local ‘source’ of

knowledgeOut-degree centrality index: measures the extentto which technical knowledgeoriginates froma firmto be used by other local firms. The indicator iscomputed on two alternative basesdichotomous:reflects the presence/absence of such

a linkagevalued:analyses the value given to each linkage by

the knowledge-user (a 0–3 range)

and

(ii) Intra-cluster knowledgelinkages

(ii) The propensity of a firm to absorb knowledge fromintra-cluster sources

In-degree centrality index: measures the extent towhich technical knowledge is acquired by/transferredtoa firm from other local firms. Again the indicator iscomputed on two alternative bases:dichotomousandvalued.

and

(iii) Different cognitivepositions in the clusterknowledge system

(iii) A firm’s degree of interconnection with theintra-cluster knowledge system

Betweenness: measures the degree of cognitiveinterconnectedness of a firm on the basis of itspropensity to be in-between of other firms’knowledge linkages.

Indicators of the different roles of a firm in the localknowledge system combine (i) and (ii) above inorder to assess the balance between a firm’s roleas source and absorber of knowledge flowswithin the cluster.

In-degree/Out-degree centrality index (I/O C.I.):measures the ratio between the knowledge receivedand that transferred by each firm

If I/O C.I is >1: the firm is a net ‘absorber’ ofknowledgeIf I/O C.I is <1: the firm is a net ‘source’ ofknowledgeIf I/O C.I is about 1, the firm engages in the mutualexchange of knowledge


Table 2 (Continued)

Hypotheses and concepts Explanation/elaboration of concepts Measure adopted

Hypotheses 3 In order to identify the extent to which knowledgeinteractions in the cluster are structured withinsubgroups of highly interconnected firms, threegraph-theoretic indicators are used. See, AppendixSection B.II. for further detail.

Core and peripheral groups: core-peripheryanalysis allows the identification of a cohesivesubgroup of core firms and a set of peripheral firmsthat are loosely interconnected with the core.

The structuring of the clusterknowledge system inways reflecting distinct‘cognitive sub-groups’ k–core (k = 4): a cohesive subgroup in which each

firm is connected to at least four other firms in thesubgroup

Source: Authors’ own specification of indicators, with sociometric indexes taken fromWasserman and Faust (1994)andBorgatti et al. (2002).

ever, as shown inTable 3, the degree of externalopenness is not homogeneous across the cluster firms,as some firms tend to establish more linkages thanothers.

Given this heterogeneity, a non-parametric correla-tion test was run between the level of firms’ absorptivecapacity and their degree of external openness (Hy-pothesis 1). The test shows a significant correlationbetween those two variables: the Kendall taub cor-

relation coefficient is 0.45 withp< 0.01. This resultconfirms Hypothesis 1: firms with higher absorptivecapacity tend to interconnect more to external sourcesof knowledge than other firms (seeTable 4).

According to our results, the existing knowledgebase of the cluster firms appears to shape the hetero-geneous propensity to interconnect with extra-clustersources of knowledge. This seems consistent with theidea that firms with higher levels of absorptive capacity

Table 3Linkages with extra-cluster sources of knowledge (number of firms with at least one knowledge link to the knowledge sources indicated)

Extra-cluster sources Among the firms with overall ‘openness’ to external sources that was

Above the average On the average Below the average

Research institutes 9 9 4Ceviuc (University Catolica)

Knowledge transfer 8 6 3Joint research projects 5 1 0

Centro Tec. Vid Y Vino (University Talca)Knowledge transfer 7 8 3Joint research projects 1 2 1

Fac. CC. Agronomicas (University Chile)Knowledge transfer 6 2 0Joint research projects 2 1 0

INIAKnowledge transfer 2 0 1Joint research projects 4 1 0

Business associations 8 7 2Vinas de Chile

Knowledge transfer 6 3 0Chilevid

P

S

Knowledge transfer 3Corporacion Chilena de Vino

Knowledge transfer 4

rivate firms: consultants and suppliers 10Domestic knowledge transfer 10Foreign knowledge transfer 9

ource: Authors’ own data.

4 1

4 1

10 610 64 4


Table 4Correlation between external openness and absorptive capacity (non-parametric correlation: Kendall taub coefficient)

Averageabsorptivecapacity

External openness (above the average) (n= 10) 0.74External openness (on the average) (n= 10) 0.07External openness (below the average) (n= 12) −0.67Kendall taub correlation between external openness

and absorptive capacity0.45**

Source: Authors’ own data.∗∗ Coefficient is significant withp< 0.01.

are cognitively closer to extra-cluster knowledge andcan, therefore, operate more easily than other firms asinterfaces or nodes of connection of the cluster with theexternal environment.

5.2. Local inter-firm knowledge exchange: howcognitive positions vary in the cluster

It is also evident that firms do not participate inthe local knowledge system in an even and homoge-neous way. Visual inspection (seeFig. 1) suggests thatfirms tend to interconnect differently to one another:

in particular one group of firms (centre of the figure)are linked, transferring and receiving knowledge fromeach other. In contrast, another group of firms (top left)remain cognitively isolated.

In order to test whether firms that were more cog-nitively interconnected in the cluster knowledge sys-tem also had higher absorptive capacities (Hypothesis2(a)), we ran a second correlation test. As indicated inTable 5, this indicated statistically significant relation-ships between the absorptive capacity and the differentcentrality indexes.

