Technovation 31 (2011) 87–104

Contents lists available at ScienceDirect

Technovation

0166-49

doi:10.1

n Corr

E-m

journal homepage: www.elsevier.com/locate/technovation

Patterns of technological learning within the knowledge systems of industrialclusters in emerging economies: Evidence from China

Bin Guo n, Jing-Jing Guo

School of Management, Zhejiang University, Hangzhou 310058, China

a r t i c l e i n f o

Available online 18 November 2010

Keywords:

Technological learning

Knowledge system

Industrial cluster

Knowledge transmission mechanism

Knowledge spanner

Cognitive subgroup

72/$ - see front matter & 2010 Elsevier Ltd. A

016/j.technovation.2010.10.006

esponding author. Fax: +86 571 88206827.

ail address: guob@zju.edu.cn (B. Guo).

a b s t r a c t

Through an interview-based exploratory study and a follow-up survey-based quantitative analysis, this

paper investigates the technological learning pattern in terms of structure and mechanisms of interaction

within the knowledge system of two industrial clusters in China. Unlike the recent studies that suggest

that industrial cluster comprises disconnected leader-centered communities, we argue that the different

leader-centered communities within the knowledge systems of industrial clusters are not disconnected

from each other. Instead, those communities are inter-connected through the so-called ‘knowledge

spanning mechanisms’. Regarding the interaction dimension of technological learning pattern, this paper

argues that in analyzing learning behavior in the knowledge networks of industrial clusters, it is necessary

to synthesize the learning opportunity perspective and the absorptive capacity perspective to better

understand and explain the similarities and dissimilarities in technological learning behavior among

different cluster types, across cognitive subgroups, and between product innovation and process

innovation. Our study reveals that in the context of emerging countries, the following four factors are

decisive for technological learning opportunities inside the knowledge networks of industrial clusters:

the underlying complexity of technology in clusters, the inter-connectedness between product and

process, path dependency in knowledge searching, and the incremental nature of a cluster’s technological

development.

& 2010 Elsevier Ltd. All rights reserved.

1. Introduction

A cluster is a form of network that occurs within a geographiclocation, in which the proximity of firms and institutions ensurescertain forms of commonality and increases the frequency andimpact of interactions (Porter, 1998, p. 226). During the past twodecades, industrial clusters and their evolution have drawn lots ofattention from both academic and policy arenas. One importantreason for this is the fact that the emergence and roles of industrialclusters as an important industrial organization in the economicsystem have significantly challenged and changed the traditionalrules of industry competition (Porter, 1990; Giuliani, 2007). Undersuch circumstances, policy-makers do not focus solely on largeenterprises in building national capabilities, and geographicalclusters of firms are seen as drivers of national competitivenessand economic growth. Therefore, the question of how to promotethe formation, development and upgrading of industrial clustershas been emphasized in policy-making for regional developmentaround the world since the 1990s.

ll rights reserved.

After the market reformation in 1979, industrial clustersdeveloped very rapidly in both number and scale in China. Clustershave been a significant component of the provincial economies ofcoastal China (Kang, 2007), which is one of the most importantcontributors to the higher economic growth rate in the easterncoastal regions of China, as compared with that of their inlandcounterparts (Zhang et al., 2004). Zhejiang province, one of themost prosperous coastal regions in China and the host area of thesample clusters in the present study, is home to many industrialclusters ranging from labor-intensive (e.g., socks, neckties, andcigarette lighters) (Hessler, 2007) to capital-intensive products(e.g., metalworking products, electric parts, and automobile parts)(Marukawa, 2006). One interesting fact is that most firms are small,except for a few core firms in these industrial clusters. However,when these small firms agglomerate, they have achieved strongcompetitive strength and market presence. For instance, Wenzhoumakes 70 percent of the world’s cigarette lighters. Forty percent ofthe world’s neckties are made in Shengzhou (Hessler, 2007).Datang, the so-called Socks City of China, produces an astoundingnine billion pairs of socks each year (Barboza, 2004). As described ina report in Los Angeles Times, ‘‘China’s advantages in the global

marketplace are moving well beyond cheap equipment, material and

labor. The country also exploits something called clusteringy.China

has created giant industrial districts in distinctive entrepreneurial

B. Guo, J.-J. Guo / Technovation 31 (2011) 87–10488

enclaves such as Datang. Each was built to specialize in making just one

thingy’’ (Lee, 2005).Among studies on industrial clusters, one characteristic trend is

that the knowledge-based perspective is widely used in analyzinglearning and innovation behaviors therein. As suggested in Baptistaand Swann (1998, p. 538), one of the main reasons behind theexistence and success of clusters is the pervasiveness of knowledgeexternalities or spillovers. The knowledge and learning processes ofthe main actors are key elements to understanding the rise, growthand transformation of a cluster (Breschi and Malerba, 2001).Furthermore, the crucial role of knowledge and learning can beclearly demonstrated by the various definitions of an industrialcluster in the literature. Besides Porter (1998), a few otherresearchers have also expressed similar ideas. For instance, indus-trial clusters are defined in Giuliani and Bell (2005, p. 47) as‘‘geographic agglomerations of economic activities that operate inthe same or inter-connect sectors’’; Morosini (2004, p. 307) statedthat ‘‘an industrial cluster is a socioeconomic entity characterizedby a social community of people and a population of economicagents localized in close proximity in a specific geographic region.Within an industrial cluster, a significant part of both the socialcommunity and the economic agents work together in economic-ally linked activities, sharing and nurturing a common stock ofproduct, technology and organizational knowledge in order togenerate superior products and services in the marketplace’’.Similarly, an industry cluster is defined by Rosenfeld (1997,p. 10) as ‘‘a geographically bounded concentration of similar,related or complementary businesses, with active channels forbusiness transactions, communications and dialogue, that sharespecialized infrastructure, labor markets and services, and that arefaced with common opportunities and threats’’. Although thesedefinitions have some slight differences, they share three commonpoints. Firstly, all of them posit that geographic proximity andeconomical linkages among cluster firms are the basic character-istics of industrial clusters. Secondly, they stress that individualfirms in clusters have certain forms of commonality such as accessto specialized factors, a supply of intermediate products, infra-structures and cultural embeddedness. Among those forms ofcommonality, one of the most important elements is a commonstock of knowledge (e.g., the knowledge embedded in the pooling ofspecialized workers) that is created and shared by firms inside acluster. Thirdly, they all argue that firms in clusters have frequentinteractions, which are mainly reflected in the acquisition ofknowledge, as well as in sharing, diffusing and creating it. A hostof linkages among cluster members results in a whole greater thanthe sum of its parts (Porter, 1998, p. 81). As a result, learningthrough networking and by interacting is seen as the crucial forcepulling firms into clusters and the essential ingredient for the on-going success of an innovative cluster (Breschi and Malerba, 2001).

Generally speaking, the knowledge perspective literature onindustrial clusters can be categorized into two strands. One is theMarshallian perspective, and the other is the localized knowledgespillovers (LKS) perspective (Breschi and Malerba, 2001; Maskell,2001b; Giuliani, 2007). As for the Marshallian perspective, theresearch focus in past literature was heavily on transaction-basedproduction systems, instead of learning-based knowledge systemsin industrial clusters. Besides, they usually hypothesized that‘knowledge in the air’ is pervasively distributed and freely shared.Local firms are generally assumed to be more willing to shareknowledge with others because common norms and values haveprevented cheating and opportunistic behavior (Harrison, 1992).By contrast, the LKS Perspective asserted that knowledge systemsand production systems obviously overlap, but that they are notidentical (Bell and Albu, 1999, p. 1723). In order to solve theinherent ambiguity of the concept of localized knowledge spil-lovers, which to date is considered by many as a ‘black box’ (Breschi

and Lissoni, 2001), the literature also emphasized the need to placefirm-level learning at the center of cluster analyses with theobjective of understanding the nature and characteristics of acluster’s innovative process (Bell and Albu, 1999; Maskell, 2001a;Martin and Sunley, 2003; Giuliani, 2005, 2007). In particular, someof the recent cluster studies have emphasized that knowledge is notdiffused evenly ‘in the air’ (Power and Lundmark, 2004; Giulianiand Bell, 2005). Rather, innovation-related knowledge is diffused inclusters in a highly selective and uneven way (Lissoni, 2001;Morrison, 2004; Giuliani, 2005, 2007; Boschma and ter Wal, 2007).

Undoubtedly, these recent studies are helpful in opening up theabove-mentioned ‘black box’. However, much effort is still requiredin this direction. First of all, two crucial drawbacks exist in theexisting studies. On the one hand, when attention has turned totechnological change in clusters in developing countries, thesematerials-centered structures and flows have usually remained atthe center of the analysis (Bell and Albu, 1999). However, asindicated earlier, we need to distinguish knowledge systems fromthe associated ‘production systems’ that comprise materials-cen-tered systems of production and trade. On the other hand, amongstudies on learning and innovation in clusters, most are aboutclusters in developed countries. In industrialized countries, clus-tering often occurs in high-tech (e.g., science parks) or design-intensive branches and involves substantial product and processinnovations (Altenburg and Meyer-Stamer, 1999). It should bepointed out that the manufacturing clusters in developing coun-tries have some different characteristics as compared to those indeveloped countries. For manufacturing clusters in developingcountries, innovation strategies depend more on imitation than oninnovation, and they are more market-led than technology-driven.Besides, most of the knowledge in clusters is not concerned withcore technology and research activities but instead has to do withknow-how and skills in the more down-stream phase of innova-tion. Therefore, the present study attempts to gain some insightinto the knowledge system by empirically examining the roles andinfluencing mechanism of technological learning and knowledgespillover in manufacturing clusters.

Another more important problem is that some recent studiesfrom the LKS perspective have adopted a single-mechanismanalysis while ignoring the potential complementary and/or sub-stitution relationship among different knowledge transmissionmechanisms. We argue that it is necessary to bring local and non-local, formal and informal, knowledge transfer and knowledgespillover, and personal and impersonal knowledge acquisitionchannels into our research on knowledge network and technolo-gical learning in industrial clusters. Furthermore, most importantof all is that the big picture of the structure of knowledge systemsinside clusters in the past literature (e.g., Lissoni, 2001; Morrison,2004; Giuliani, 2005; Boschma and ter Wal, 2007) is described ascomprising a few disconnected leader-centered communities, inwhich the leading producers act as gatekeepers for their ownnetwork. We argue that different leader-centered communitieswithin clusters are not disconnected. Instead, these communitiesare inter-connected through what are called ‘knowledge spanning

mechanisms’. We attempt to provide some empirical evidence forthe existence of such knowledge spanning mechanisms in thepresent study.

The goal of the present study is to investigate the technologicallearning pattern in terms of structure and mechanisms of interac-tion within the knowledge system of industrial cluster through aninterview-based exploratory study and a follow-up survey-basedquantitative analysis of two industrial clusters in Zhejiang pro-vince, China. Specifically, in this paper, we define a producer-

centered community as a group of associated units (e.g., suppliers–producers–clients) along the supply chain that formed throughproduction and trade relationships in the production network of an

B. Guo, J.-J. Guo / Technovation 31 (2011) 87–104 89

industrial cluster. Meanwhile, cognitive subgroups are defined asgroups of firms with similar knowledge transmission mechanismsand a similar position in the knowledge network of an industrialcluster. With regard to the structure dimension of the technologicallearning pattern, we strive to unfold the existence, roles andcharacteristics of different cognitive subgroups inside clusterknowledge networks. Our argument is that the leader-centeredsubgroups are not disconnected from each other, and that knowl-edge spanning mechanisms play an essential role in connectingthem as an integrated knowledge system. As to the interactiondimension of the technological learning pattern, we will explorethe dissimilarities in knowledge transmission mechanisms acrossdifferent cognitive subgroups, as well as in different types ofindustrial clusters. We expect that such differences in knowledgetransmission mechanisms depend not only on the learning cap-abilities of cluster firms but also on the learning opportunity facedby them. The paper is organized as follows. In Section 2, we outlinethe theoretical framework and define the research hypotheses ofthis study. Section 3 explains the methodology applied in thisresearch. Section 4 presents the preliminary evidence by introdu-cing an exploratory study on a cooling tower cluster. After that, theempirical results with regard to the research hypotheses arepresented and discussed in Section 5. Finally, a concluding discus-sion is provided for the findings, theoretical implications, limita-tions and directions for future research.

2. Literature review and research hypotheses

2.1. The channels of knowledge acquisition

In clusters, agglomerations of firms in related industries providea pool of technical knowledge and expertise and a potential base ofsuppliers and users of innovations (Maskell, 2001b; Porter, 1998).The effective transmission and sharing of knowledge withinclusters is vital to the formation of competitive advantage at boththe firm level and the cluster level (Malecki, 2010). These networksof producers, suppliers and users, through which firms can con-stantly search for external sources of knowledge in order todiversify and broaden their knowledge base (Li and Tang, 2010),play an especially important role when technological knowledge isinformal or tacit in nature (Feldman and Florida, 1994, p. 220).Nonetheless, in traditional industrial district literature, local firmsare generally assumed to be immersed in the ‘industrial atmo-sphere’ and are more willing to share knowledge with othersbecause of common values and relational embeddedness(Harrison, 1992). Thus, the knowledge externalities in a clusterare ‘in the air’ (Marshall, 1920, p. 225) and available to firms locatedinside automatically and naturally. Consequently, little attentionhas been paid to the fact that the opportunity to access externalknowledge and the capacity to acquire and utilize that knowledgeare divergent.

