Knowledge and Information Networks in an Italian Wine Cluster

Abstract

The aim of this article is to analyse the nature and extent of knowledge and information networks in an Italian wine cluster. Moreover, the relation between firms’ characteristics and the knowledge network structure is also explored. The empirical findings show that knowledge is unevenly distributed in clusters and that networks of knowledge and information differ a great deal in terms of their structure. In fact, knowledge flows are restricted to a tightly connected community of local producers, differing in terms of knowledge assets, innovation behaviour and overall economic performance with respect to the rest of the firms in the cluster.

1. Introduction

Following Marshall's (1920) seminal contribution, which introduced the concept of “industrial atmosphere”, many scholars have stressed the public nature of knowledge in geographically bounded areas such as industrial districts, clusters, milieux innovateurs, local production and innovation systems.1 Many studies have underlined the role of personal, face-to-face interaction as an effective way to transfer knowledge locally (Asheim, 1994; Audretsch & Feldman, 1996; Becattini, 1990; Brusco, 1996; Feldman, 1999; Saxenian, 1994) related to some of the basic properties of knowledge; that it is idiosyncratic, contextual, sticky and tacit. Since knowledge is incorporated in the skills of individuals, learning mainly occurs through personal interaction, the necessary and, to some extent, sufficient conditions of which are physical proximity and local embeddedness. Thus, it is generally claimed that members of a geographical agglomeration benefit from local knowledge spillovers (LKS), because they are spatially close and embedded in local networks of informal contacts.

Some scholars have recently challenged this traditional approach (Boschma & Frenken, 2006; Breschi & Lissoni, 2001a, 2001b; Capello & Faggian, 2005; Lissoni, 2001; Malmberg & Maskell, 2002) by arguing that it overlooks the very different types of knowledge flows in local agglomerations, and in turn it fails to distinguish between those flows that are freely available (e.g. information) and those that are not (e.g. tacit knowledge). They have argued that underlying the well-accepted belief that informal contacts represent an easy conduit for transferring tacit knowledge in geographically bounded areas there is a somewhat ambiguous definition of knowledge and information as interchangeable concepts, sometimes referred to as rumours, ideas or know-how. Thus, they suggest that the nature of knowledge flows and the mechanisms through which they circulate need more detailed investigation.

In this view, physical distance is not the only and main factor explaining the spatial diffusion of knowledge (Boschma, 2005; Staber, 2001). Several recent works have shown that knowledge diffusion is influenced by other kinds of distance than geographical distance, i.e. institutional distance (Rallet & Torre, 2005), cognitive distance (Nooteboom, 1999) and social distance (Breschi & Lissoni, 2009), and that, therefore, knowledge spreads unevenly among members of a local agglomeration of economic activities (Giuliani & Bell, 2005; Lissoni & Pagani, 2003). This unevenness also arises from the inherent heterogeneity that characterizes firms’ learning trajectories and their mechanisms for building capabilities (Dosi, 1988; Nelson & Winter, 1982), which, in turn, give rise to the different absorptive capacities of firms (Cohen & Levinthal, 1990). Thus rather than being an undifferentiated and homogeneous population of firms, these agglomerations may embody different competing networks, characterized by structural differences (e.g. dispersed, centralized, cohesive) (Giuliani, 2007; Lazerson & Lorenzoni, 1999; Rabellotti & Schmitz, 1999).

This article aims to contribute to this field of studies by undertaking a detailed investigation of the structure and constituent properties of knowledge and information networks in a cluster, simply defined here as a geographical agglomeration of sectorally specialized firms. Moreover, following a methodological framework proposed in Giuliani and Bell (2005) and Giuliani (2007), we investigate the relationship between some characteristics (e.g. size, export strategy, knowledge base) of the firms, their structural position in the knowledge network as well as their extra-cluster linkages.