The variation between the different correlationstatistics is also illuminating. We observe that amongthese the highest correlations are between absorp-tive capacity and Out-degree centrality, with both di-chotomous and valued data (Kendall taub = 0.523 and0.532). This suggests that absorptive capacity influ-ences the propensity of firms totransferknowledge toother local firms and hence to be net ‘sources’ of tech-nical knowledge within the cluster system. For the In-degree centrality and betweenness indexes, the correla-tions are weaker, but still significant. This suggests thateven at lower levels of absorptive capacity, firms mightbe linked to the local knowledge system, provided thataminimumabsorptive capacity threshold is reached.

F a grap . Note: Ana iamete

ig. 1. The local knowledge system in the Colchagua Valley:rrow from i to j indicates that i transfers knowledge to j. The d

hical representation source: UCINET 6 on author’s own datar of the nodes is proportional to firms’ absorptive capacity.


Table 5Correlation between absorptive capacity and centrality indexes

Indexes of different communication patterns

Out-degree C.(dichotomous)

Out-degree C.(valued)

In-degree C.(dichotomous)

In-degree C.(valued)

Betweenness

Absorptive capacity 0.523** 0.532** 0.399** 0.323* 0.291*

Source: Authors’ own data (non-parametric correlation: Kendall taub coefficient), see Appendix Section B for definitions of the indexes.∗ Correlation is significant withp< 0.05.

∗∗ Correlation is significant withp< 0.01.

This association between firms’ absorptive capac-ities and their propensity to take different cognitivepositions in the cluster knowledge system (Hypothe-sis 2(b)) is shown more directly inTable 6. This in-troduces a set of classifications of the firms’ cognitivepositions in the cluster—knowledge transferers, mu-tual exchangers, absorbers or isolates—as described inthe first column in the Table.

The Table shows clearly that average absorptive ca-pacity varies considerably across the different cogni-tive positions. Particularly interesting is the differenceobservable between the first three groups: sources, mu-tual exchangers and absorbers and the last one, whichis characterised by isolated firms. This result supportsthe idea that a threshold for inter-firm knowledge ex-change exists, so that when firms’ absorptive capacityis very low, the cognitive distance with other firms’knowledge bases becomes too high (i.e. infinite) andthe firms tend to be isolated. Correspondingly, thosefirms that are sufficiently above the minimum thresh-old have a higher probability of being interconnectedwith other local firms. Given the way the questions

Table 6Firms’ absorptive capacities and cluster cognitive positions

Cognitive positions in the cluster Absorptivecapacitymeasure

Sources (n= 5) 1.00

A

M

I

S

were centred on problem-solving and performance im-provement, these linked firms, in contrast to the isolatedfirms, are likely to improve the quality of their produc-tion by virtue of such linkages. On the basis of theseresults we accept Hypothesis 2(b).

5.3. Structure of the intra-cluster knowledgesystem

In order to analyse how knowledge flows werestructured within the cluster knowledge system, weadopted graph theoretical measures for identifying cog-nitive subgroups within the cluster, by which we meansubgroups of firms that have established more re-lations with members internal to the subgroup thanwith non-members (Alba, 1973). In particular, we ap-plied core/periphery models to our data (Borgatti andEverett, 1999). These allow the identification of centraland peripheral poles within cluster knowledge systems.

In the case of the Colchagua cluster we observe theformation of a clear core-peripheral knowledge struc-ture where (a) firms in thecoretend to be highly inter-connected among themselves, whereas (b) peripheralfirms tend to establish loose linkages with the core firmsand virtually no interconnections with other peripheralfirms. More specifically, we show inTable 7, the den-sity of the four types of relations namely: core-to-core(top left), core-to-periphery (top right), periphery tocore (bottom left) and periphery to periphery (bottomright).12 Density is highest for core-to-core relations( sferk ted,t y pe-r butt t the

et.

firms with an In/Out degree centrality ratio >1bsorbers (n= 5) 0.65firms with an In/Out degree centrality ratio <1utual exchangers (n= 8) 0.07firms with an In/Out degree centrality ratio = 1

solates (n= 14) −0.88firms with In and Out centralities

approximating to 0

ource: Authors’ own data.

0.571), which means that core firms tend to trannowledge more often within the core. As expechey are also identified as sources of knowledge bipheral firms (core-to-periphery density is 0.155),his relation is much looser than the previous one. A

12 For core/periphery analysis we adopted a directional datas


Table 7Density of linkages within and between core and peripheral groups

The density of linkages(knowledge transfer fromrow to column)

Averageabsorptivecapacity

Core Periphery

Core (nC = 14) 0.571 0.155 0.58Periphery (nP = 18) 0.083 0.026 −0.45

Source: UCINET 6 applied to author’s own data.Note:The densityof a network is the total number of ties divided by the total numberof possible ties.

same time core firms do not mention peripheral firms assources of technical advice (periphery-to-core densityis very low 0.083) and even less do peripheral firmstransfer or receive knowledge from other peripherals(periphery-to- periphery density is 0.026).

The association between these core/periphery po-sitions and the absorptive capacity of the constituentfirms is interesting. Core firms have an average absorp-tive capacity of 0.58, while the same data for peripheralfirms is 0.45. This seems consistent with the idea thatcore firms, having higher absorptive capacities, boostlocal processes of incremental learning and stimulatesome peripheral firms to ask for technical advice, al-though these relations are not nearly as intense as thosewithin the core group itself.

To improve understanding of this core sub-group,we undertook a further step in the analysis, the identifi-cation of 4-cores within the cluster. This identifies firmsthat have established at least four knowledge linkageswith other firms of the sub-group. This is carried out by

taking into account the whole set of knowledge links,independently of their estimated importance. As shownin Fig. 2, we find a complete network of interrelatedfirms, which correspond to the core identified above.