During the last decade, a large body of literature began to takeinto account the influence of the uneven distribution and selectivetransmission of knowledge on innovation networks of clusters.Lazerson and Lorenzoni (1999) criticize the tacit assumptionthroughout most of the literature that all district firms arerelatively homogeneous. Giuliani and Bell (2005) find that knowl-edge is not diffused evenly ‘in the air’ but instead flows within acore group of firms characterized by advanced absorptive capa-cities. Similarly, Giuliani (2007) reveals that innovation-relatedknowledge is diffused in clusters in a highly selective and unevenway, and that this pattern is found to be related to the hetero-geneous and asymmetric distribution of firm knowledge bases inthe clusters. A substantial number of studies have suggested that tofully understand the process of localized learning and innovation,

one needs to place firm-level learning at the center of clusteranalyses with the objective of understanding how firm-level andcluster-level learning processes interact (Bell and Albu, 1999;Maskell, 2001a; Martin and Sunley, 2003; Giuliani, 2007;Boschma and ter Wal, 2007).

There are numerous ways for knowledge to flow in clusters.Clustering enables firms to benefit from a ‘collective learningprocess’, operating ‘‘through skilled labor mobility within the locallabor market, customer–supplier technical and organizationalinterchange, imitation processes .y and informal ‘cafeteria’effects’’ (Camagni, 1991, p. 130). As a whole, the previous studieshave stressed the importance of labor mobility and spin-off of newfirms for the transfer of knowledge and local know-how (Capelloand Faggian, 2005; Giuliani, 2007). Labor mobility is likely to speedup knowledge dissemination by creating links between firms,workplaces and institutions and thus to create new combinationsof knowledge embodied in people (Power and Lundmark, 2004,p. 1027). In addition, the previous studies emphasize the crucialinfluence of external linkages (especially the linkages with localsuppliers, key clients, universities and research institutions). Forexample, Keeble and Wilkinson (1999) indicate that the mainmechanisms for knowledge transmission in industrial clustersinclude the linkages between suppliers and customers and themakers and users of capital equipment, and collaborative relation-ships between firms. Zeng et al. (2010) also show that the verticaland horizontal cooperation with customers, suppliers and otherfirms plays a distinct role in the innovation process of Chinesesmall- and medium-size enterprises. Similarly, as discussed inNadvi (1996), US consultants played a major role in enabling firmsin the cluster to improve their product quality in the Sialkotsurgical instrument cluster.

In shaping a generalized pattern of knowledge sharing andtransmission within clusters, however, some problems still remainunsolved in the literature. As discussed in Giuliani and Bell (2005)and Boschma and ter Wal (2007), the existing studies have stressedthe importance of intra-cluster knowledge sharing while greatlyunderestimating the effects of extra-cluster knowledge on theintra-cluster knowledge system. As a matter of fact, the ‘openness’of cluster knowledge systems and their capacity to inter-connectwith extra-cluster sources of knowledge seem especially importantin technologically lagging regions and industries (Bell and Albu,1999). Moreover, knowledge transmission mechanisms include theintentional type (i.e., knowledge transfer) and the unintentionalone (i.e., knowledge spillover). As for the unintentional type ofknowledge transmission, which refers to the effect of researchperformed in one economic unit on improving technology in asecond economic unit without the latter’s having to pay for it(Griliches, 1979, 1992), is one of the main reasons behind theexistence and success of clusters (Baptista and Swann, 1998). Suchknowledge flows due to spillovers can be the outcome of economictransactions, free-sharing agreements or some agents’ failure toappropriate the outcome of their own innovation efforts (Breschiand Lissoni, 2001). Generally speaking, most of the existingliterature pays a great deal of attention to intentional and/orformal mechanisms while to some extent ignoring the contributionof unintentional and/or informal mechanisms. One of the fewexceptions is an empirical study by Dahl and Pedersen (2004). Witha sample of engineers in a regional cluster of wireless commu-nication firms in Northern Denmark, they find that informalcontacts represent an important channel of knowledge diffusion.Moreover, the influence of informal contacts within industrialclusters in emerging economies is not clear yet. Taking this intoaccount, the present study will investigate knowledge transmis-sion mechanisms in terms of formal and informal, intentional andunintentional, internal and external, and vertical (between clientsand suppliers) and horizontal (among rival firms) at the same time.

B. Guo, J.-J. Guo / Technovation 31 (2011) 87–10490

2.2. The structure of cluster knowledge system

Combinations of internal change-generating resources withexternal knowledge resources, and the links between them, havebeen described as industrial ‘‘knowledge systems’’ (Bell and Albu,1999). Knowledge systems are different from production systems,as the latter refer to networks of autonomous but interdependentproducers (Saxenian, 1991) and include the product designs,materials, machines, labor inputs, and transaction linkagesinvolved in the production of goods (Bell and Albu, 1999). Belland Albu (1999) argue that knowledge systems are not identical toproduction systems and stress that, although they may interactwith each other, their interactions are highly variable and poorlyunderstood. In addition, some recent studies have noticed theinfluence of knowledge base on the structure of knowledge system(Cantner et al., 2010; Giuliani, 2005). For example, based on acomparison of three rather different regional innovator networks,Cantner et al. (2010) suggests that regions with a strong knowledgebase that are specialized in broad technology fields tend to haverelatively fragmented network structures.

Several recent studies have argued that an industrial cluster iscomposed of a few leader-centered networks (or say, communities)that are probably more disconnected from each other and in whichleading firms will behave as knowledge gatekeepers for their owncommunities (Lissoni, 2001; Morrison, 2004; Giuliani, 2005;Boschma and ter Wal, 2007). For instance, Lissoni (2001) investi-gates a Brescia mechanical cluster in Italy and concluded thatknowledge, rather than flowing freely within the cluster bound-aries, circulates within a few smaller ‘epistemic communities’, eachcenters around the mechanical engineers of a single machineproducer in the district and involves a selected number of suppliersand customers’ technicians. It should be pointed out that apotential problem in Lissoni (2001) is that it reached the aboveconclusion by only investigating two types of knowledge transmis-sion mechanisms in terms of interpersonal contact and labormobility, whereas other mechanisms such as impersonal channelsor inter-community linkages were not explored. Extending thefocus on interpersonal contacts by asking ‘‘technicians from theleading firms whether they contacted and asked for or gavetechnical advice to colleagues of other local firms’’, Morrison(2004) asserts that leading firms absorb external knowledge andspread it only to their own network of clients and providers, thusarguing that leading firms cannot be interpreted as knowledgegatekeepers. Similarly, Morrison (2004) ignores the fact that thereare a wide variety of knowledge transmission mechanisms inclusters and that one mechanism may share a complementaryand/or substitution relationship with another. Another study byGiuliani (2005) focuses only on inter-firm horizontal relationships,whereas vertical linkages are not explored. Through analyzing thetechnical advice and support derived from interpersonal andformal linkages, Giuliani (2005) states that in spite of pervasivebusiness interactions, inter-firm knowledge flows are strikinglylimited to cohesive subgroups of firms. Similarly, Boschma and terWal (2007) limit their attention to intentional knowledge transferfrom technical support and market knowledge provided by thefirm, and those with whom the firm was involved in researchcollaboration, while ignoring the influence of unintentional knowl-edge transmission. This treatment could lead to a biased conclu-sion. In fact, when compared with knowledge transfer throughformal contracts and relationships, knowledge spillover is moreessential to knowledge sharing and creation in the knowledgenetworks of industrial clusters.

In summary, it must be pointed out that the above studiesexemplify the following limitations when conceptualizing thetechnological learning pattern inside the knowledge networks ofindustrial clusters. First, the effectiveness of transmission for

different knowledge acquisition channels might be varied underdifferent contexts of knowledge transmission. The above studieshave neglected the fact that there co-exist multiple learningmechanisms with potential complementary and/or substitutioninterrelationships. Second, with respect to clusters in developingcountries, in addition to interpersonal ties, hiring and imitation areoften important to knowledge sharing, especially in transmittingtacit knowledge unintentionally (Bell and Albu, 1999; Schmitz andNadvi, 1999). Keeping these in mind, we attempt to examine andvalidate the question of whether the leader-centered communitieswithin the knowledge system of two industrial clusters in China arereally disconnected from each other. If not, what mechanismswould make them interconnect and interact? Actually, if thosesubgroups or communities in clusters are really disconnected fromeach other, geographical proximity will be meaningless to innova-tion networks (Maskell, 2001b). As a consequence, the firmsubgroups inside clusters will almost turn into traditional lea-der-centered production networks resembling the Fordist model.

Proposition 1. For manufacturing clusters in emerging economies,

there exist multiple connecting mechanisms in bridging different

leader-centered communities within the cluster knowledge system

in terms of technological gatekeepers and knowledge spanners. The

former acts as an intra-community connecting mechanism, whereas

the latter is an inter-community connecting mechanism. Through these

connecting or bridging mechanisms, different leader-centered com-

munities are combined into a partially inter-connected knowledge

system instead of an isolated one with disconnected communities.

2.3. Learning mechanisms in different cognitive subgroups

Knowledge gatekeepers have been a particular focus of existingstudies. The gatekeeper plays an active role in screening andacquiring external sources of knowledge that appear to be appro-priate for the firm and then for the cluster’s members, transcodingexternal knowledge to make meaningful complex knowledge, andtransferring and disseminating internal accumulated knowledge totheir partner firms (Morrison, 2004, p. 8). Furthermore, the open-ness of cluster knowledge systems and their capacity to inter-connect with extra-cluster sources of knowledge seems especiallyimportant in technologically lagging regions, industries or coun-tries (Bell and Albu, 1999; Giuliani and Bell, 2005). However, itshould be pointed out that the existing studies are too muchconcerned with the roles of leading firms as gatekeepers in theirown networks, while neglecting the nature and effects of inter-linkages among different leader-centered communities. Addition-ally, they generally pay a good deal of attention to the influences ofinterpersonal ties and formal exchanges, but much less attention toother knowledge transmission mechanisms that might be evenmore important. In this paper, one of our research foci is thequestion of what are the specific differences in knowledge trans-mission mechanisms across different cognitive subgroups in acluster.

In addition, Lazerson and Lorenzoni (1999) criticized the view ofindustrial districts as an undifferentiated community of smallfirms. In the knowledge network of an industry cluster, firmsubgroups with diverse learning patterns are different in theirpositions within the network and in their ways of acquiringknowledge from external linkages. For instance, in Giuliani andBell (2005), different cognitive roles played by cluster firms wereanalyzed in terms of the number and nature of intra-cluster andextra-cluster knowledge linkages. Five types of roles were identi-fied: those of the technological gatekeepers, active or weak mutualexchangers, external stars and isolated firms.

Much external technological knowledge is acquired and diffusedinto clusters through leading firms as technological gatekeepers in

B. Guo, J.-J. Guo / Technovation 31 (2011) 87–104 91

clusters because of their large number of technical linkages withsources outside the clusters. Thereby, in line with the view ofabsorptive capacity, the gatekeepers must possess sufficient absorp-tive capacities to absorb external knowledge effectively. This meansthat they need to carry out in-house R&D activities consistently.Although the previous research suggested that these gatekeepersneed not be the largest-sized in the clusters, firms of this type areusually the largest in manufacturing clusters of Zhejiang province,China, due to the fact that sustaining internal R&D activities requiressufficient financial resources. Consequently, we assert that large firmsin industrial clusters are more dependent on in-house R&D inacquiring technical knowledge. By contrast, small-sized firms,accounting for the largest portion of firms in industrial clusters, findit difficult to obtain enough financial resources because of thecommon problem of financial constraints in developing countries.Thus, their short-term orientations are much stronger in R&Dinvestments. In general, they seldom carry out active in-house R&Dand rely heavily on the copying and imitation of existing products onthe market, so as to reduce both their market risk and the cost ofproduct development. As for those firms that are medium-sized, ascompared with small firms, they usually have stronger incentives andmore financial resources to do internal R&D with an aim of facilitatingthe effective absorption of external technologies. On the other hand,unlike leading firms that rely more on internal R&D, the medium-sized firms are generally inclined to use multiple knowledge acquisi-tion channels to acquire external technological knowledge, in order tomaintain flexibility in technical strategies and to reduce the cost oftechnology acquisition. This inclination indicates that their knowl-edge searching strategies are more likely to be breadth-first ratherthan depth-first.

Proposition 2. Differences in selecting knowledge transmission

mechanisms exist across producer groups (or, say, cognitive sub-

groups) inside clusters in emerging economies. In acquiring external

knowledge, large-sized firms as cluster technological gatekeepers rely

more on in-house R&D, whereas small-sized firms are inclined to

imitation. As for medium-sized firms, they tend to use more diverse

channels in knowledge acquisition than do small or large firms.