In doing so, we contribute to the debate in the literature in several ways. First, our study goes beyond the often anecdotal evidence on the relevance and diffusion of informal chit-chat in district-like areas by providing a detailed measure of knowledge and information flows shared through informal relations. Such new evidence contributes to a growing area of studies which use social network analysis to investigate linkages among firms and the different actors in clusters or alike (Boschma & Ter Wal, 2007; Giuliani, 2007; Graf, 2007; Kauffeld-Monz & Fritsch, 2007; Morrison, 2008; Samarra & Biggiero, 2008). Second, with respect to these network studies, it reaches some new and interesting results on the role of intra-cluster linkages with respect to relationships with actors external to the cluster. This contrasts with the prevailing view, which takes for granted the positive association between firms’ learning opportunities and network cohesiveness. Too much clustering or network closure can indeed be detrimental to learning dynamic at local level (Boschma, 2005). In our case study we show that the larger and more successful producers in the local system are not interested in forging internal knowledge linkages with the local small and very small firms; larger firms remain in the periphery of the knowledge network and strengthen their linkages with sources of knowledge external to the cluster, whereas the smaller firms are highly interconnected and communicate with external sources of knowledge only through a local broker (e.g. extension agency), which plays the role of gatekeeper for the local system.

The article is organized as follows. Section 2 reviews the main issues in the literature with a focus on the distinction between knowledge and information (Section 2.1) and on the structure of knowledge networks (Section 2.2). Section 3 briefly describes the local system under investigation. Section 4 describes the methodology and discusses the sample and the survey design. Section 5 presents the findings of the network analysis, focusing on the topology of information and knowledge networks. Section 6 concludes.

2. The Theoretical Framework

2.1 Information and Knowledge Diffusion in Clusters

A large body of empirical and theoretical literature in the fields of industrial and regional studies concurs that in contexts such as industrial districts, clusters and local production systems, informal relations are key channels for the diffusion of knowledge. The line of reasoning in this literature is that local and personal (i.e. tacit) knowledge is primarily exchanged by people who are involved in its creation or by those that are part of the same local community.2

Following Marshall (1920), Becattini (1990) goes back to the concept of “industrial atmosphere” and emphasizes the importance of localized knowledge externalities that accrue from face-to-face contacts and the co-location of people and firms. In this context, access to information and knowledge appears unintentional and facilitated by geographical proximity and by the fact that the different actors in the cluster (i.e. entrepreneurs, technicians, workers) have common cultural values, communication codes and behavioural norms (Maskell, 2001). According to this view, informal contacts allow knowledge to be shared by cluster members, while outsiders are excluded, since they are not embedded in the local community.

Audretsch and Feldman (1996), in the context of US innovative activities, provide robust empirical evidence of the existence of a positive relationship between spatial clustering, localized knowledge spillovers and firms’ innovative output. The presence of LKS explains why firms tend to co-locate, and informal relations (i.e. face-to-face contacts) appear to be the relevant mechanisms for the transmission of tacit knowledge. The view that knowledge spillovers are highly localized is also expressed in a number of other econometric studies which show that physical proximity matters since it increases the actors’ probability of contacts and, hence, the flow of information exchange among them (Audretsch, 1998; Jaffe, 1989; Jaffe et al., 1993). In these studies, knowledge is considered to be a public good which spreads pervasively within a spatially bounded area.

Regional economists have criticized this interpretation of space because it focuses only on geographical proximity; they stress the additional importance of institutional and cultural proximity. A number of contributions have referred to the concept of innovative milieu to account for the learning processes occurring at local and network levels (Camagni, 1991; Capello, 1999; Keeble & Wilkinson, 1999; Rallet & Torre, 2005). In this literature, learning is seen as a collective, social process involving people who share strong social and cultural values. Informal relations within the milieu, along with other mechanisms (e.g. spin-offs; labour mobility; user–producer interactions), contribute to sustaining the diffusion of knowledge at local level, which is considered a club good within the boundaries of the cluster (Capello, 1999; Capello & Faggian, 2005). The mileu approach clearly makes the point that it is not only firms’ geographical proximity but also their embeddedness that influences the process of innovation in clusters. However, in common with those contributions focusing mainly on the spatial dimension of proximity, throughout most this literature there is the tacit assumption that clusters are a homogeneous community of firms and entrepreneurs, which have the same culture, origins and knowledge base.