These results support Hypothesis 3. In particular,the local knowledge system has the structural charac-teristics of a core/periphery set where knowledge in-teractions are clearly concentrated within a subset ofcore firms. Furthermore, consistent with previous sec-tions, this core group is formed by firms that have, onaverage, higher absorptive capacities than the firms inthe periphery. The data are consistent with the exis-tence of a single community of fairly well connected,skilled knowledge workers who tend to exchange moreknowledge within the community (i.e. within the core)than outside it. In contrast, with their relatively weakknowledge base, peripheral firms are not part of the coreknowledge community in the cluster. In other words,we observe that firms’ absorptive capacities and the par-ticipation of their professional personnel in knowledgecommunities are interwoven elements which shape thestructure of the local knowledge system.

5.4. Linking intra- and extra-cluster knowledgesystems: the role of technological gatekeepers

In this section, we bring together the data about (a)the external openness of firms and (b) the ‘cognitiveposition’ of firms within the local knowledge system.Combining these parameters we identified five mainl g tot ed

F ET 6 o adopted as e node

ig. 2. The core group: an analysis of 4-cores Source: UCINymmetrised version of the original dataset. The diameter of th

earning patterns within the cluster, correspondinhe following five types of ‘cognitive role’, as indicat

n author’s own data. Note: The linkages are undirected as wes is proportional to firms’ absorptive capacity.


Table 8Differing firm-centred cognitive roles in cluster learning

Source: Authors’ own data.

in the cells inTable 8that are highlighted inside theheavy broken lines. More detailed characteristics ofthe firms playing these five types of role are indicatedin Table 8.

(a) Technological gatekeepers (TGs):firms that havea central position in the network in terms ofknowledge transfer to other local firms and thatare also strongly connected with external sourcesof knowledge.

(b) Active mutual exchangers (AMEs):firms that forma central part of the local knowledge system withbalanced source/absorber positions within thecluster. They also have relatively strong externallinks. Although they are less strongly connectedto external sources than the TG firms, they behavein a similar way to ‘technological gatekeepers’by bridging between external sources and localabsorbers of knowledge.

(c) Weak mutual exchangers (WMEs)consist of firmsthat are similar to AMEs in that they are welllinked to external knowledge sources and play arelatively balanced source and absorber role withinthe cluster. However, compared with AMEs, theyare less well connected to other firms in the cluster.

(d) External Stars (ES):13 firms that have establishedstrong linkages with external sources, but havelimited links with the intra-cluster knowledge sys-tem. These weak intra-cluster links are primarilyinward and absorption–centred.

(e) Isolated firms (IFs):are poorly linked at both thelocal and extra-cluster levels.

Tables 8 and 9indicate that firms playing three ofthe five cognitive roles (TGs, AMEs and WMEs) areconnected actively into the cluster knowledge systemand contribute positively to its learning processes. Theyconstitute the core of the absorptive capacity of thecluster. The other two groups’ cognitive links with thecluster system are much more marginal.

All the firms contributing to the cluster’s absorp-tive capacity engage in a combination of three keyactivities: acquiring knowledge from outside thecluster, generating new knowledge through their ownintra-cluster experimentation, and contributing tointra-cluster diffusion. The strength of this combina-tion of positive roles differs between the groups and is

13 This term is taken fromAllen (1977)where it describes individ-uals having similar positions within firms i.e. with strong links toexternal sources of knowledge plus weak links to the internal knowl-edge system.


Table 9Heterogeneity in the cluster knowledge system

Characteristics of firms’ differing cognitive roles in the cluster knowledge system

ES (n= 3) TG (n= 5) AME (n= 4) WME (n= 3) IF (n= 10) Others (n= 7) Total (n= 32)

1. Absorptive capacityComposite index 1.7 1.00 0.16 0.04 −0.70 – 0

Within whichIntensity of experimentation 3.0 3.0 1.75 1.66 0.7 – 1.59

2. OpennessExternal openness 10.33 11.0 8.5 8.9 3.1 – 7.12

3. Intra-cluster cognitive positionsAverage Out-degree centrality

Dicot 3.0 7.2 6.0 3.0 0.36 – 2.90Valued 5.3 12.8 11.5 7.3 0.73 – 5.37

Average In-degree centralityDicot 4.3 4.0 5.0 3.0 0.5 – 2.90Valued 8.3 6.6 10.25 7.6 1.1 – 5.37

Ratio In/Out degree centralityDicot 1.44 0.58 0.8 1.0 1.5 – 0.85Valued 1.56 0.51 0.9 1.0 1.6 – 0.86

Betweenness 20.0 43.2 26.75 8.0 0.2 – 13.90

Source: Authors’ own data.

most evident in the case of the TGs and AMEs. Theyhave relatively high Out-degree centrality indexes, andthis is particularly so with respect to the valued dataindexes, reflecting the qualitative importance attachedto the knowledge they transfer to other firms. There arenevertheless substantial differences between these twogroups. The TG firms play a striking role as net sourcesof knowledge for the cluster system (with In/Out degreeindexes of only about 0.5), while the AMEs act onlymarginally as net sources (with In/Out degree indexesaround 0.8–0.9). Behind this difference lie differingfirm-level capacities. Compared to the TGs, the AMEsundertake more modest local experimentation andhave more limited intra-firm knowledge resources. Incontrast, the TGs are not just well connected to externalknowledge sources. They are also significant creatorsof knowledge in their own right, demonstrating a highintensity of local experimentation; and this is backedby relatively strong intra-firm knowledge resources.These are typically ‘advanced’ firms that operate veryclose to the technological frontier14 and whose pro-

14 The wine industry has gone through substantial scientific andtechnological changes over recent decades. Both old world produc-ers (e.g. France) and new world producers (e.g. California, Australia,South Africa, etc.) have contributed to define a widely accepted fron-tier of technology in this industry (seePaul, 2002).

duction is oriented towards the exportation of premiumwines.