2.4. Learning mechanisms: vertical- vs. horizontal-type clusters

Clusters can be categorized according to two types, the hor-izontal and vertical. Leaving aside their spatial structure, industrialclusters have usually been defined in terms of the materials thatthey use and the goods that they produce. For instance, ‘horizontal’clusters are typically defined by the similarity of the firms’products; in ‘vertically’ linked clusters, the rows of materials andgoods constitute the key linkages (Bell and Albu, 1999, p. 1722).However, for either vertical-type or horizontal-type clusters, thereare both horizontal linkages and vertical linkages in a given cluster.The horizontal dimension of the cluster consists of firms withsimilar capabilities that carry out similar activities, and the verticaldimension is composed of firms with dissimilar but complemen-tary capabilities that carry out complementary activities (Maskell,2001b, p. 927).

With regard to horizontal-type clusters, the goods produced andthe technology used are highly similar between producers. There-fore, direct learning opportunities among producers will be limitedbecause of the lesser amount of heterogeneity in terms of technol-ogy and expertise, which makes the producers more likely toacquire external knowledge through vertical linkages. In contrast,for vertical-type clusters, the typical structures of the productionand knowledge systems are composed of several leader-centeredcommunities inside which producers, suppliers and customers areconnected vertically. Almost all of the leaders are large in size whencompared with other firms in the same cluster, and they usually

play the role of knowledge providers in their communities. Most ofthe knowledge accessible inside the communities is directly relatedto technical problem solving in regular production (e.g., qualityproblem solving, small technology improvements of products, andincremental improvements of the manufacturing process toenhance efficiency). Producers mainly acquire knowledge aboutnew product ideas and process adaptation triggered by productinnovation and major process restructuring through informal andunintentional channels in horizontal linkages.

In addition, for horizontal-type clusters, on the one hand,because of the high similarity in products and technology acrossproducers, most of the producers’ learning opportunities comefrom improvements in the manufacturing process rather than inthe products, thus promoting the likelihood of adopting verticallinkages and increasing the degree to which they do so. On theother hand, knowledge embedded in process improvements ishighly tacit; hence, strong vertical linkages (among producers,suppliers and customers) are more efficient in transferring this kindof knowledge, as demonstrated by the case of Brescia machineproducers in Italy (Lissoni, 2001). Nonetheless, for vertical-typeclusters, there exist more improvement opportunities in productsdue to the technological complexity of products. As a result, thoseproducers that are competitors or potential competitors in themarket will have more opportunities and incentives to learn fromeach other horizontally.

Proposition 3. Vertical linkages and horizontal linkages have varying

degrees of influence on the effectiveness of technological knowledge

acquisition in different types of clusters in emerging economies.

Vertical-type clusters are more prone to use horizontal linkages in

product and process innovation. By contrast, horizontal-type clusters

are inclined to rely on vertical linkages.

2.5. Learning mechanisms: product innovation vs. process innovation

In manufacturing clusters in China, most firms are relativelysmall in size except for few leading firms inside the clusters. Thesesmall-sized firms devote great attention to incremental improve-ments in the manufacturing process. As mentioned in Maskell(2001b), besides internal development and pre-competitiveresearch through university, etc., at least as important is theinvestment in incremental ‘low-tech’ learning and innovation thattakes place when firms in even fairly traditional industries handleand develop mundane day-to-day operations. The cost-cuttingprocess improvements and low-cost adaptation and replication ofmachinery allow small firms to upgrade their production processesat a relatively rapid rate and low cost (Bell and Albu, 1999).

Generally speaking, knowledge regarding process innovationthat spills over is mainly ‘tacit’, i.e., highly contextual and difficultto codify, and it is therefore more easily transmitted through face-to-face contact and personal relationships. In clusters, productinnovation and improvements are ultimately embodied in themarket explicitly. Thus, impersonal learning mechanisms such asimitation and reverse engineering can be effective in knowledgeacquisition. Compared with product innovation, process innova-tion and improvements are mainly completed inside firms, andthey are usually difficult observe directly from outside the firms.Furthermore, to better cope with market changes and uncertainty,as well as to satisfy specific or customized demands of customers,many improvements are adaptive adjustments that accommodatechanges in product design and production process. Knowledge inthe case of such improvements is largely embodied in the personalskills of engineers and technical workers, taking the form oftechnological know-how. Therefore, compared in terms of productinnovation, cluster firms are more dependent on personal

B. Guo, J.-J. Guo / Technovation 31 (2011) 87–10492

mechanisms (e.g., interpersonal ties and learning-by-hiring) inacquiring technological knowledge in process innovation.

Proposition 4. Cluster firms in emerging economies have different

inclinations in choosing knowledge transmission mechanisms in

product innovation and process innovation. Specifically, firms are

more likely to choose personal mechanisms in process innovation,

whereas they are more prone to select impersonal mechanisms (e.g.,

imitation) in product innovation.

2.6. The influence of technological complexity on technological

learning pattern

In the context of developing countries, the complexity oftechnology involved in the related product sector(s) of a clusteris an important contingent variable in analyzing industrial clustersbecause the underlying complexity of technology in any productsector is likely to have a significant influence on the kind ofknowledge system needed to support high rates of technologicaldynamism (Bell and Albu, 1999). The increase of technologicalcomplexity involved in the products and processes of clusters willexert multiple influences on a knowledge system and on thetechnological learning of clusters. First of all, compared with thosefirms at a low level of technological complexity, cluster firms tendto require more technological knowledge in designing and improv-ing products at a high level of technological complexity. Moreover,an increase in technological complexity will also give rise to closerrelationship and more frequent interactions among producers andupstream and down-stream actors (e.g., clients and suppliers)along with the value chain of the production system in the processof product design and improvements. Another consequence is that,to reduce the technological risk associated with a high level oftechnological complexity, the producers in the cluster will tend tolearn through multiple knowledge acquisition channels.

Technological complexity also affects the usage of specificknowledge transmission mechanisms across cluster firms. Alongwith the increase in technological complexity, the cost of techno-logical knowledge acquisition tends to increase simultaneously,and hence technological complexity will impose a stronger influ-ence on firms’ choices regarding knowledge transmission mechan-isms. In order to reduce the cost of knowledge acquisition,producers will be inclined to choose those mechanisms that havebeen adopted by their competitors and that have been proven to beeffective in acquiring knowledge. Besides, the increase in techno-logical complexity will result in a decrease in the number of sourcesthat can provide qualified technological knowledge. The abovefactors will eventually lead to a similar pattern in choosing andusing knowledge transmission mechanisms across cluster firms. Inother words, along with the increase in technological complexity,the diversity in the usage of specific knowledge transmissionmechanisms will be lower among cluster firms.

Proposition 5. Technological complexity has a strong influence on

technological learning patterns in industrial clusters of emerging

economies. The higher the level of the complexity of technology

involved in that cluster, the more active the cluster’s firms will be in

using multiple knowledge transmission mechanisms to acquire tech-

nological knowledge, and the less diversity in the usage of the same

knowledge transmission mechanisms will exist among the cluster’s

firms.

3. Research design—sample, data and methodology

In the present study, we selected two industrial clusters inZhejiang Province as our research sites: the Shangyu cooling towercluster as a vertical type, and the Haining warp-knitting cluster as a

horizontal type. Our study included two phases. The first phase wasan exploratory qualitative study comprising interviews, the aim ofwhich was to preliminarily identify the cognitive subgroups andthe knowledge transmission channels used by those subgroups(including both knowledge transfer and knowledge spillover) inclusters. The sample cluster in this phase is the Shangyu coolingtower cluster, which was established in 1976 and now has 30producers in total. After a number of pilot interviews, we carriedout a round of formal interviews with five key informants fromthree large-sized firms in the cluster (one of them was the origin ofthis cluster and is the largest enterprise currently), all of whom areexperienced engineers and managers with a technology back-ground. Key informants were chosen as follows: (1) all of them haveworked in the cluster for more than ten years; (2) they boast richmobility experiences with several producers in the cluster. Asposited by Giuliani and Bell (2005), these experienced engineersand managers with a technology background are usually the bestinformants about the history and current characteristics of thefirms, and they are also more important key nodes in the cognitiveinter-connections between firms. The information collectedthrough interviews can also help us to study the potential inter-actions and inter-connecting mechanisms among different cogni-tive subgroups in the industrial clusters, and thereby validateProposition 1. In addition, we have discussed this with someinformants (senior managers and experienced engineers) in otherindustrial clusters (e.g., the Wenzhou Rui’an automobile compo-nent cluster, the Shaoxing textile cluster, and the Wenzhou Liushilow-voltage apparatus cluster) to ensure and validate the general-izability of the technological learning pattern within the knowl-edge system of an industrial cluster. At the same time, following theguidance of an expert informant, we combined the literatureconcerned with knowledge acquisition channels and designedthe questionnaire for the knowledge transmission mechanismsurvey in the second phase of our study.

In the second phase of our study, we did a survey on producers inthe Shangyu cooling tower cluster (a vertical cluster) and theHaining warp-knitting cluster (a horizontal cluster). The warp-knitting cluster of Haining dates back to 1980s. At that time, onlyone or two firms possessed domestic middle-and-low gradedwarp-knitting machines, and they mainly made low-end marketproducts such as garment fabrics and mosquito netting. In the late1990s, the Haining warp-knitting cluster developed rapidly tobecome the largest and the most competitive industry base in theChinese warp-knitting industry under support from the localgovernment. At present, among three types of warp-knittingproducts for garments, decorations, and industrial usage, lamp-box fabric and geotextile fabric rank first in China and account formore than a 35% market share of the Chinese warp-knittingindustry. The first national testing center of warp-knitting productswas established in Haining. By the end of 2006, there are 325 warp-knitting firms in total in Haining with a gross industrial output of1.56 billion US dollars (2006).

For the Shangyu cooling tower cluster, we got 35 usableresponses from 15 firms (one of the total 16 sample firms wasomitted due to missing data). Because of the relatively smallnumber of sample firms from the Shangyu cluster, we collectedmultiple responses from those firms and computed the averages asthe scores used in the analysis to increase the reliability of the data.For the Haining warp-knitting cluster, we obtained 93 responses atfirst, three of which were discarded because of (in one case) missingdata or (in two cases) non-producer firms. In this way, we finallygot 90 usable responses from 90 firms. It should be pointed out thatwe did not report the response rate of our questionnaires as isconventional practice. This is because we gained the help of thelocal government and several key informants in completing thesurvey on these two clusters, and hence all questionnaires handed

B. Guo, J.-J. Guo / Technovation 31 (2011) 87–104 93

out were returned and completed with a high degree of quality. Theaggregate sales of our sample firms from Shangyu cluster were 29.9million US dollars in 2005 and 41.2 million US dollars in 2006,which covered 58.3% and 59.2% of the total sales of the wholecluster in the respective years. The aggregate sales of the samplefirms from the Haining cluster were 402.2 million US dollars in2005 and 560.7 million US dollars in 2006, which made up 40.7%and 35.9% of the total sales of the whole cluster in therespective years.

Our questionnaire contained two parts. The first part was aboutbasic information, including the firm’s year of localization, salesrevenues in 2005 and 2006, number of employees, and lines ofbusiness, as well as respondents’ related information (title, jobtenure, and years of work experience in the line of business). Thesecond part was about the effectiveness of different knowledgeacquisition channels in product innovation and process innovation.Based on Bell and Albu (1999, p. 1725) and Harabi (1997, p. 629)and on information that we collected in the first phase of the study,we designed the effectiveness items concerning each of thefollowing 17 knowledge acquisition channels in the survey:informal contacts with employees in intra-cluster innovatingfirms; informal contacts with employees in extra-cluster innovat-ing firms; hiring away employees with experience at intra-clusterfirms; hiring away employees with experience at extra-clusterfirms; reverse engineering products and technology from intra-cluster (competing) firms; reverse engineering products andtechnology from extra-cluster (competing) firms; internal devel-opment; collaborative development with university and/or inde-pendent research units; collaborative development with intra-cluster firms; collaborative development with extra-cluster firms;technical support and training by key clients; contacts with pieceparts and material suppliers; contacts with equipment and man-ufacturing service suppliers; professional training by business andtrade bodies such as chambers of commerce, business links or tradeassociations; technology licensing; patent disclosure; and publica-tions and open technical meetings. We asked the respondents toanswer the questions on the basis of a 7-point Likert scale (1¼notat all effective; 4¼moderately effective; 7¼very effective); thebasic question is stated as follows: ‘‘For each of the followingchannels by which your firm may acquire technological knowledge,please indicate the effectiveness and importance you attach to itaccording to its persistence and quality’’. The question was askedonce for product innovations and once for process innovations.