A number of recent contributions in the field of geography of innovation have shifted their attention to the learning dynamics in clusters at firm level (Boschma & Frenken, 2006; Breschi & Lissoni, 2001a, 2009; Giuliani & Bell, 2005; Malmberg & Maskell, 2002). These approaches point to the fundamental role of social interactions in the creation and diffusion of knowledge and explore the extent to which geography mediate in this process. Some of these scholars have also challenged the widespread belief that “industrial districts, which by definition, rely upon long established and homogeneous social networks, are best placed to diffuse and produce tacit knowledge” (Lissoni, 2001, p. 1480). This claim relies on the argument that tacit knowledge, being personal and specific (Polanyi, 1962), cannot be communicated by word of mouth; therefore, knowledge is not a public good, freely available to all clusters members, it is rather a club good, circulating within a few small “epistemic communities”. These communities are formed by small groups of people (e.g. technicians, professionals), who work in the same technical fields—although in different firms, come up against the same technical problems, rely on similar heuristics and procedures on how to conduct research and own common views about who is allowed to access their knowledge and which part of it can be released (Cowan et al., 2000). In these more selective contexts as opposed to whole clusters, informal contacts might function as channels along which knowledge is exchanged, as shown in several empirical works on collective invention and knowledge diffusion (Allen, 1983; Rogers, 1982; Schrader, 1991; von Hippel, 1987). These latter contributions show that technicians feel themselves to be part of a cohesive professional community, in which reputation and status play a key role in shaping interactions. They also show that exchanges take the form of trading; technicians are willing to release crucial information on the basis of reciprocity (Schrader, 1991; von Hippel, 1987). The above implies that knowledge is not given for free, but is exchanged through barter, and the preconditions for exchange are trust, mutual recognition and long-term relationships (Carter, 1989).

However, in addition to exchange of technical advice, technicians use interpersonal networks for a number of other different purposes: to share information about job opportunities, markets, use of machinery, inputs, new regulations, etc. (Cross et al., 2001; Granovetter, 1973; Stuart & Sorenson, 2003). Therefore, contacts established between peers (e.g. entrepreneurs, workers, researchers) do not necessarily entail transfer of tacit knowledge; it is more likely that they serve to share information about who knows whether a customer is reliable or not or whether a machinery is well functioning or not. Some recent contributions using social network analysis seem to confirm this argument (Boschma & Ter Wal, 2007; Giuliani, 2007; Morrison, 2008). They show that firms in clusters or alike are connected through a variety of formal and informal networks (e.g. business networks, advice networks, technological networks, managerial networks), but only a minor fraction of these contacts concern knowledge exchanges. Thus, much of the evidence showing that people interact through informal means—the often cited cafeteria effect—does not support the claim that the content of these exchanges is knowledge, which conversely is selectively appropriated by few members of the local community.

The above considerations suggest that the literature hype about the importance of clustering for knowledge diffusion has contributed to the creation of a myth (Dahl & Pedersen, 2004). In this article, we aim at investigating the nature of knowledge which is passed on through informal contacts by distinguishing between those that convey know-how (i.e. knowledge) and those that vehicle know what or declarative knowledge (i.e. information), being the former mainly tacit and the latter mainly codified (Lissoni, 2001).3 Moreover, given that informal contacts are established for different purposes and respond to different motivations, we expect that the structure of information networks will differ from that of knowledge networks.

To distinguish between knowledge and information is relevant since it enables to shed some lights on the nature of knowledge that effectively circulates through informal relations. The ultimate implication is that the co-location of firms in clusters or alike may not be a sufficient condition for accessing knowledge, as often claimed by the literature, but on the contrary, the access to knowledge may be restricted only to a few actors.

It follows that it is also interesting to investigate which actors are able to participate in those knowledge networks, and what characteristics do they have. Therefore, in order to address this issue, we focus on the structure of the knowledge network and investigate which are the firm-specific characteristics that ensure access to knowledge. These points are further discussed in the next section.