These TG firms, therefore, tend to be local deposi-tories of technical novelties, which they apply and con-textualise in their economic practice. Their technolog-ically advanced position is normally acknowledged bythe rest of the firms in the cluster15 and this spurs thelatter to ask for advice. This explains their asymmet-ric position as substantial net sources of knowledge inthe cluster. It is pertinent to note that these TG firmsin the majority of the cases have vertically integratedoperations located within the cluster and are, therefore,well embedded in the local area.16 Their willingness toengage in unreciprocated knowledge transfer to otherlocal firms, may reflect the positive externalities asso-ciated with this. In a wine area, such as Colchagua,which is currently investing in achieving internationalacknowledgement for the production of high qualitywines, the improvement of every producer in the areais likely to generate positive marketing-related exter-

15 A question specifically addressed this issue. Respondents wereasked to provide three names of firms that they considered advancedin the cluster, with respect to their degree of technical modernisationand the quality of wine produced.16 This does not mean that such firms are all locally-owned. Indeed

in three of the five cases the ownership is wholly or partially foreign.


nalities for the whole area, and these may outweigh thepossible cost to these firms associated with imbalancedknowledge transfer relationships with competing firms.

Among the relatively active participants in the clus-ter knowledge system, the WME group has some sim-ilar attributes to the AME firms (moderate levels ofopenness and experimentation). However, with a verylow level of absorptive capacity, the WME firms havemuch more modest knowledge resources to draw on. Itis perhaps not surprising, therefore, that they seem to besought out only infrequently to contribute knowledge toother firms, as reflected in the low Out-degree central-ity indexes (3.0 and 7.3 for the dichotomous and valueddata respectively). At the same time they have, simi-larly, low levels of In-degree centrality. Consequently,although they have a balanced exchange of knowledgewith other firms (an In/Out degree ratio of 1.0), thoseinteractions take place at a much lower level than in thecase of the AME firms.

The two marginal groups of firms differ widely. TheIFs have extremely low scores on almost all the indi-cators. They have very low absorptive capacity; theyundertake almost no experimentation; and they acquirealmost no knowledge from extra-cluster sources. Notsurprisingly, they are rarely sought out as knowledgesources by other firms. But it is striking that they alsorarely seek out knowledge from other cluster firms(demonstrating by far the lowest In-degree centralityindexes in the sample: 0.5 and 1.1). These firms, ac-counting for nearly one-third of the sample, are barelyc

thek byf citya desa thec ancef lingt wnl en-n playa ter.O wl-

d ina velso withe

edge to other cluster firms (reflected in very low Out-degree indexes: 3.0 and 5.3). On the other, as reflectedin their high In-degree centrality values, they seek outadvice from the ‘advanced’ firms inside the cluster, par-ticularly the technological gatekeepers, although theytend not to reciprocate the transfer of knowledge.

In other words, although these external star firms areperhaps best positioned of all the cluster firms to makepositive contributions to the cluster knowledge system,they rarely do so. To the extent that they engage with theintra-cluster knowledge system, this is primarily aboutextracting and absorbing cluster knowledge, not aboutcontributing to it. This pattern appears not to reflectthe ownership status of the firm,18 but may reflect theproduction structure of the firms. They are mainly ver-tically disintegrated subsidiaries of large national wineproducers that base their main operations elsewhere,and are, therefore, not well embedded in the local area.This might partially explain why their behaviour con-trasts so sharply with that of the other advanced firmsthat do operate as local technological gatekeepers.

In summary, then, because of the limited role playedby these two groups of firms, the overall technologicaldynamism of the cluster as a whole seems to bedriven by less than half of the sample firms—theTGs, AMEs and, to a lesser extent, the WMEs (12 outof 32). Although this is a relatively dynamic clusterthat is moving ‘upwards’ in international markets, itstechnological dynamism is not driven by a widespreadcommunity of technologically dynamic firms oper-a ter.I mst dgec elyp s ofk bero that

lysis.H starsw iningfi : oneT rly,c s int d theo y thefi les off ion int

onnected to the cluster knowledge system at all.17

The External Stars are marginally connected tonowledge system in a different way. They havear the highest index of firm-level absorptive capamong all the firms in the cluster (1.7), and this incluhigh intensity of experimentation carried out in

luster. This suggests they face a low cognitive distrom extra-cluster sources of knowledge, enabhem to draw heavily on those sources for their oearning and innovation (as reflected in the high opess index of 10.3). At the same time, though, theyhighly imbalanced cognitive role inside the clusn the one hand, they pass on very little of their kno

17 In addition, there are four other isolated firms, not includeny of the selected groups of firms. These have, similarly, low lef connection to other firms in the cluster, though stronger linksxtra cluster sources of knowledge.

ting, similarly, and pervasively across the clusnstead, it is driven by a relatively small group of firhat is organised within a core interacting knowleommunity, surrounded by greater number of largassive firms that occasionally absorb elementnowledge from the core group or, in a small numf cases, directly from external sources. Moreover,

18 The small number of observations precludes meaningful anaowever, it may be pertinent to note that, while two of the threeere subsidiaries of large national (not foreign) firms, the remave of the seven nationally-owned subsidiaries were as followsG, one WME, one IF and two in the ‘Other’ categories. Similalear conclusions cannot be drawn about the six foreign firmhe cluster. Half of them were technological gatekeepers anthers fell into more isolated categories. Ongoing research brst author in other wine clusters suggests that the cognitive rooreign companies depends more on the duration of the locathe cluster than on their foreignness.