After data collection through questionnaires, we categorized theabove-mentioned 17 knowledge acquisition channels into seventypes based on their similarities, i.e., interpersonal ties (informalcontacts with employees in intra-cluster innovating firms; infor-mal contacts with employees in extra-cluster innovating firms;contacts with raw material and parts suppliers; contacts withequipment and manufacturing service suppliers), learning-by-hiring (hiring away employees with experience at intra-clusterfirms, or at extra-cluster firms), imitation (reverse engineeringproducts and technology from intra-cluster firms, or from extra-cluster firms), independent R&D, training (technical support andtraining by key clients; professional training by business and tradebodies), collaborative development (collaborative developmentwith universities and/or independent research units; collaborativedevelopment with intra-cluster firms, collaborative developmentwith extra-cluster firms), and codified knowledge (technologylicensing, patent disclosure, and publications and open technicalmeetings). Using the above seven types of knowledge transmissionmechanisms, we identified the cognitive subgroups with differenttechnological learning patterns through cluster analysis method.Cluster analysis was performed first using Ward’s method, anagglomerative technique that provides a guide for the number ofclusters present in the data. The proximities between variables

were calculated using squared Euclidean distance. Then theanalysis was repeated using the cluster partitions as ‘seed points’for a K-means cluster analysis, which is a divisive technique. Assuggested in Leask and Parker (2007), this procedure has the effectof acting as a cross-check for the Ward’s method results. Thus, theresults presented in this paper are based on the K-means method.

4. An exploratory study of Shangyu cooling tower cluster,China

The origin of the cooling tower cluster in Shangyu can be tracedback to the year 1976. Since 1975, a professor from ShanghaiJiaotong University (China) had transferred the original technologyto three firms in Shangyu, Shangyu Fans Factory (now a publiclylisted company named Shangfenggaoke in the Shanghai SecurityMarket), Shangyu Baiguan Electrical Motor Factory (a specializedproducer of electrical motors used in cooling tower systems), andLianfeng Fiber-reinforced Plastics (FRP) Factory (Lianfeng). As forLianfeng, it was initially established in 1976 as a township-and-village enterprise, and has become the largest cooling towerproducer in the cluster by now with total assets of 56.7 millionUS dollars (in 2006). In fact, the Shangyu cooling tower clusteroriginated from it.

After that, in the late 1980s and early 1990s, a number ofproducers with small-scale cooling tower products successivelywent into this sector, and the cooling tower cluster gradually cameinto being. As for the origin of these producers, some of them werespin-offs from Lianfeng begun by former firm employees (mostlyex-managers and ex-salesmen) who recognized the potentiallyprofitable business opportunity and decided to exploit it by startingtheir own businesses. Besides, some former suppliers of Lianfengentered into the manufacturing of whole cooling tower systemsthrough forward integration by utilizing their complementaryassets (e.g., distribution channels). These new producers hiredaway a few technical engineers and workers from Lianfeng. In thisway, technical knowledge and know-how about product design,manufacturing, installation, and system debugging was diffusedinto the cluster rapidly. At the same time, more and morespecialized suppliers emerged and established steady and on-going linkages with their upstream producers.

4.1. Cognitive subgroups in the cluster

A multi-level networking structure (see Fig. 1 and will bediscussed in the following sub-sections) has been shaped in therelationships among producers and suppliers in clusters. As theresearch focus of the present study, the producers in the Shangyucooling tower cluster can be placed into the following subgroups:

Producer subgroup 1: Leading firms: This type of firm is largeand a system producer. Such firms usually play the role of gate-keepers in the localized knowledge network and establish inde-pendent R&D function for internal development. In the case of theShangyu cooling tower cluster, they can be further divided into twokinds. One kind is the core firms, such as the above-mentionedLianfeng. They have strong technological capabilities and relativelybroader product lines, and they produce a large portion of thecomponents on their own through vertical integration. Anotherkind is the second-best firms that are often spin-offs from the corefirm in the cluster. They generally adopt a focused strategy withregard to specific industrial applications or regional markets (e.g.,two of such producers mainly provide large-scale products forindustrial clients in electric power and petrochemical industries),integrating their internally developed technology expertise intoproduct development to make products more attractive to theirclients.

B. Guo, J.-J. Guo / Technovation 31 (2011) 87–10494

Producer subgroup 2: Close followers: These firms are med-ium-sized in the context of the cluster. They do not invest too muchin in-house R&D. However, they largely emphasize acquiringtechnology and knowledge from outside, and they usually utilizethe breadth-first strategy in searching for external technologicalknowledge (i.e., using as many diverse channels as possible).

Producer subgroup 3: Small producers: They are small-sizedfirms with few product lines that focus on regional markets orsmall-scale products, and they mainly undertake the low-coststrategy in market competition. As far as the technology strategy isconcerned, they invest little in internal developments and mainlydepend upon the copy and imitation strategy in productdevelopment.

The materials, components and services provided by suppliersin the cluster include: raw materials (e.g., plastic, resin, andfiberglass; the suppliers of such materials are almost all outsidethe cluster); plastic components (e.g., PVC packing, ABS splashdevice, and PVC drift eliminators); fans (e.g., aluminum alloy, FRP;the suppliers of such components come from both inside andoutside the cluster); electrical motors (with both intra-cluster andextra-cluster suppliers); other components, equipments and man-ufacturing services, etc. According to their roles and features in theknowledge network, the suppliers of the cluster can be categorizedinto subgroups as follows:

Supplier subgroup 1: Specialized suppliers: They are similar tothe specialist suppliers in Saxenian (1991). As the specializedsuppliers of the cluster, they provide components such as fans,electrical motors and plastic components for local producers.

Supplier subgroup 2: Common suppliers: They often providematerials and components for several sectoral or regional produc-tion systems other than the cooling tower cluster. For example, thecooling tower sector is only one of the sectors that the plasticcomponent suppliers serve. In fact, they manufacture most of theirplastic products for other industries. Therefore, they acquire morebasic knowledge from their own main sectors and transfer thatknowledge into the cooling tower cluster by acting as bridging ties.

Supplier subgroup 3: Equipment and manufacturing servicesuppliers: These firms make manufacturing equipment and

Leader-centered community

Equipment and manufacturing

Producer Knowledge gatekeeper

Specialized suppliers (exclusive)

ProducerKnowledge gate

Specialized suppliers (exclusive)

Specialized suppliers in common acting as inter-community linkages

Intra-community connecting

Fig. 1. The structure of cluster knowledge sy

provide manufacturing services (e.g., machine maintenance) forthe cluster. They play an important role in facilitating the improve-ment of production efficiency and the diffusion of technical know-how related with the manufacturing process in the cluster.

4.2. Technological gatekeeper acting as intra-community connecting

mechanism

In the context of developing countries, large firms generally playan important role in the production systems of industrial clusters.However, whether those large firms play a similarly important rolein the clusters’ production systems as they do in the role ofgatekeepers in knowledge systems is still not clear (Rabellotti,1995; Nadvi, 1996). We found that leading firms (including corefirms and second-best producers) play a crucial role in knowledgesystems. As technological gatekeepers, they are actively involved inscreening, acquiring, transcoding and transferring external knowl-edge into the cluster (Morrison, 2004; Carbonara, 2004).

These technological gatekeepers also promote the diffusion oftechnological knowledge inside cluster through knowledge codi-fication. As leading firms, they accumulate their knowledge andexpertise via learning and practicing in the process of productdesign, manufacturing, installing, selling and problem-solving.Such knowledge is mostly embodied in individual persons and istacit in the form of experience and technological know-how. Part ofit is transferred through internal interpersonal contacts, appren-ticeships and mentoring systems. At the same time, the core firmsoften assign experienced engineers to collect and arrange theseexperiences and know-how and to turn them into codified knowl-edge in terms of manuals of maintenance procedures, materialspecifications, detailed product designs and blue-prints, particularlabor skills and operating routines. Industry norms and standardsare also important elements of technological learning. This isbecause in industrial clusters, only those suppliers who can provideproducts according to related industrial standards are regarded asqualified and hence able to enter into the supply chain system of anindustrial cluster.

service suppliers

keeper

Producer Knowledge gatekeeper

Specialized suppliers (exclusive)

Common suppliers acting as inter-industry linkages

mechanism Inter-community bridging mechanism

stem in Shangyu cooling tower cluster.

B. Guo, J.-J. Guo / Technovation 31 (2011) 87–104 95

It is worth noting that we found little evidence of formaltechnical know-how exchange recognized or supported by firmtop managers among those leading firms. This finding is consistentwith Morrison (2004), in which it was observed that the leadingfirms appear to be well-connected with external sources of knowl-edge as well as with their own network of clients and providers, butthat few horizontal linkages have been detected between leaders.As a core firm in the cluster, Lianfeng is able to produce most of thecomponents internally. Only those components that it cannotproduce, such as electrical motors, fiberglass cloth, resin, and fansfor large-scale cooling towers, are provided by its specializedsuppliers. In addition, its new product developments are generallycarried out with external partners from universities and indepen-dent research institutes. As for the second-best firms, they alsoseldom collaborate with each other. Nonetheless, the componentsuppliers that they have in common make them interact indirectly,since the component suppliers usually and unintentionally spreadthe information about new products and technical problem-solving from one producer to another. In this way, those suppliersin common play the role of bridging ties in transferring knowledgeand information among leader-centered communities. This findingsuggests that this kind of knowledge spanning mechanism can beparticularly important for knowledge sharing, diffusion and spil-lover among producers in the knowledge networks of industrialclusters.

4.3. Knowledge spanner acting as inter-community bridging

mechanism

Although few formal technical linkages have been found amongthe leading firms as mentioned above, the different leader-centeredcommunities within the cluster knowledge system are not dis-connected with each other. Instead, they are combined into apartially inter-connected knowledge system through the so-called’knowledge spanning mechanisms’ in the present study (see Fig. 1).Specifically, our investigation revealed that in the Shangyu coolingtower cluster, there exist the following knowledge spanners andknowledge spanning mechanisms, through which the various localleader-centered communities in the same cluster, as well as theclusters located in the same or neighboring regions or industries,can interact and be inter-linked as a network.

First, those leading producer-centered communities have somespecialized suppliers in common (e.g., suppliers of fans, packedtowers) that act as inter-community linkages (or, say, bridgingties). The specialized suppliers actively transfer related technolo-gical knowledge and information from one community to anotherthrough interpersonal contacts or transfer the technologicalknowledge embodied in improved parts and components acrosscommunities through component supply and later imitation andreverse engineering by the recipient firms.

Second, the common suppliers can serve as inter-industrylinkages to bring the technology and knowledge created in othersectors (or other regions) into the local cluster. For example, themold makers and suppliers for fiber-reinforced plastics (FRP), resinand electrical motors provide products and services not only for theShangyu cooling tower cluster but also for multiple industries orregions. Sometimes they are not even local firms. Thus, they oftenact as bridging ties that transfer the technological knowledge ofother industries, regions and clusters to local cluster firms.

Third, equipment and manufacturing service suppliers some-times play the role of knowledge spanners among differentcommunities in the cluster. On the one hand, these equipmentsuppliers are in contact with most of the producers in the clusterand establish long-term business relationships with some of them.During such business contacts, they will leak the information about

product and process improvements achieved by one producer tothe others via interpersonal contacts. On the other hand, in gettingin touch with producers, the manufacturing equipment and serviceproviders will learn about the improvements in equipmentachieved by the producers themselves. Besides, during the processof designing and producing customized equipments for producers,they can also absorb knowledge regarding product improvementsaccomplished by both sides. All of this knowledge will be incorpo-rated into the product improvements and into new productdevelopments by the manufacturing equipment and service pro-viders. Hence, the knowledge spillovers in terms of knowledgeembodied in improved equipment can benefit all communities. It isalso in this way that the distributive technological improvementsin and among various communities can be integrated as a co-invention process in clusters, which accelerates the rate of tech-nological advance (Nuvolari, 2004) and is a source of productivetechnological development (Meyer, 2006). Such distributive co-invention can help the cluster to realize the economy of flexibilityin innovation through the agglomeration of firms with diversespecialized expertise.

Fourth, we observed an interesting and cluster-specific knowl-edge spanning mechanism in the Shangyu cooling tower cluster. Afew technical employees with special technical expertise are hiredby leading firms. At the same time, they also take part-time jobs forother producers under the implicit consent of their own employers.Thus, these ‘co-employed’ workers constitute a kind of cluster-sharing expertise or asset, helping producers inside the cluster toshare and transfer some technological knowledge across differentcommunities. Typically, as in the case of mold making for FRPparts/components, such cluster-sharing expertise is related tocritical jobs that require a high degree of professional skill. Atthe early formation stage of the Shangyu cooling tower cluster, onlyLianfeng and a few other large-sized firms had full-time moldworkers. Furthermore, as for small-sized firms, the amount of FRPmold making was small, and technicians with special expertisewere relatively rare; thus, almost none of the small-sized firmshired such workers at all. FRP molds in the cluster were at this stageentirely made by the mold workers of Lianfeng through part-timework. To this day, mold making in the cluster is mostly donethrough the temporary hire of part-time engineers and experts,which intentionally diffuses knowledge and technology among theproducers and across the leader-centered communities.

Therefore, the leader-centered communities within the knowl-edge system of an industrial cluster are actually not disconnectedfrom each other. Rather, they are inter-connected through tech-nological gatekeepers as intra-community connecting mechanismand knowledge spanners as inter-community bridging mechanism,in spite of the fact that the inter-community interactions andconnectedness are uneven and selective, not pervasive and collec-tive. The knowledge spanning mechanism can be particularlyimportant for knowledge sharing, diffusion and spillover amongproducers and leader-centered communities in the knowledgenetworks of industrial clusters. Thus, Proposition 1 is confirmed inthis case.