2.2 Knowledge Networks Structure and Firm Characteristics

As to begin with and in line with the above, we can argue that knowledge networks, differently from chit-chat conversations, are intentionally formed by their members. Indeed, when technicians need specific technical advice, they purposefully search out and select those colleagues who they believe are better endowed to provide effective solutions to their problems (Schrader, 1991). Since nurturing these social and professional relations is costly and time consuming—and also because “technical assistance comes with the obligation to reciprocate later on” (Carter, 1989, p. 155)—technicians turn to their smaller community of acquaintances (i.e. the epistemic or professional community), within which they search for peers whose competencies and experience are not too far removed from theirs, otherwise acquisition and reciprocation would be difficult and interactions would not generate any benefits (Bathelt et al., 2004; Boschma, 2005). Therefore, exchanges occur if a certain degree of similarity and complementarity in knowledge assets holds. The above indicates that interactions aimed at obtaining specific problem-solving knowledge do not emerge from unplanned and occasional contacts; rather they are structured and shaped according to the specific characteristics of the exchangers. The analyses, both theoretical and empirical, of the specific structures and properties of networks affecting knowledge diffusion and their relations with firm characteristics are the focus of a large body of literature (Ahuja, 2000; Coleman, 1988; Cowan, 2004; Powell & Grodal, 2005; Rauch & Casella, 2001).

Along these lines and drawing on the methodology used in a previous study that has applied social network analysis to examine knowledge diffusion in Italian and Chilean wine clusters (see Giuliani & Bell, 2005; Giuliani, 2007), we examine the structure of the knowledge network and relate it to the firm features in another Italian wine cluster.

From the seminal contributions by Nelson and Winter (1982) and Cohen and Levinthal (1990), we know that firms are able to search, absorb or share knowledge flows according to their technological capabilities (i.e. absorptive capacity) and inherited knowledge base. These capabilities affect their position and role in the local knowledge network and consequently shape its structure and, as it is shown in some recent studies, there is a positive correlation between firms’ knowledge base and centrality in the cluster's knowledge network (Boschma & Ter Wal, 2007; Giuliani, 2007; Giuliani & Bell, 2005). In this view, what really matters is the knowledge endowment of the individual firm, thus the stronger the firm's absorptive capacity, the deeper its embeddedness in the local web of knowledge ties. In the above argument, there is an implicit assumption that firms are willing to share, and also that they have knowledgeable counterparts in their neighbourhoods with which they can profitably interact. In fact, agents are keen to interact with partners able to return valuable knowledge (Schrader, 1991). Thus, firms that are better endowed with knowledge assets are at the core of the local knowledge network.

Conversely, if this assumption does not hold, we can instead expect that firms with stronger knowledge bases may search for potential partners outside the local area, with the ensuing effect of reducing their local embeddedness. This means that the actors at the core of the local knowledge network are those with weaker knowledge bases, which collaborate more intensively with other cognitively and spatially proximate actors, since they are unable to reach distant knowledge sources.

Some recent studies on industrial districts (Belussi et al., 2003; Boschma & Lambooy, 2002) have shown the emergence of a similar scenario characterized by a few leading firms, sometimes playing the role of knowledge gatekeepers, through which new ideas enter and reinvigorate the local production system. In some cases, however, lead firms can prevent knowledge leakage for strategic purposes, by sharing it with only a restricted group of partners (e.g. collaborators, subcontracting firms) and keeping it marginal for the others. In other cases, dominant firms, either because of their strategic power or their stronger competencies, may feel themselves too cognitively distant from the local system in which they are located and they therefore search for knowledge connections with outsiders. In these cases, we would expect that firms better endowed with knowledge assets are loosely connected or isolated with respect to the core of the local knowledge network.

The above discussion suggests that firms in clusters are willing to invest in local knowledge networks if they are interested in tapping into the local knowledge base; otherwise they may choose to search for knowledge outside the cluster. This is of particular relevance in a globalizing economy in which firms increasingly have many opportunities to connect with distant actors.