Fig. 3. Approximate location of wine cluster firms with differing cognitive roles, Colchagua Valley, Chile.

core group is itself organised in a hierarchical knowl-edge structure that has a few very capable, knowledge-creating, and external knowledge-acquiring firms atthe ‘top’, acting as the main sources of new knowledgefor the cluster. Below that, firms with progressivelylower levels of these qualities shift from being netcontributors into the cluster knowledge system tobeing net absorbers of that circulating knowledge.

It is interesting to note also that this hierarchy doesnot seem to be influenced in any way by geography.Fig. 3 indicates the physical location of firms in thecluster, distinguished by their cognitive roles. All thefirms are distributed along the valley running from SanFernando in the East to Marchihue and Lolol in the Westand their spatial propinquity and cognitive roles seem

unrelated. Indeed, in the subcluster of firms aroundPerallillo technological gatekeepers are closely locatedwith all the other types of firms. Similarly, in the sub-clusters around Santa Cruz and Nancagua, the closestneighbours to the technological gatekeepers are cogni-tively isolated at the local level.

6. Conclusions

The results of the analysis in this paper callinto question the extent to which clusteringper seinfluences the learning behaviour of cluster firms.In Colchagua at least, the spatially clustered wineproducers demonstrated a wide range of different


communication and learning behaviours. For some,learning links with other organisations ran stronglyoutside the valley, almost exclusively so in somecases; and a substantial number of other firms werealmost totally isolated from any learning processesat all whether within the cluster or outside. Amongthe firms that did demonstrate cluster-centred learningrelationships, the cognitive positions and roles variedwidely. This heterogeneity, associated with the dif-ferences in firms’ absorptive capacities, suggests thata cluster is a complex economic and cognitive spacewhere firms establish knowledge linkages not simplybecause of their spatial proximity but in ways that areshaped by their own particular knowledge bases.

Consequently, the results shed light on the relation-ship between meso and micro within the cluster. Insteadof the common emphasis given to ways in which themeso-level characteristics of clusters influence microbehaviour, this study highlights the importance of theopposite direction of influence. It was the capacities ofindividual firms to absorb, diffuse and creatively ex-ploit knowledge that shaped the learning dynamics ofthe cluster as a whole.

This direction of relationship was exemplifiedparticularly clearly in the case of one meso charac-teristic that has been suggested as important for thelonger run growth of a cluster: its openness to externalknowledge, or more specifically its capacity to acquireexternal knowledge and absorb it into its productionactivities. In the Colchagua cluster, this meso-levela dgeb thec ualfi ac-q yc city.H nelsi acto glys rms.

be-t byA -s thisfi wl-e t int andu

157–158), they were highly structured. Also, although,like Owen-Smith and Powell (2004), we have high-lighted the importance of networks nodes in shapingthe structure of interactions, we have shown the signif-icance of their cognitive rather than organisational andinstitutional characteristics. This leads into, but leavesopen, questions about why firms with these characteris-tics behaved in the ways we describe. In particular, whydid firms with very similar cognitive characteristics be-have as differently as the Technological Gatekeepersand the External Stars? Also, what factors drive the in-teractions among the members of the knowledge com-munity? These questions are not pursued here but arebeing examined in further research by the first author.

Our conclusions also prompt speculative questionsabout the long term evolution of cluster cognitive sys-tems of this type. Our cross sectional data do not throwlight directly on that dynamic, but the indirect illumi-nation they provide does prompt reflection about thecircumstances that might underpin it. Recall key fea-tures of the current situation: one group of firms withthe strongest knowledge bases, the most intensive in-house experimentation and the strongest links to exter-nal knowledge sources were the External Stars whichcontributed very little to the intra-cluster learning sys-tem. At the same time, a relatively small number ofother firms played strong positive roles in acquiringor developing new knowledge and diffusing it morewidely in the cluster; and finally nearly one-third ofthe sample firms were disconnected from the system.F n bee

sivea m. Ag torsi nalS orel allyi tema therd tra-c nantd glyc no-l ersr ’ ofk rs).T ide-

bsorptive capacity was determined by the knowleases of the firms. This was not simply a matter ofluster capacity being an aggregation of the individrms’ capacities, since the channels of knowledgeuisition and diffusionbetweenthe firms were also keomponents of the overall cluster absorptive capaowever, the density and structure of those chan

nto and within the cluster, and hence their impn the extent of learning in the cluster, were stronhaped by the knowledge bases of the individual fi

These conclusions align with the distinctionween relational and spatial proximity highlightedmin and Cohendet (2004). However, in several repects they differ from other recent contributions ineld. In particular, although spatially bounded knodge interactions within the cluster were importan

he learning process, instead of being “unstructurednplanned”, arising “by chance” (Malmberg, 2003, pp.

rom that base, several directions of evolution canvisaged. Two seem particularly interesting.

One would be towards a much more pervand less polarised knowledge and learning systereater number of firms would act as net contribu

nto the knowledge system (in particular, with Extertars and Active Mutual Exchangers behaving m

ike Technological Gatekeepers), and the internsolated firms would either connect into the syss knowledge acquirers or exit the industry. The oirection would be towards a system in which exluster sources of knowledge became the domirivers of learning and innovation in an increasinompetitive market, with firms among the Techogical Gatekeepers and Active Mutual Exchangeducing their willingness to act as net ‘sourcesnowledge (i.e. behaving more like the External Stahe cluster knowledge system would then turn ‘ins


out’, exhibiting a different kind of polarised structure.The most technologically advanced and dynamic firmswould concentrate their knowledge links with actorsoutside the spatial cluster, contributing little or nothingto the intra-cluster system. Increasingly isolatedother cluster firms, without their links to sources ofnew knowledge, would fail to compete in growingmarkets.