5. Empirical results and discussion

5.1. Descriptive statistics of the sample firms

The sample for the survey comprised responses from 15 firms inthe Shangyu cooling tower cluster and 90 firms in the Hainingwarp-knitting cluster, respectively. Table 1 reports descriptivestatistics on firm-level characteristics, including during what yearsthe firms began operation in the cluster, the firms’ sales in 2005 and2006, and the firms’ number of employees. Like many clusters in

Table 1Descriptive statistics of the sample firms.

Characteristics of firms by Shangyu cooling tower

cluster (N¼15)

Haining warp-knitting

cluster (N¼90)

Year of localization

Up to the 1970s 1 –

1980s 1 6

1990s 8 17

2000s 5 67

Sales in 2005 (million RMB yuan)

0–10 10 33

10–50 3 39

Z50 2 18

Sales in 2006 (million RMB yuan)

0–10 6 21

10–50 7 40

Z50 2 29

Number of employees

0–50 5 6

50–150 8 57

Z150 2 27

Note: 1 US dollar¼8.1917 RMB yuan (2005); 1 US dollar¼7.8087 RMB yuan (2006).

B. Guo, J.-J. Guo / Technovation 31 (2011) 87–10496

developing countries, member firms in Chinese clusters are mostlyof small and medium enterprises (Schmitz and Nadvi, 1999).

According to their effectiveness in acquiring technical knowl-edge, the top five among the 17 knowledge transmission channelsare reported in Table 2 for both the Shangyu cooling tower clusterand the Haining warp-knitting cluster. The results indicate thatbesides the interpersonal contacts mentioned by Morrison (2004),interpersonal contact and labor mobility by Lissoni (2001), inter-personal and formal linkages by Giuliani (2005), and researchcollaboration by Boschma and ter Wal (2007), imitation is also animportant knowledge transmission mechanism in clusters. Firmsalong the horizontal dimension of the cluster are constantly giventhe opportunity to imitate the proven or foreseeable success ofothers while adding some ideas of their own (Maskell, 2001b). Inaddition, other knowledge sources up and down the industry valuechain can be important, too; these include key clients, raw materialand part suppliers, and equipment and manufacturing servicesuppliers. As a consequence, rather than just focusing on a singleknowledge transmission mechanism, we should analyze theknowledge network in cluster from multiple mechanism perspec-tives, in accordance with formal and informal, intentional andunintentional, internal and external, and vertical and horizontallinkages.

5.2. Comparison of learning mechanisms for different cognitive

subgroups

As mentioned in the ‘Research Design’ section, we collected datafrom a survey about the effectiveness of seventeen knowledgeacquisition channels in the process of product innovation andprocess innovation for producers in the clusters. We computed thefirm-level scores for the seven learning mechanisms in terms ofinterpersonal ties, learning-by-hiring, imitation, independent R&D,training, collaborative development, and codified knowledge. Afterthat, we divided the producers in the clusters (including theShangyu cooling tower cluster and the Haining warp-knittingcluster) into several groups through cluster analysis of the aboveseven learning mechanisms. The result shows that, consistent withour exploratory study of the Shangyu cooling tower cluster in thesecond part of Section 4, the producers in both clusters can becategorized into three groups. This validates that there are three

typical cognitive groups in a cluster knowledge network accordingto their specific uses of learning mechanisms. The finding alsosuggests that there exist some inter-connections between knowl-edge networks and production networks in clusters.

Tables 3 and 4 present the firm-level characteristics of differentcognitive subgroups (as a result of the above-mentioned clusteranalysis) in the Shangyu cooling tower cluster and the Hainingwarp-knitting cluster, respectively, as well as the effectiveness ofthe seven learning mechanisms for an average firm in eachcognitive subgroup in product innovation and process innovation.The results presented in Tables 3 and 4 reveal a consistent structurein the knowledge systems for the Shangyu and Haining clusters:

Leading firm subgroup: The size of producers in this subgroup ismuch larger than the average level of cluster firms. Specifically, asfor the average size of the leading firms, it is approximately twicethat of all cluster firms by sales in 2005 and 2006, and it is alsosubstantially larger than that of other cognitive subgroups in termsof the number of employees in 2006. With regard to technologicallearning mechanisms, the leading firms pay the most attention tointernal development in both product and process innovationamong all the cognitive groups in the cluster, which is the mosteffective and important of all the seven learning mechanisms forthem to use in acquiring knowledge.

Close follower subgroup: The producers in this subgroup aremedium-sized. To some extent, they possess certain internaldevelopment capabilities, as reflected by the results in Tables 3and 4: the effectiveness of internal development as a knowledgetransmission mechanism for them is lower than it is for the leadingfirm subgroup, but it is higher than for the small producersubgroup. Compared to the leading firm subgroup and its inclina-tion toward the depth-first strategy of knowledge searching, theseclose followers are more inclined to adopt the breadth-firststrategy in the process of knowledge searching. Table 3 showsthat as in the Shangyu cluster, the close follower subgroup hashigher scores than the leading firm subgroup for the seven learningmechanisms except for learning-by-hiring and internal develop-ment in product and process innovation. Compared with theaverage level of cluster firms, the close follower subgroup displayshigher scores for the effectiveness of the seven learning mechan-isms in both product and process innovation except for learning-by-hiring in the case of product innovation. With regard to theHaining cluster, such inclination toward the breadth-first strategy

Table 2Top five in 17 knowledge acquisition channels in Shangyu and Haining clusters.

SHANGYU cooling tower cluster HAINING warp-knitting cluster

Product innovation

Internal development 5.76 Equipment and manufacturing service supplier 4.79

Intra-cluster imitation 5.47 Intra-cluster imitation 4.25

Extra-cluster imitation 5.38 Intra-cluster hiring 3.94

Intra-cluster interpersonal ties 5.22 Raw material and parts suppliers 3.93

Key clients 4.97 Key clients 3.66

Process innovation

Intra-cluster imitation 5.54 Equipment and manufacturing service supplier 5.22

Internal development 5.44 Intra-cluster imitation 4.69

Extra-cluster imitation 5.33 Intra-cluster hiring 4.66

Intra-cluster interpersonal ties 5.09 Raw material and parts suppliers 4.21

Collaborative development with university and independent research unit 4.96 Key clients 3.88

Table 3Comparison of learning mechanism among different cognitive subgroups: Shangyu cooling tower cluster.

Cognitive subgroups

within industrial

cluster

Averages for the

leading firm

subgroup (a)

Averages for the

close follower

subgroup (b)

Averages for the

small producer

subgroup (c)

T-test for

equality of

means (a–b)

T-test for

equality of

means (a–c)

T-test for

equality of

means (b–c)

Cluster

averages

Number of firms 5 2 8

Sales (2005) (million

US dollars)

4.08 1.61 0.78 1.99

Sales (2006) (million

US dollars)

5.63 2.29 1.06 2.75

Number of employees 123 72 43 73

Product innovation

Interpersonal

communication

4.44 5.29 4.64 �0.85nn (�4.70) �0.20 (�0.91) 0.65 (1.82) 4.66

Hiring 4.64 4.25 5.04 0.39 (0.87) �0.40 (�1.12) �0.79 (�1.60) 4.80

Imitation 5.25 6.09 5.23 �0.83n (�3.57) 0.02 (0.10) 0.85n (3.06) 5.35

Internal development 6.67 6.17 5.08 0.50y (2.00) 1.58nnn (7.82) 1.08nn (3.77) 5.75

Training 4.05 5.84 4.66 �1.78n (�3.27) �0.61 (�1.55) 1.18y (1.92) 4.61

Collaborative

development

4.34 6.06 4.60 �1.71n (�4.06) �0.25 (�0.96) 1.46nnn (5.01) 4.71

Codified knowledge 3.29 4.89 3.74 �1.59 (�1.41) �0.44y (�1.93) 1.15 (1.03) 3.74

Process innovation

Interpersonal

communication

4.38 4.88 4.32 �0.50 (�0.78) 0.06 (0.18) 0.55 (1.04) 4.42

Hiring 4.49 5.17 4.77 �0.68 (�1.37) �0.28 (�0.64) 0.40 (1.18) 4.73

Imitation 5.59 5.92 5.11 �0.33 (�0.85) 0.49 (1.63) 0.81y (2.04) 5.38

Internal development 6.10 6.00 4.90 0.10 (0.27) 1.20n (2.45) 1.10nn (3.54) 5.44

Training 3.68 5.50 4.44 �1.82nn(�4.36) �0.75 (�1.51) 1.06 (1.33) 4.33

Collaborative

development

4.25 5.22 4.65 �0.97y (�2.37) �0.40 (�1.12) 0.57 (1.14) 4.59

Codified knowledge 3.39 5.17 3.80 �1.77n (�2.87) �0.40 (�0.98) 1.37y (2.04) 3.85

The figures in parentheses are T-statistics.

y po0.10.n po0.05.nn po0.01.nnn po0.001.

B. Guo, J.-J. Guo / Technovation 31 (2011) 87–104 97

seems not to be so significant as for the Shangyu cluster. However,for the close follower subgroup, the scores for the effectiveness oflearning-by-hiring and imitation in both product and processinnovation and interpersonal contacts in the case of productinnovation are the highest among all three cognitive subgroups.The results indicate that the close follower subgroup depends onmore diverse learning mechanisms to acquire technological knowl-edge, while the leading firm subgroup puts more emphasis oninternal development.

Small producer subgroup: The producers in this subgroup aregenerally small in size, especially in the Shangyu cluster. Theirinputs in internal development are much fewer than those for theleading firm subgroup and the close follower subgroup. With

technological learning mechanisms, they give more attentions toimitation (especially intra-cluster imitation). As shown in Tables 3and 4, among all the seven learning mechanisms, those that yieldthe highest scores for the small producer subgroup are imitation inproduct and process innovation for the Shangyu cluster and theHaining cluster.

The results also show that there exist some differences intechnological learning mechanisms for the two clusters, mainlyreflected by the difference in the effectiveness and importance ofinternal development in acquiring technological knowledge. As awhole, as a means of knowledge acquisition, internal developmentappears more important in the Shangyu cluster than in the Hainingcluster at both the subgroup level and cluster levels (see also

B. Guo, J.-J. Guo / Technovation 31 (2011) 87–10498

Table 7). This can be attributed, at least in part, to the fact that theproducts in the Shangyu cluster are of a higher technologicalcomplexity than those in the Haining cluster. As a result, the firmsin the Shangyu cluster have to carry out continuous internaldevelopment to enhance their absorptive capabilities and theirabilities to improve their products and manufacturing process.

Moreover, the results indicate that to a certain extent, the twoclusters are different in their focus on technological innovation. Inthe Shangyu cluster, the effectiveness of internal development forproduct innovation is greater than for process innovation (seeTable 3). By contrast, the importance of internal development forprocess innovation in the Haining cluster is more significant thanthat for product innovation at both the subgroup level and clusterlevels (see Table 4). This varying importance of internal develop-ment to product and process innovation can be traced back to thedifferent technological characteristics of the product and manu-facturing processes of the two clusters. For the Shangyu cluster, inaddition to the high technological complexity of its product, theproducts need to be designed and improved according to theclients’ customized requirements in most cases, which forces theproducers to implement corresponding adjustments in both pro-duct design and manufacturing process. Thus, the producers attacha balanced sense of importance to the improvements of bothproducts and process, and they invest a large amount of resourcesin internal development. Contrary to the case of the Shangyucluster, the degree of technological diversity in products is muchlower in the Haining cluster. The producers focus more on themanufacturing process. They need to incrementally improve themanufacturing process, with the aim of flexibly adapting to thechanges in products, enhancing production efficiency and productquality, and reducing production costs.

Table 4Comparison of learning mechanism among different cognitive subgroups: Haining war

Cognitive subgroups

within industrial

cluster

Averages for the

leading firm

subgroup (a)

Averages for the

close follower

subgroup (b)

Averages for the

small producer

subgroup (c)

Number of firms 19 34 37

Sales (2005) (million

US dollars)

10.30 3.02 2.89

Sales (2006) (million

US dollars)

13.66 4.51 4.10

Number of employees 175 119 107

Product innovation

Interpersonal

communication

3.86 3.90 3.34

Hiring 3.61 4.04 2.77

Imitation 3.74 3.96 3.59

Internal development 4.37 2.88 2.49

Training 3.79 3.22 2.84

Collaborative

development

2.74 2.21 2.04

Codified knowledge 2.25 1.79 1.61

Process innovation

Interpersonal

communication

4.25 4.24 3.84

Hiring 4.26 4.40 3.80

Imitation 4.18 4.37 4.08

Internal development 5.00 3.32 2.81

Training 4.00 3.34 2.96

Collaborative

development

3.11 2.45 2.41

Codified knowledge 2.42 1.93 1.64

The figures in parentheses are T-statistics.

y po0.10.n po0.05.nn po0.01.nnn po0.001.