3. The “Colline Novaresi” Wine Cluster

The study is based on the collection of primary firm level data in the wine cluster of “Colline Novaresi”, located in Piedmont, a region producing many appellation wines, which are among some of the best known Italian wines (e.g. Asti Spumante, Barolo and Barbaresco).4 As far as exports are concerned, Piedmont is the second biggest exporting region in Italy after Veneto, with a share of 23% of total Italian wine exports in 2007.

Although its contribution to regional production is much less than the Southern part of the region, the local system of “Colline Novaresi” is historically recognized as a wine cluster, characterized by 300-odd micro- and small firms, most of them being grape growers.5 Some of the wines produced in the area are classified as wines of quality produced in identifiable regions Vin de qualité produit dans une région déterminée (VQPRD); these include four DOC wines and one DOCG. In general, local firms cultivate autochthonous vines and produce local varieties of wine; therefore, the local terroir is one of their key competitive assets.

During the last two decades, notwithstanding its old tradition, the area's contribution to total regional production (nowadays around 2%) has been decreasing, with a sharp reduction in the area dedicated to vines (the total number of hectares decreased from 12,000 in the 1950s to 800 hectares today). The average size of the wineries in the area is very small (on average 0.4 hectares), even smaller than the national average (0.8 hectares).6

Despite the average very small size of the firms, according to our findings (see Section 5.3) many local firms are not falling behind; on the contrary, they seem to have taken measures designed to increase efficiency, improve the quality of wines, adopt new technologies and introduce innovations.

The local branch of Vignaioli Piemontesi plays a leading role in encouraging modernization. Vignaioli Piemontesi is a regional association, the largest in Italy, which provides technical assistance to producers in viticulture and related fields. As discussed in Morrison and Rabellotti (2007), Vignaioli Piemontesi is a key player in the Piedmont Wine Regional Research System, connecting small and marginal producers to several sources of knowledge, such as the university in Turin and other regional and national research institutions. The extension agency contributes to diffusing information on the newest technological advancements and best practice which would otherwise not be accessible to small producers, through its magazine, through demonstrations to farmers or through a direct consultancy activity.

This process of modernization is a key strategy to survive in the wine industry, an industry known for being traditional but recently having experienced intensive technical change. In a global market characterized by a shift in demand from bulk to quality wines, and by an increasing number of competitors from the “new world”, access to knowledge is a key competitive asset. From this follows the relevance to investigate how knowledge circulates among firms, through intra-cluster linkages with respect to linkages with actors external to the cluster in a wine local system, such as “Colline Novaresi” under analysis in this article.

4. Methodology

4.1 The Population and Data Collection

Primary firm level data have been collected via a structured questionnaire enquiring about intra- and extra-cluster flows of information and knowledge. With regard to selection criteria, the firms interviewed were identified according to their activity and location: from the total population in the Novara Chamber of Commerce database, we extracted the population of wine makers (i.e. wineries that produce wine on their own), thus excluding grape growers and wine traders.7 Table 1 summarizes some of the structural characteristics of the 26 wine makers interviewed.

Table 1. The population of wine-making firms (n = 26)

	N	%
Number of employees
1–3	10	38.5
3–10	11	42.3
≥10	5	19.2
Year of foundation
1990 to today	9	34.6
1950–1990	7	26.9
Before 1950	10	38.5

Given that in the “Colline Novaresi” wine cluster, as it is very common in Italy, wineries are generally family-run and mostly individual businesses, firm's technician very often coincides with the owner. Consequently, our interviews were directed to the owner, being also the oenologist and/or the viticulturist of the firm and therefore the best informed person in the firm about its broad range of activities and external relations, in particular those related to technical issues.8

In addition to some general background information, the questionnaire includes (a) a section on firms’ economic performance (e.g. total sales, exports, main destination markets), (b) a section on innovation activities (e.g. amount of investments in new technologies, experimentation activities carried out), (c) a section on firms’ endowment of human capital (e.g. level of education and training of qualified personnel, external consultants) and (d) two relational sections on intra- and extra-cluster informal linkages aimed at exchanging information and knowledge.