Both these directions of change would be likelyto enhance the overall growth and competitiveperformance of wine production in the ColchaguaValley. But they would result in two very differentmeanings of ‘cluster’. One would conform to theconventional expectations of cluster analysts: a coher-ent cluster-centred knowledge system would act as apositive influence on the innovative activities of thespatially associated firms. The other would virtuallyeliminate the role of spatial clustering as an influenceon the learning and innovative activities of firms thatwould remain geographically, but not cognitively,associated.

In that context, if the patterns reported here arewidespread, interesting questions arise about policy. Itseems fairly clear that, since learning and innovationare driven primarily by the knowledge bases (absorp-tive capacities) of individual firms, measures designedmerely to foster spatial agglomeration may have lim-ited influence—a view consistent with that ofBreschiand Malerba (2001). Similarly, measures designed tofoster intra-cluster communication and collaborationm lesi dgeb . Inc rms’k ngere tiona

A

avo( .,C rk.T io-v or’ss e-u ts on

an earlier draft. Financial support provided for the firstauthor’s doctoral research by the EU Marie Curie Pro-gramme, the UK Economic Social Research Counciland the Italian Ministry for Education, University andResearch is gratefully acknowledged.

Appendix A. Absorptive capacity

Absorptive capacity has been measured by applyinga Principal Component Analysis to the following fourcorrelated variables:

A.1. Variable 1: Human resources

This variable represents the cognitive backgroundof each firms’ knowledge skilled workers on the oftheir degree of education. According to previous stud-ies regarding returns to education, we assume thatthe higher the degree of education the higher is theircontribution to the economic returns of the firm. Onthis assumption we weight each knowledge skilledworker differently according to the degree attained sothat:

Human resource= 0.8 × degree+ 0.05× master

+0.15× doctorate

A weight of 0.8 has been applied to the number of grad-uate employees in the firm which include also those thatr asest r ofe ve aP

n int ofw aret

As

s thec nedr easti gehM ted

ight not do much to change firms’ cognitive rof those also are shaped primarily by their knowleases as well as strong underlying motivationsontrast, measures focused on strengthening finowledge bases might be expected to lead to stroxtra-cluster links, greater new knowledge creand more intensive intra-cluster diffusion.

cknowledgements

The authors are grateful to Christian Diaz BrS.A.G., Chile) and Marcelo Lorca Navarro (I.N.I.Ahile) for their invaluable help during the fieldwohanks go also to Mario Cimoli, Jorge Katz and Ganni Stumpo for their support during the first authtay at CEPAL and to Simona Iammarino, Aldo Gna and three anonymous referees for commen

eceived higher levels of specialisation. In such che value adds up a further 0.05 times the numbemployees with masters and 0.15 for those that hah.D.Only degrees and higher levels of specialisatio

echnical and scientific fields related to the activityine production (i.e. agronomics, chemistry, etc.)

aken into account.

.2. Variable 2: Months of experience in the wineector

This variable has been included as it representognitive background of each of the abovementioesources in temporal terms. Time is in fact at lndicative of the fact that accumulation of knowledas occurred via ‘learning by doing’ (Arrow, 1962).ore in detail, the variable is the result of a weigh


mean of the months of work of each knowledge skilledworker in the country and abroad:

Months of experience in the sector = 0.4× no.of months (national) + 0.6× no. of months (interna-tional).

To the time spent professionally abroad we at-tributed a higher weight because the diversity of theprofessional environment might stimulate an activelearning behaviour and a steeper learning curve. Thelearning experiences considered are those realised inthe wine industry only.

A.3. Variable 3: Number of firms in which eachknowledge skilled worker has been employed

This variable includes the professional experience inother firms operating in the wine industry. Also in thiscase we weighted differently national and internationalexperiences, giving to the latter a higher weight.

Number of Firms = 0.4× no. of firms (na-tional) + 0.6× no. of firms (international)

A.4. Variable 4: Experimentation

In this case, the level of experimentation at firm levelhas been calculated according to the following scale:

(0) for no experimentation;(1) when some form of experimentation is normally

hea-

( uc--

( ainonerch

( ainmore

or

om-p ptivec

Appendix B. Sociometric measures

B.I. Degree centralitydepends on the links that onenode has with the other nodes of the network. It is asimple measure because it counts the direct ties withother nodes. It can be calculated both for undirectedand directed graphs. In this study, we computed bothIn-degree and Out-degree centrality. In-degree countsthe number of ties incident to the node; Out-degreecentrality the number of ties incident from the node.

CD(ni ) = d(ni )

where d(ni ) is the sum of the nodes adjacent to thatnode.B.II. Actor betweennesscentrality is a measure

of centrality that considers the position of nodes in-between the geodesic (i.e. shortest path) that link anyother node of the network.

Let gjk be the proportion of all geodesics linkingnode j and node k which pass through node i, the be-tweenness of node i is the sum of allgjk where i, j andk are distinct.

CB(ni ) =∑

j<kgjk

(ni )

gjk

This index has a minimum of zero whenni falls on nogeodesics and a maximum which is (g− 1) (g− 2)/2(g= total nodes in the network) which is the number ofpair of nodes not includingni .B.III. Core/periphery modelsare based on the no-

t he-s ctedt ss ofn ohe-s thec pe-r sityr totalt

isa thern

A

eringt ter

carried out but only in one of the activities of tproductive chain (either in viticulture or vinifiction);

2) when is led in at least two activities of the prodtive chain (normally in both viticulture and vinification);

3) when at least two activities of the productive chare marked and the firm has been engaged injoint research project with a university or a resealab in the last 2 years;

4) when at least two activities of the productive chare marked and the firm has been engaged inthan one joint research project with a universitya research lab in the last 2 years.