5.3. Comparison of learning mechanisms for vertical- and horizontal-

type cluster

The above-mentioned 17 knowledge transmission channelsin the survey can also be divided into three types as follows:(1) horizontal linkages, including intra-cluster and extra-clusterinter-firm linkages in terms of interpersonal ties, hiring, imitationand collaborative development; (2) vertical linkages, includinglinkages with clients, material and parts suppliers, and equipmentand manufacturing service suppliers; (3) infrastructural linkages,including linkages with universities and independent researchunit, as well as with business and trade bodies. Thus, we caninvestigate the influences of the different linkage types on theeffectiveness of those linkages in firms’ knowledge acquisition.

The results reported in Table 5 show that as a whole, the firmsare more dependent on horizontal linkages than vertical linkagesfor the Shangyu cooling tower cluster as a typical vertical-typecluster. For both product innovation and process innovation, intra-cluster horizontal linkages have the highest cluster averagesamong intra-cluster horizontal linkages, extra-cluster horizontallinkages, vertical linkages, and infrastructural linkages. Regardingthe leading firm subgroup, the scores for intra-cluster horizontallinkages are much higher than those for other linkage types. As forthe small producer subgroup, its scores for intra-cluster and extra-cluster horizontal linkages are higher than those for vertical andinfrastructural linkages. The close follower subgroup is the onlyexception, the scores of which are at a relatively high level for alltypes of linkages. This result is consistent with our findings inSection 5, which show that the close follower subgroup is prone tochoose diverse mechanisms of knowledge transmission, i.e., thebreadth-first strategy in knowledge acquisition.

p-knitting cluster.

T-test for

equality of

means (a–b)

T-test for

equality of

means (a–c)

T-test for

equality of

means (b–c)

Cluster

averages

4.47

6.23

126

�0.05 (�0.45) 0.51nnn (4.38) 0.56nnn (6.90) 3.66

�0.44n (�2.28) 0.84nnn (4.55) 1.27nnn (12.61) 3.43

�0.22y (�1.74) 0.14 (1.00) 0.36nn (3.06) 3.76

1.49nnn (9.65) 1.88nnn (11.05) 0.40nn (3.25) 3.03

0.57nnn (4.43) 0.95nnn (8.94) 0.38nnn (4.11) 3.18

0.53nnn (3.74) 0.70nnn (4.86) 0.17y (1.93) 2.25

0.45nnn (4.05) 0.63nnn (5.32) 0.18n (2.36) 1.81

0.01 (0.05) 0.41n (2.58) 0.40nnn (3.57) 4.08

�0.13 (�0.61) 0.47n (2.04) 0.60nnn (3.70) 4.12

�0.18 (�1.18) 0.10 (0.70) 0.29nn (2.69) 4.21

1.68nnn (7.02) 2.19nnn (10.84) 0.51nn (3.03) 3.46

0.66nnn (4.35) 1.04nnn (7.92) 0.38nn (3.12) 3.32

0.65nn (3.40) 0.69nnn (3.66) 0.04 (0.32) 2.57

0.49nn (3.13) 0.78nnn (5.04) 0.29nnn (3.46) 1.91

Ta

ble

5C

om

pa

riso

no

fth

ety

pe

so

fli

nk

ag

ea

mo

ng

dif

fere

nt

cog

nit

ive

sub

gro

up

s:S

ha

ng

yu

coo

lin

gto

we

rcl

ust

er.

Co

gn

itiv

esu

bg

rou

ps

wit

hin

ind

ust

ria

lcl

ust

er

Pro

du

ctin

no

va

tio

nT

-te

stfo

re

qu

ali

tyo

fm

ea

ns

Pro

cess

inn

ov

ati

on

T-t

est

for

eq

ua

lity

of

me

an

s

Ho

rizo

nta

lli

nk

ag

es

Ve

rtic

al

lin

ka

ge

s

(c1)

Infr

ast

ruct

ura

l

lin

ka

ge

s(d

1)

a1–

b1

a1–

c 1a

1–

d1

Ho

rizo

nta

lli

nk

ag

es

Ve

rtic

al

lin

ka

ge

s

(c2)

Infr

ast

ruct

ura

l

lin

ka

ge

s(d

2)

a2–

b2

a2–

c 2a

2–

d2

Intr

a-

clu

ste

r

(a1)

Ex

tra

-

clu

ste

r

(b1)

Intr

a-

clu

ste

r

(a2)

Ex

tra

-

clu

ste

r

(b2)

Av

era

ge

sfo

rth

ele

ad

ing

firm

sub

gro

up

4.9

14

.52

4.2

24

.36

0.3

9(1

.50

)0

.70n

(2.8

4)

0.5

6(1

.84

)5

.05

4.3

84

.14

3.9

80

.68y

(2.0

6)

0.9

2nn

(3.4

1)

1.0

8n

(2.8

1)

Av

era

ge

sfo

rth

ecl

ose

foll

ow

er

sub

gro

up

5.3

85

.42

5.8

35

.67

�0

.04

(�0

.19

)�

0.4

6(�

0.8

8)�

0.2

9(�

0.8

1)

5.2

15

5.3

95

.50

.21

(0.8

3)

�0

.18

(�0

.29

)�

0.2

9(�

0.5

8)

Av

era

ge

sfo

rth

esm

all

pro

du

cer

sub

gro

up

4.9

95

.03

4.4

54

.44

�0

.04

(�0

.20

)0

.54n

(2.5

4)

0.5

5n

(2.7

8)

4.9

34

.85

4.1

4.4

80

.08

(0.2

2)

0.8

3n

(2.9

5)

0.4

5(1

.60

)

Clu

ste

ra

ve

rag

es

5.0

24

.91

4.5

64

.58

0.1

1(0

.68

)0

.46n

(2.2

8)

0.4

4n

(2.3

5)

5.0

14

.71

4.2

94

.45

0.3

0

(1.3

4)

0.7

2nn

(3.3

8)

0.5

6n

(2.4

1)

Th

efi

gu

res

inp

are

nth

ese

sa

reT

-sta

tist

ics.

ypo

0.1

0.

npo

0.0

5.

nn

po

0.0

1.

B. Guo, J.-J. Guo / Technovation 31 (2011) 87–104 99

By contrast, the results reported in Table 6 indicate that as atypical horizontal cluster, the firms are more dependent on verticallinkages for Haining warp-knitting cluster. No matter what clusterlevel or cognitive subgroup level, the score for vertical linkages ismuch higher than that for horizontal and infrastructural linkages inthe case of product innovation or process innovation. Therefore,Proposition 3 receives support.

5.4. Comparison of learning mechanisms for product innovation and

process innovation

Proposition 4 suggests that the producers have different propen-sities for the selection of technological learning mechanisms in thecase of product innovation or process innovation. On the contrary,Table 7 reveals that for the Haining warp-knitting cluster, inter-personal contacts, learning-by-hiring and imitation are the three mosteffective mechanisms for technological learning in both productinnovation and process innovation. In addition, the ranking ordersfor the scores of the seven technological learning mechanisms aresimilar in both product innovation and process innovation. For theShangyu cooling tower cluster, the scores for internal development,imitation and learning-by-hiring are the three highest among allseven technological learning mechanisms for both product innova-tion and process innovation, and almost without exception, the scoresfor these seven technological learning mechanisms have the sameranking order in both product innovation and process innovation. As awhole, the results demonstrate a very similar pattern for product andprocess innovation in terms of technological learning mechanisms.

Tables 8 and 9 present the top seven in all 17 knowledgeacquisition channels in product innovation and process innovationfor all three cognitive subgroups in the knowledge system of a cluster.When comparing product innovation with process innovation in theShangyu cooling tower cluster, we see that they share almost thesame top seven knowledge transmission channels (although withslightly different ranking order) with the leading firm subgroup,except for extra-cluster hiring in the case of product innovation andintra-cluster hiring in the case of process innovation. For closefollower subgroup, the difference is reflected in intra-cluster andextra-cluster inter-firm collaborative development in the case ofproduct innovation and in technology licensing and equipmentsuppliers in the case of process innovation. The only exceptions forthe small producer subgroup are extra-cluster interpersonal contactin product innovation and intra-cluster inter-firm collaborativedevelopment in process innovation. In the case of the Hainingwarp-knitting cluster, a very similar result is reported in Table 9.For the leading firm and close follower subgroups, the same top sevenknowledge transmission channels have been found in cases of bothproduct innovation and process innovation. As for the small producersubgroup, the small producers’ knowledge acquisition channels inproduct innovation and process innovation are alike to a large degree.The only difference has to do with intra-cluster interpersonal contactin product innovation and extra-cluster hiring in process innovation.On the whole, the results reported in Tables 7–9 suggest thatProposition 4 is not supported. On the contrary, we find that for acertain cognitive subgroup, the firms tend to use similar knowledgetransmission mechanisms in cases of both product innovation andprocess innovation.

Several plausible factors can account for the similar learningpatterns that we found in product innovation and process innova-tion. First of all, most of the technological improvements in theclusters are incremental in nature and are triggered more bymarket pull than technology push. Thus, most of the knowledgein clusters is related to adaptive and incremental improvements,instead of to core technology and core products. The increm-ental nature of the innovation process results in the close

Ta

ble

6C

om

pa

riso

no

fth

ety

pe

so

fli

nk

ag

ea

mo

ng

dif

fere

nt

cog

nit

ive

sub

gro

up

s:H

ain

ing

wa

rp-k

nit

tin

gcl

ust

er.

Co

gn

itiv

esu

bg

rou

ps

wit

hin

ind

ust

ria

lcl

ust

er

Pro

du

ctin

no

va

tio

nT

-te

stfo

re

qu

ali

tyo

fm

ea

ns

Pro

cess

inn

ov

ati

on

T-t

est

for

eq

ua

lity

of

me

an

s

Ho

rizo

nta

lli

nk

ag

es

Ve

rtic

al

lin

ka

ge

s(c

1)

Infr

ast

ruct

ura

l

lin

ka

ge

s(d

1)

c 1–

a1

c 1–

b1

c 1–

d1

Ho

rizo

nta

lli

nk

ag

es

Ve

rtic

al

lin

ka

ge

s(c

2)

Infr

ast

ruct

ura

l

lin

ka

ge

s(d

2)

c 2–

a2

c 2–

b2

c 2–

d2

Intr

a-

clu

ste

r(a

1)

Ex

tra

-

clu

ste

r(b

1)

Intr

a-

clu

ste

r(a

2)

Ex

tra

-

clu

ste

r(b

2)

Av

era

ge

sfo

rth

ele

ad

ing

firm

sub

gro

up

3.2

63

.12

4.6

33

.24

1.3

7nnn

(8.2

6)

1.5

1nnn

(9.4

0)

1.4

0nnn

(9.1

1)

3.6

73

.62

4.8

93

.68

1.2

2nnn

(5.7

4)

1.2

8nnn

(5.4

6)

1.2

1nnn

(6.0

7)

Av

era

ge

sfo

rth

ecl

ose

foll

ow

er

sub

gro

up

3.7

82

.76

4.2

92

.81

0.5

1nnn

(5.2

1)

1.5

3nnn

(18

.15

)

1.4

9nnn

(19

.01

)

4.1

93

.08

4.5

12

.96

0.3

2nn

(2.8

0)

1.4

3nnn

(10

.69

)

1.5

5nnn

(14

.15

)

Av

era

ge

sfo

rth

esm

all

pro

du

cer

sub

gro

up

2.8

22

.61

3.7

22

.49

0.9

0nnn

(10

.89

)

1.1

1nnn

(12

.17

)

1.2

3nnn

(11

.86

)

3.3

23

.32

4.1

42

.64

0.8

2nnn

(7.2

6)

0.8

2nnn

(6.1

1)

1.5

1nnn

(13

.33

)

Clu

ste

ra

ve

rag

es

3.2

82

.78

4.1

32

.76

0.8

5nnn

(10

.06

)

1.3

5nnn

(18

.47

)

1.3

6nnn

(17

.55

)

3.7

33

.29

4.4

42

.97

0.7

2nnn

(7.7

8)

1.1

5nnn

(11

.95

)

1.4

6nnn

(16

.19

)

Th

efi

gu

res

inp

are

nth

ese

sa

reT

-sta

tist

ics.

nn

po

0.0

1.

nnn

po

0.0

01

.

B. Guo, J.-J. Guo / Technovation 31 (2011) 87–104100

inter-connection between product and process improvements.Many process improvements are triggered by specific require-ments in product design and improvements. Another factor isconcerned with relational and structural embeddedness, whichhave significant influences on firm behaviors in a cluster(Dayasindhu, 2002). Specifically, relational and structural embedd-edness give rise to the significant path-dependency in knowledgesearching behavior. In other words, the firms in a cluster areinclined to acquire technological knowledge from those sourceswith which they have previously built up steady relationships. Thisbehavior inclination will help to reduce the cost of screening andacquiring technological knowledge because of the firms’ long-termrelationship with knowledge sources. As a result, the cluster firmstend to acquire and transfer technological knowledge through thesame or similar knowledge transmission mechanisms.