Principal Component Analysis extracted one conent, which we adopted as a measure of absorapacity.

ion of a two-class partition of nodes, namely, a coive subgraph (the core) in which nodes are conneo each other in some maximal sense and a claodes which are more loosely connected to the cive subgroup but lack any maximal cohesion withore. The analysis sets the density of the core toiphery ties in an ideal structure matrix. The denepresents the number of ties within the group onies possible (Borgatti and Everett, 1999).B.IV. k-coreis a subgraph in which each node

djacent to at least a minimum number k of the oodes in the subgraph.

ppendix C. External openness

External openness has been measured considhe knowledge linkages of firms with extra-clus


sources of knowledge. In the analysis, we have groupedthe linkages into 10 sources and channels of extra-cluster knowledge. The importance of each source forthe transfer of technical knowledge into the firm is mea-sured on a 0–3 scale, where 0 stands for ‘no importance’and 3 for ‘maximum importance’.

References

Alba, R.D., 1973. A graph theoretic definition of a sociometric clique.Journal of Mathematical Sociology 3, 113–126.

Allen, T.J., 1977. Managing the Flow of Technology: TechnologyTransfer and the Dissemination of Technological Informationwithin the R&D Organization. MIT Press, Cambridge, MA.

Altenburg, T., Meyer-Stamer, J., 1999. How to promote clusters:policy experiences from Latin America. World Development 27(9), 1693–1713.

Amin, A., Cohendet, P., 2004. Architectures of Knowledge. Firms,Capabilities and Communities. Oxford University Press, Oxford.

Anderson, K., Norman, D., 2001a. Global wine production, con-sumption and trade: 1961–1999. A statistical compendium. CIESWine Statistical Compendium, Centre for International Eco-nomic Studies, University of Adelaide, Australia.

Anderson, K., Norman, D., 2001b. Globalisation and the worlds’wine markets. CIES Discussion Paper 0143, Centre for Interna-tional Economic Studies, University of Adelaide, Australia.

Arrow, K.J., 1962. The economic implications of learning by doing.Review of Economic Studies 29, 155–173.

Asheim, B.T., 1994. Industrial districts, inter-firm cooperation andendogenous technological development: the experience of de-veloped countries. In: UNCTAD, Technological Dynamism inIndustrial Districts: An Alternative Approach to Industrializa-tion in Developing Countries?, United Nations, New Yorkand

A gra-iew

B phi-18,

B n thean,

s in

B bale.

B icalorld

B icalternsh to

B truc-

Borgatti, S.P., Everett, M.G., Freeman, L.C., 2002. Ucinet 6 for Win-dows. Analytic Technologies, Harvard.

Breschi, S., Lissoni, F., 2001. Knowledge spillovers and local innova-tion systems: a critical survey. Industrial and Corporate Change10 (4), 975–1005.

Breschi, S., Malerba, F., 2001. The geography of innovation and eco-nomic clustering: some introductory notes. Industrial and Cor-porate Change 10 (4), 817–833.

Brown, J.S., Duguid, P., 1991. Organizational learning andcommunities-of-practice: Toward a unified view of working,learning and innovation. Organization Science 2 (1), 40–57.

Camagni, R., 1991. Local milieu, uncertainty and innovation net-works: towards a new dynamic theory of economic space. In:Camagni, R. (Ed.), Innovation Networks: Spatial Perspectives.Belhaven Press, London, pp. 121–142.

Capello, R., 1999. Spatial transfer of knowledge in high technologymilieux: learning versus collective learning processes. RegionalStudies 33 (4), 353–365.

Cantwell, J., Iammarino, S., 2003. Multinational Corporations andEuropean Regional Systems of Innovation. Routledge, London,New York.

Carter, A.P., 1989. Know how trading as economic exchange. Re-search Policy 18, 155–163.

Cassiolato, J., Lastres, H., Maciel, M., 2003. Systems of Innovationand Development: Evidence from Brazil. Edward Elgar, Chel-tenham.

Cohen, W.M., Levinthal, D.A., 1989. Innovation and learning: thetwo faces of R&D. The Economic Journal 99, 569–596.

Cohen, W.M., Levinthal, D.A., 1990. Absorptive capacity: a newperspective on learning and innovation. Administrative ScienceQuarterly 35, 128–153.

Coleman, J., 1990. Foundations of Social Theory. Harvard UniversityPress, Cambridge, MA.

Creplet, F., Dupouet, O., Kern, F., Mehmanpazir, B., Munier, F.,firms.

C anderi-

D hoy.

D nal

D terns7.

D .G zione

prese

G ap-Pre-inore

G f re-, The

Geneva, pp. 91–142.udretsch, D., Feldman, M.P., 1996. R&D spillovers and the geo

phy of innovation and production. American Economic Rev86 (3), 630–640.

aptista, R., 2000. Do innovation diffuse faster within geogracal clusters? International Journal of Industrial Organization515–535.

ecattini, G., 1989. Sectors and/or districts: some remarks oconceptual foundations of industrial economics. In: GoodmE., Bamford, J. (Eds.), Small Firms and Industrial DistrictItaly. Routledge, pp. 123–135.

ecattini, G., Rullani, E., 1993. Sistema locale e mercato gloEconomia e Politica Industriale 80, 25–49.

ell, M., Albu, M., 1999. Knowledge systems and technologdynamism in industrial clusters in developing countries. WDevelopment 27 (9), 1715–1734.

elussi, F., Gottardi, G., 2000. Models of localised technologchange. In: Belussi, F., Gottardi, G. (Eds.), Evolutionary Patof Local Industrial Systems. Towards a Cognitive Approacthe Industrial District. Ashgate Publishing Ltd., Aldershot.

orgatti, S.P., Everett, M.G., 1999. Models of core/periphery stures. Social Networks 21, 375–395.