In addition, we find that whether we are talking about productinnovation or process innovation, equipment and manufacturingservice suppliers are important acquisition channels of technolo-gical knowledge, with the highest scores among all channels for thethree cognitive subgroups in the Haining warp-knitting cluster (seeTable 9). This result indicates that equipment and manufacturingservice suppliers play the role of an active knowledge spanner inthe Haining warp-knitting cluster. Through them, information andknowledge about product and process improvements is effectivelytransferred, diffused and shared across cognitive subgroups. Theresults reported in Tables 8 and 9 also clearly indicated that incomparison to the case of the Haining warp-knitting cluster, thedifferences in knowledge acquisition channels between productinnovation and process innovation are more significant in theShangyu cooling tower cluster. To a great extent, such significantdifferences, consistent with the findings in the second part ofSection 5, are caused by the relatively high level of technologicalcomplexity in the Shangyu cooling tower cluster.

5.5. Technological complexity and technological learning pattern

We use variation coefficients to measure the degree of variation forthe firms within the same cluster in the effectiveness of a certainmechanism of knowledge transmission. Variation coefficients arecalculated as the ratio of the standard deviation to the mean of thefirm scores for a certain knowledge transmission mechanism. Asshown in Table 7, whether we are talking about product innovation orprocess innovation, the cluster averages for all seven of the knowledgetransmission mechanisms in the Shangyu cluster are higher thanthose for the Haining cluster, while the variation coefficients for allseven mechanisms, except training, in the Shangyu cluster are lowerthan those in the Haining cluster. As mentioned above, the level oftechnological complexity in the Shangyu cooling tower cluster ishigher than that in the Haining warp-knitting cluster. Thus, the resultsindicate that Proposition 5 receives partial support.

Another interesting finding is that in both process innovation andproduct innovation, the variation coefficients of the two clusters forimitation are the lowest of those for all seven knowledge transmissionmechanisms. This result means that for these two clusters, imitationas a learning mechanism is widely adopted and is also important forthe firms in the clusters. Meanwhile, it provides some empiricalevidence that we should bring those informal and impersonalmechanisms into our analysis of technological learning in clusters.

6. Summary and conclusions

6.1. Summary of findings

Through an interview-based exploratory study and a follow-upsurvey-based quantitative analysis, this paper investigates the

Table 7Comparison of learning mechanisms for Haining warp-knitting cluster and Shangyu cooling tower cluster.

Haining warp-knitting cluster Shangyu cooling tower cluster T-test for equality of means (a–b)

Cluster averages (a) Variation coefficient Cluster averages (b) Variation coefficient

Product innovation

Interpersonal communication 3.66 0.13 4.66 0.10 �1.00nnn (�7.80)

Hiring 3.43 0.22 4.80 0.13 �1.38nnn (�6.57)

Imitation 3.76 0.13 5.35 0.08 �1.59nnn (�11.51)

Internal development 3.03 0.29 5.75 0.14 �2.72nnn (�11.05)

Training 3.18 0.17 4.61 0.19 �1.43nnn (�6.16)

Collaborative development 2.25 0.22 4.71 0.15 �2.46nnn (�16.85)

Codified knowledge 1.81 0.24 3.74 0.20 �1.93nnn (�14.29)

Process innovation

Interpersonal communication 4.08 0.13 4.42 0.13 �0.34n (�2.23)

Hiring 4.12 0.19 4.73 0.15 �0.61nn (�2.79)

Imitation 4.21 0.12 5.38 0.10 �1.16nnn (�8.24)

Internal development 3.46 0.32 5.44 0.18 �1.98nnn (�6.47)

Training 3.32 0.19 4.33 0.23 �1.00nn (�3.78)

Collaborative development 2.57 0.24 4.59 0.14 �2.02nnn (�11.86)

Codified knowledge 1.91 0.27 3.85 0.24 �1.93nnn (�8.00)

The figures in parentheses are T-statistics.

n po0.05.nn po0.01.nnn po0.001.

Table 8Comparison of the top seven in all 17 knowledge acquisition channels for product innovation and process innovation: Shangyu cooling tower cluster.

Subgroup Averages for the leading firm subgroup Averages for the close follower subgroup Averages for the small producer subgroup

Product

innovation

Internal development 6.67 Key clients 6.67 Extra-cluster imitation 5.27

Intra-cluster Imitation 5.60 Collaborative development with

university and independent research

unit

6.33 Intra-cluster imitation 5.25

Intra-cluster interpersonal contact 5.45 Extra-cluster imitation 6.33 Extra-cluster hiring 5.13

Collaborative development with

university

and independent research unit

5.27 Internal development 6.17 Internal development 5.08

Extra-cluster imitation 5.18 Intra-cluster imitation 6.00 Intra-cluster interpersonal contact 5.04

Extra-cluster hiring 4.80 Intra-cluster collaborative

development

6.00 Intra-cluster hiring 4.98

Key clients 4.65 Extra-cluster collaborative

development

5.83 Extra-cluster interpersonal contact 4.92

Process

innovation

Internal development 6.10 Internal Development 6.00 Extra-cluster imitation 5.25

Intra-cluster imitation 6.05 Extra-cluster imitation 6.00 Intra-cluster imitation 5.15

Intra-cluster interpersonal contact 5.37 Collaborative development with

university and

independent research unit

6.00 Intra-cluster interpersonal contact 4.94

Extra-cluster imitation 5.20 Intra-cluster imitation 5.83 Extra-cluster hiring 4.92

Collaborative development with

university

and independent research unit

5.15 Key clients 5.83 Internal development 4.90

Intra-cluster hiring 4.80 Technology licensing 5.50 Intra-cluster collaborative

development

4.88

Key clients 4.58 Equipment suppliers 5.33 Intra-cluster hiring 4.75

B. Guo, J.-J. Guo / Technovation 31 (2011) 87–104 101

technological learning pattern within the knowledge systems ofindustrial clusters in China. More specifically, we attempt toexamine three related issues: first, which cognitive subgroups(i.e., groups of firms with same or similar technological learningcharacteristics in a cluster) generally exist in the knowledgenetworks (or knowledge systems) of industrial clusters? Further-more, what are the roles that they play in knowledge systems, andwhat are their characteristics in technological learning? Second, weargue that the different leader-centered communities withinclusters are not disconnected. Instead, those communities areinter-connected through ‘knowledge spanning mechanisms’. We

attempt to provide some empirical evidence for the existence ofsuch knowledge spanning mechanisms in the present study. Third,we attempt to investigate whether there are any differences intechnological learning mechanisms across different cognitivesubgroups in the same cluster, between product innovation andprocess innovation for the same cognitive subgroup, and amongdifferent cluster types, and to find out the reasons for suchdifferences.

Our study finds that producer cognitive subgroups in theanalyzed clusters consist of three types: the leading firm subgroup,the close follower subgroup and the small producer subgroup. The

Table 9Comparison of the top seven in all 17 knowledge acquisition channels for product innovation and process innovation: Haining warp-knitting cluster.

Subgroup Averages for the leading firm subgroup Averages for the close follower subgroup Averages for the small producer subgroup

Product

innovation

Equipment and manufacturing

service suppliers

5.00 Equipment and manufacturing

service suppliers

5.03 Equipment and manufacturing

service suppliers

4.46

Key clients 4.68 Intra-cluster hiring 4.97 Intra-cluster imitation 3.95

Internal development 4.37 Intra-cluster imitation 4.76 Material and parts suppliers 3.62

Material and parts suppliers 4.21 Material and parts suppliers 4.12 Extra-cluster imitation 3.24

Intra-cluster hiring 4.05 Intra-cluster interpersonal contact 3.79 Key clients 3.08

Intra-cluster imitation 3.89 Key clients 3.74 Intra-cluster hiring 2.95

Extra-cluster imitation 3.58 Extra-cluster imitation 3.15 Intra-cluster interpersonal contact 2.92

Process

innovation

Equipment and manufacturing

service suppliers

5.26 Equipment and manufacturing

service suppliers

5.32 Equipment and manufacturing

service suppliers

5.11

Internal development 5.00 Intra-cluster hiring 5.32 Intra-cluster imitation 4.41

Key clients 4.74 Intra-cluster imitation 5.26 Intra-cluster hiring 4.05

Material and parts suppliers 4.68 Intra-cluster interpersonal contact 4.41 Material and parts suppliers 3.89

Intra-cluster hiring 4.63 Material and parts suppliers 4.29 Extra-cluster imitation 3.76

Intra-cluster imitation 4.21 Key clients 3.91 Extra-cluster hiring 3.54

Extra-cluster imitation 4.16 Extra-cluster imitation 3.47 Key clients 3.43

B. Guo, J.-J. Guo / Technovation 31 (2011) 87–104102

Leading firm subgroup comprises large and system producers. Theygenerally act as gatekeepers in local knowledge networks andestablish R&D departments inside their enterprises to undertakeinternal development activities. The producers in the close follower

subgroup are medium-sized. They invest less in internal develop-ment as compared with the leading firm subgroup, but they pay agreat deal of attention to acquiring technology and knowledge fromexternal sources. Usually, they employ the breadth-first strategy inknowledge searching and use as many diverse channels as possible.The firms in the Small producer subgroup are small in size andmainly adopt the low-cost strategy in market competition withlimited product lines. Regarding technological strategy, they investlittle in internal development, and hence, their technologicalcapabilities are relatively weak. As a result, they mostly undertakethe copy and imitation strategy in product development.

By further analyzing the differences in knowledge transmissionmechanisms and the reasons for these differences, our study findsthat vertical-type clusters depend more on horizontal linkages forknowledge acquisition, whereas horizontal-type clusters rely moreon vertical linkages. As for horizontal-type clusters, the directlearning opportunities among producers are restricted because ofthe great similarities between products and technology for thoseproducers. By contrast, for vertical-type clusters, large firms play amore active role as knowledge suppliers. The knowledge acquiredby cluster members inside their communities is largely relatedto specific technical problem solving in regular production(e.g., dealing with problems of product quality, small changes inproducts, and incremental improvements in the manufacturingprocess), while new product ideas, process adaptations related tonew products and major process restructuring mostly need to beattained through informal and formal channels in horizontallinkages. Furthermore, as far as horizontal-type clusters are con-cerned, a large proportion of learning opportunities for producerscome from improvements in the manufacturing process ratherthan from products due to the above-mentioned similarities inproducts and technology. Thus, vertical linkages are involved morefrequently in knowledge transmission among cluster firms. Thestrong ties in the vertical linkages among producers, suppliers andclients are more effective and efficient in transferring such knowl-edge regarding process improvements that is highly tacit. On theother hand, for vertical-type clusters, there are more opportunitiesfor products to be improved because of the technological complex-ity and systematic nature of these products, which provides thenecessary basis for horizontal learning among the producers ascompetitors.

Another interesting finding is that for product innovation andprocess innovation, the producers in the clusters exhibit very similartechnological learning patterns at both the cluster level and cognitivesubgroup level. Several plausible factors can account for the similarlearning patterns in product innovation and process innovation. Onereason is that most technology improvements in these clusters areincremental and more market-oriented than technology-driven. As aresult, the knowledge embedded in cluster is related with the moredown-stream phase of innovation, involving detailed product design,testing, re-design and production. Due to the incremental nature ofthe innovation process, a lot of process improvements are closelycoupled with and triggered by the improvements in products.Another reason is that the knowledge searching behavior of clusterfirms has strong path-dependency due to the relational and structuralembeddedness inside industrial clusters. The cluster firms tend toacquire technological knowledge from sources with which they havealready built up long-term relationships in order to reduce thescreening, evaluating and acquiring costs of technological knowledgesearching. As a result, cluster firms are inclined to utilize the same orsimilar knowledge transmission mechanisms to acquire and transfertechnological knowledge.

The influence of technological complexity on the technologicallearning behavior of cluster firms is also investigated in the presentstudy. The results reveal that the complexity of the technologyinvolved in the related product sector(s) of a cluster is an importantcontingent variable in analyzing industrial clusters. An increase inthe technological complexity of products and processes willimpose an essential influence on technological learning behaviorin the knowledge network of industrial cluster, in terms of theamount of technological knowledge required for technology devel-opment and improvement, as well as in terms of the frequency ofthe interactions among producers, clients and suppliers. We findthat in order to reduce the technology risk wrought by a high levelof technological complexity, the producers in the cluster areinclined to utilize multiple knowledge acquisition channels intechnological learning. In addition, along with the increase intechnological complexity, the variation in the effectiveness of onemechanism among cluster members in knowledge transmissionwill decrease.

6.2. Theoretical implications

The present study differs from and contributes to the previousliterature on innovation and learning within industrial clusters in

LEARNING OPPORTUNITY

• The underlying complexity of technology • Interconnectedness between product and

• Path dependency in knowledge searching • Incremental nature in technology development

LEARNING CAPACITY• Absorptive capacity perspective

LEARNING BEHAVIOR• Structure of knowledge system • Selection and usage of Knowledge

transmission mechanisms

process

Fig. 2. Learning opportunity, learning capacity and learning behavior in industrial cluster.