2001. Consultants and experts in management consultingResearch Policy 30 (9), 1517–1535.

rowley, W.K., 2000. Chile’s wine industry: historical characterchanging geography. In: Yearbook, Conference of Latin Amcanist Geographers, pp. 87–101.

el Pozo, J., 1998. Historia del Vino Cileno. Desde 1850 hastaEditorial Universitaria, Santiago de Chile.

icken, P., Malmberg, A., 2001. Firms in territories: a relatioperspective. Economic Geography 77 (4), 345–363.

osi, G., 1997. Opportunities, incentives and the collective patof technical change. The Economic Journal 107, 1530–154

rucker, P., 1993. Post Capitalist Society. Butterworth, Oxfordambardella, A., 1993. Innovazioni tecnologiche e accumula

delle conoscenze: quale modello per le piccole e medie imnegli anni ’90? Piccola Impresa/Small Business 2, 73–89.

iuliani, E., 2002. Cluster absorptive capability: an evolutionaryproach for industrial clusters in developing countries. Papersented at the DRUID Summer Conference, Copenhagen/Els6–8 June.

rabher, G., 1993. The weakness of strong ties: the lock-in ogional development in the Ruhr area. In: Grabher, G. (Ed.)Embedded Firm. Routledge, London, pp. 1–32.


Guerrieri, P., Iammarino, S., Pietrobelli, C., 2001. The Global Chal-lenge to Industrial Districts. Edward Elgar, Cheltenham.

Haas, P.M., 1992. Introduction: epistemic communities and inter-national policy coordination. International Organization 46 (1),1–35.

Humphrey, J., Schmitz, H., 1996. The triple C approach tolocal industrial policy. World Development 24 (12), 1859–1877.

Lane, P.J., Lubatkin, M., 1998. Relative absorptive capacity andinterorganizational learning. Strategic Management Journal 19,461–477.

Lissoni, F., 2001. Knowledge codification and the geography of in-novation: the case of Brescia mechanical cluster. Research Policy30, 1479–1500.

Malmberg, A., 2003. Beyond the cluster—local milieus and globaleconomic connections. In: Peck, J., Yeung, H.W.C. (Eds.),Remaking the Global Economy. Sage, London, pp. 145–159.

Malmberg, A., Maskell, P., 2002. The elusive concept of localizationeconomies: towards a knowledge-based theory of spatial cluster-ing. Environment and Planning A 34 (3), 429–449.

Markusen, A., 1996. Sticky places in slippery space: a typology ofindustrial districts. Economic Geography 72 (2), 294–314.

Marshall, A., 1919. Industry and Trade. Macmillan, London.Marshall, A., 1920. Principles of Economics. Macmillan, London.Maskell, P., Malmberg, A., 1999. Localised learning and indus-

trial competitiveness. Cambridge Journal of Economics 23 (2),167–186.

McPherson, M., Smith-Lovin, L., Cook, J.M., 2001. Birds of aFeather: Homophily in Social Networks. Annual Review of So-ciology 27, 415–444.

Nadvi, K., Schmitz, H., 1999. Clustering and industrialization: in-troduction. World Development 27 (9), 1503–1514.

Owen-Smith, J., Powell, W.W., 2004. Knowledge networks as chan-nels and conduits: The effects of spillovers in the Boston Biotech-nology Community. Organization Science 15 (1), 5–21.

Paul, H.W., 2002. Science, Vine and Wine in Modern France. Cam-bridge University Press, Cambridge, MA.

Porter, M., 1998. On Competition. Harvard Business School Press,Boston, MA.

Rabellotti, R., 1999. Recovery of a Mexican cluster: devaluationbonanza or collective efficiency? World Development 27 (9),1571–1585.

Rogers, E.M., 1983. Diffusion of Innovations, third ed. The FreePress, New York.

S.A.G., 2000. Catastro Viticola Nacional. Servicio Agricola yGanadero. Departamento de Proteccion Agricola, Santiago deChile.

Saxenian, A.L., 1994. Regional Advantage. Culture and Competitionin Silicon Valley and Route, vol. 128. Harvard University Press,Cambridge, MA.

Schachner, M., 2002. The Next Napa? Chile’s Colchagua Valley isAiming for Big Time. Wine Enthusiast, March, Elmsford, NewYork.

Schrader, S., 1991. Informal technology transfer between firms:cooperation through informal trading. Research Policy 10,153–171.

Tapia, P., 2001. Los Vinos de Colchagua. PMC Pinnacle Worldwidey Ed. Planetavino, Santiago de Chile.

Von Hippel, E., 1987. Cooperation between rivals: informal know-how trading. Research Policy 16, 291–302.

Wasserman, S., Faust, K., 1994. Social Network Analysis. Methodsand Applications. Cambridge University Press, Cambridge, MA.

Wenger, E.C., Snyder, W.M., 2000. Communities of prac-tice: the organizational frontier. Harvard Business Review,January–February, 139–145.

The micro-determinants of meso-level learning and ...

Documents

Transcript of The micro-determinants of meso-level learning and ...