B. Guo, J.-J. Guo / Technovation 31 (2011) 87–104 103

the following ways. First, our findings contribute to a comprehen-sive understanding of patterns of technological learning within theknowledge systems of industrial clusters in the context of emergingeconomies. Some recent research has put forward the idea thatthere are several leader-centered communities in clusters, insidewhich the leading firms will act as knowledge gatekeepers for theirown networks; meanwhile, it is said that these communities aredisconnected from each other (Lissoni, 2001; Morrison, 2004;Giuliani, 2005; Boschma and ter Wal, 2007). Based on the empiricalfindings, this study argues that although there is no directtechnological collaboration and formal technology exchangeamong those leading firms as technological gatekeepers in clusters,the leader-centered subgroups within the knowledge system of anindustrial cluster are not disconnected from each other, and thatthey are inter-connected through technological gatekeepers andknowledge spanners in a selective way. Some knowledge spanning

mechanisms identified by us connect these communities as aninter-linked firm network. There are at least four types of knowl-edge spanning mechanisms, as revealed in the present study: (a)The leading producer-centered communities in clusters have somespecialized suppliers in common, which can act as inter-communitylinkages (or bridging ties) in the local knowledge system. (b) Thecommon suppliers can serve as inter-industry linkages to bring thetechnology and knowledge created in other sectors (or otherregions) into the local cluster. (c) The equipment and manufacturing

service suppliers sometimes also play the role of knowledgespanners among different communities in clusters. As such, asimilar finding has been reported in the machine tool (MT) clusterin Taiwan (Chen, 2009, p. 531) —’’MT firms manage to acquire theirdomestic counterparts’ confidential know-how with the help oftheir suppliers in the production networks. Suppliers are activefacilitators circulating information and know-how among firms incluster. These suppliers provide services for multiple local MT firmsand may very well possess specific information or know-howconcerning their respective clients.’’ (d) In some cases, there aresome cluster-specific knowledge spanning mechanisms, such asthe temporary hiring of cluster-sharing expertise in the Shangyucooling tower cluster.

Second, the present study highlights the importance of takinginto consideration the existence of multiple and diverse knowledgetransmission mechanisms in clusters, as well as the potentialsubstitution and complementary relationships among thesemechanisms, in investigating the learning and innovation behaviorwithin industrial clusters. Some of the recent studies from the LKSperspective have adopted a single-mechanism analysis whileignoring the potential complementary and/or substitution rela-tionship among different knowledge transmission mechanisms.Our findings suggest that it is necessary to bring local and non-local, formal and informal, knowledge transfer and knowledge

spillover, and personal and impersonal knowledge acquisitionchannels into the research on knowledge network and technolo-gical learning in industrial clusters. Especially for the producers inindustrial clusters, in contrast to the existence of intense verticalinteraction between producers and their customers, horizontalinteraction among producers is sparse due to their direct competi-tion in the market (Chen, 2009). Besides, the innovations on bothproducts and processes are mainly incremental in nature inindustrial clusters (Albino et al., 2006). Consequently, in additionto the intentional and formal mechanisms, the contribution ofunintentional and informal learning mechanisms (such as imita-tion and interpersonal communications among local engineers),which has been less discussed and investigated in the previousstudies, is indispensable in constructing a better, comprehensiveunderstanding of technological learning patterns in industrialclusters of emerging economies.

Third, the present study suggests the importance of synthesizingthe learning opportunity perspective and the absorptive capacity

perspective in analyzing the learning behavior within the knowledgenetworks of industrial clusters (see Fig. 2). Some recent studies (e.g.,Morrison, 2004; Giuliani, 2005, 2007; Giuliani and Bell, 2005) havefocused on the influence of absorptive capacity on the structuralposition and learning behavior of cluster firms inside the clusterinnovation networks. As a theoretical extension, our study providesstrong empirical evidence that learning capacity and learning oppor-tunity have a complementary relationship in determining learningbehavior at both firm level and cluster level. By combining these twoperspectives, we can better understand and explain the similaritiesand dissimilarities in technological learning behavior among differentcluster types, across cognitive subgroups, and between productinnovation and process innovation. Our study reveals that in thecontext of emerging economies, the following four factors are decisivefor technological learning opportunities inside the knowledge net-works of industrial clusters: the underlying complexity of thetechnology in the cluster, the inter-connectedness of product andprocess, path dependency in knowledge searching, and the incre-mental nature of cluster’s technological development.

6.3. Limitations and directions for future research

It is worth noting that this study has a number of limitations. Asalways with questionnaire research, there are some concerns as to theusage of perceptual measures (Fey and Birkinshaw, 2005, p. 618). In thisstudy, we use the self-reported and subjective items to measure theeffectiveness of different knowledge acquisition channels in productinnovation and process innovation. Future work should strive to movefurther in collecting objective firm-level data of technological learningbehavior in industrial clusters. Additionally, by collecting a large

B. Guo, J.-J. Guo / Technovation 31 (2011) 87–104104

firm-level data set, future research can statistically examine theinfluences of the four above-mentioned decisive factors concerningtechnological learning opportunity inside cluster knowledge system onfirms’ learning behavior and strategy.

The second limitation is that the sample clusters used in thepresent study might prevent us from generalizing the results, sinceall the research findings are drawn from the firm-level data in thetwo industrial clusters of Zhejiang province, China. Thus, futureresearch is needed to explore and examine the similarities anddissimilarities in similar or different industrial cluster contexts(e.g., knowledge-intensive clusters vs. low- and medium-technol-ogy clusters), as well as in the contexts of developing and/ordeveloped countries. Cross-industry and cross-national compara-tive research would enable us to investigate the generalizability ofthe results in the present study and better understand the patternsof technological learning within the knowledge systems of indus-trial clusters across industries and in emerging economies.

Acknowledgements

The authors would like express their appreciation to twoanonymous referees for their constructive and thorough commentsand Editor-in-chief Jonathan Linton for the editorial comments thatimproved this manuscript, and their gratitude to Zhigang Gu andZhong Liang for their extensive support and efforts with regard todata collection. Funding provided by the International Develop-ment Research Center (IDRC, Canada) and Zhejiang ProvincialNatural Science Foundation of China (Grant no. R6090360) aregratefully acknowledged.

References

Albino, V., Carbonara, N., Giannoccaro, I., 2006. Innovation in industrial districts: anagent-based simulation model. International Journal of Production Economics104, 30–45.

Altenburg, T., Meyer-Stamer, J., 1999. How to promote clusters: policy experiencesfrom Latin America. World Development 27, 1693–1713.

Baptistra, R., Swann, P., 1998. Do firms in clusters innovate more? Research Policy27, 525–540.

Barboza, D., 2004. Roaring China, Sweaters Are West of Socks City. The New YorkTimes, December 24, 2004.

Bell, M., Albu, M., 1999. Knowledge systems and technological dynamism inindustrial clusters in developing countries. World Development 27, 1715–1734.

Boschma, R.A., ter Wal, A.L.J., 2007. Knowledge networks and innovative perfor-mance in an industrial district: the case of a footwear district in the south ofItaly. Industry and Innovation 14, 177–199.

Breschi, S., Lissoni, F., 2001. Knowledge spillovers and local innovation systems: acritical survey. Industrial and Corporate Change 10, 975–1005.

Breschi, S., Malerba, F., 2001. The geography of innovation and economic clustering:some introductory notes. Industrial and Corporate Change 10, 817–833.

Camagni, R., 1991. Innovation Networks: Spatial Perspectives. Belhaven Press,London, New York.

Cantner, U., Meder, A., Ter Wal, A.L.J., 2010. Innovator networks and regionalknowledge base. Technovation 30 (9/10), 496–507.

Capello, R., Faggian, A., 2005. Collective learning and relational capital in localinnovation processes. Regional Studies 39 (1), 75–87.

Carbonara, N., 2004. Innovation processes within geographical clusters: a cognitiveapproach. Technovation 24, 17–28.

Chen, L.C., 2009. Learning through informal local and global linkages: the case ofTaiwan’s machine tool industry. Research Policy 38 (3), 527–535.

Dahl, M.S., Pedersen, Christian Ø.R., 2004. Knowledge flows through informalcontacts in industrial clusters: myth or reality? Research Policy 33, 1673–1686.

Dayasindhu, N., 2002. Embeddedness, knowledge transfer, industry clusters andglobal competitiveness: a case study of the Indian software industry. Technova-tion 22, 551–560.

Feldman, M.P., Florida, R., 1994. The geographic sources of innovation: technologicalinfrastructure and product innovation in the United States. Annals of theAssociation of American Geographers 84 (2), 210–229.

Fey, C.F., Birkinshaw, J., 2005. External sources of knowledge, governance mode, andR&D performance. Journal of Management 31 (4), 597–621.

Giuliani, E., 2005. The structure of cluster knowledge networks: uneven andselective, not pervasive and collective. DRUID Working Paper No. 05-11,Copenhagen.

Giuliani, E., 2007. The selective nature of knowledge networks in clusters: evidencefrom the wine industry. Journal of Economic Geography 7, 139–168.

Giuliani, E., Bell, M., 2005. The micro-determinants of meso-level learning andinnovation: evidence from a Chilean wine cluster. Research Policy 34, 47–68.

Griliches, Z., 1979. Issues in assessing the contribution of research and developmentto productivity growth. Bell Journal of Economics 10 (1), 92–116.

Griliches, Z., 1992. The search for R&D spillovers. Scandinavian Journal of Economics94 (Suppl.), 29–47.

Harabi, N., 1997. Channels of R&D spillovers: an empirical investigation of Swissfirms. Technovation 17, 627–635.

Harrison, B., 1992. Industrial districts: old wine in new bottles? Regional Studies26 (5), 469–483.

Hessler, P., 2007. China’s Instant Cities. National Geographic Magazine, June 2007.Kang, Y., 2007. Made in China: coastal industrial clusters and regional growth. Issues

in Political Economy 16.Keeble, D., Wilkinson, F., 1999. Collective learning and knowledge development in

the evolution of regional clusters of high technology SMEs in Europe. RegionalStudies 33 (4), 295–303.

Lazerson, M.H., Lorenzoni, G., 1999. The firms feed industrial districts: a return to theItalian source. Industrial and Corporate Change 8, 235–266.

Leask, G., Parker, D., 2007. Strategic groups, competitive groups and performancewithin the U.K. pharmaceutical industry: improving our understanding of thecompetitive process. Strategic Management Journal 28, 723–745.

Lee, D., 2005. The New Foreign Aid. Los Angeles Times, April 10, 2005.Li, H.-L., Tang, M.-J., 2010. Vertical integration and innovative performance: the

effects of external knowledge sourcing modes. Technovation 30 (7/8), 401–410.Lissoni, F., 2001. Knowledge codification and the geography of innovation: the case

of Brescia mechanical cluster. Research Policy 30, 1479–1500.Malecki, E.J., 2010. Everywhere? The geography of knowledge. Journal of Regional

Science 50 (1), 493–513.Marshall, A., 1920. Principles of Economics. MacMillan, London.Martin, R., Sunley, P., 2003. Deconstructing clusters: chaotic concept or policy

panacea? Journal of Economic Geography 3, 5–35.Marukawa, T., 2006. The geography and history of industrial clusters in Zhejiang

Province, China. Working Paper, November 2006.Maskell, P., 2001a. The firm in economic geography. Economic Geography 77,

329–344.Maskell, P., 2001b. Towards a knowledge-based theory of the geographical cluster.

Industrial and Corporate Change 10, 921–943.Meyer, P.B., 2006. The Airplane as a Collective Invention. The Office of Productivity

and Technology, US Bureau of Labor Statistics.Morosini, P., 2004. Industrial clusters, knowledge integration and performance.

World Development 32, 305–326.Morrison, A., 2004. Gatekeepers of knowledge within industrial districts: who they

are, how they interact. CESPRI Working Paper No. 183, Milano, Italy.Nadvi, K.M., 1996. Small firm industrial districts in Pakistan. IDS D. Phil Thesis,

University of Sussex, Brighton.Nuvolari, A., 2004. Collective invention during the British Industrial Revolution: the

case of the Cornish pumping engine. Cambridge Journal of Economics 28,347–363.

Porter, M.E., 1990. The Competitive Advantage of Nations. Macmillan Press, London.Porter, M.E., 1998. Clusters and the new economics of competition. Harvard

Business Review (November–December), 77–90.Power, D., Lundmark, M., 2004. Working through knowledge pools: labour market

dynamics, the transference of knowledge and ideas, and industrial clusters.Urban Studies 41, 1025–1044.

Rabellotti, R., 1995. External economies and cooperation in industrial districts: acomparison of Italy and Mexico. IDS D. Phil Thesis, University of Sussex,Brighton.

Rosenfeld, S., 1997. Bringing business clusters into the mainstream of economicdevelopment. European Planning Studies 5, 3–23.

Saxenian, A., 1991. The origins and dynamics of production networks in SiliconValley. Research Policy 20 (3), 423–437.

Schmitz, H., Nadvi, K., 1999. Clustering and Industrialization. World Development27, 1503–1514.

Zeng, S.X., Xie, X.M., Tam, C.M., 2010. Relationship between cooperation networksand innovation performance of SMEs. Technovation 30 (3), 181–194.

Zhang, Z.M., To, C., Cao, N., 2004. How do industry clusters success: a case study inChina’s textile and apparel industries. Journal of Textile and Apparel, Tech-nology and Management 4, 1–10.