Relational data were collected through a “roster recall” method: interviewees were presented with a complete list of the other firms in the cluster and were asked to name all the firms with which they exchanged knowledge and information. Similarly, a roster including the main supporting organizations located either within or outside the local area and wineries outside the cluster was shown to the firms.9

4.2 Some Definitions

In this study, we focus on flows of knowledge and information taking place through informal networks defined as linkages among individuals via face-to-face contacts, which are not mediated by marked-based mechanisms. Therefore, winery technicians (often coinciding with the owner) are depicted as nodes in the network and their linkages as flows of knowledge and information connecting them. Both the individuals connected through direct ties and those linked to them can in principle accrue benefit from the sharing of knowledge and information circulating in the network.

A key issue is how to make the distinction between information and knowledge operational (Lundvall & Johnson, 1994). Information is considered generic information (i.e. declarative knowledge or know what). Information exchanges can involve a wide range of business issues. The question is:

Do you have any informal contact with employees—or the owner—of the following firms (see the list) aimed at exchanging information (about business opportunities, markets, providers, inputs, machineries or technologies)?

(Please indicate the frequency of such interaction: none = 0; occasional = 1; frequent (every month) = 2; highly frequent (weekly) = 3).

Knowledge exchanges are defined as technical advice to solve problems and the question is:

If you experience a technical problem in your routine activity, to which of the following firms (see the list) do you turn for technical support?

(Please rate the importance you attach to the advice provided for solving your problem on the basis of the following scale: none = 0; low = 1; medium = 2; high = 3).

As a control, respondents were also asked to indicate the firms that had benefited from their technical support. These questions allow identifying the sub-networks that arise from informal conversations among peers—for example, those that oenologists conduce with colleagues to be informed about market opportunities—from those that imply a learning process. The former are likely to be highly frequent and based on weak ties. The latter entails higher stability, need of reciprocation and purposiveness. In other words, they are based on strong ties and, according to the literature, are more effective in transferring complex and tacit knowledge (Sorenson et al., 2006)

Based on the relational data, we built a directional database in which every node can be a source (outward arrow) and/or a receiver (inward arrow) of information/knowledge.

Furthermore, respondents were also asked to indicate and qualify their relationships with distant actors, in terms of both geographical and social/cognitive distance. We included in this group local and national business associations, research institutions such as universities, technical schools and laboratories, extension agencies and wineries located outside the local cluster.

In the next section, we use socio-metric techniques and graph theory (Wasserman & Faust, 1994) to analyse the structural characteristics of the networks established in the local area and to study how knowledge circulates among actors.10

5. The Main Empirical Findings

5.1 Structural Properties of the Networks

A visual examination of the information and knowledge networks (Figure 1) shows that, as expected, they are different: the former being quite dense and highly connected and the latter less dense with nodes connected by relatively fewer linkages and with some isolated nodes disconnected from the local network. The differences are confirmed by some of the measures presented in Table 2: overall density measures are consistently higher for the information network than for the knowledge network, these differences being always statistically significant. As far as non-directional ties are concerned, on average, each firm is in touch with half (50%) of the actors in the information network; conversely, the number of contacts drops when knowledge flows are considered (18%). It should be noted that such a gap is stronger for non-directional density than for mutual density. This result suggests that contacts through which knowledge circulates, although undoubtedly fewer than those in the information network, are possibly based on mutual and stronger relationships. These findings are confirmed by the average number of contacts established by each winery.

Table 2. Average density measures in the information and knowledge networks

	Type of relation	Information network	Knowledge network	Bootstrap paired sample t-test
Average density	Mutual^a	0.35 (0.47)	0.14 (0.35)	4.97∗
	Non-directional^b	0.50 (0.49)	0.18 (0.39)	6.84∗
	Valued^c	0.60 (0.94)	0.34 (0.87)	3.47∗
Average number of contacts^d	Mutual	8.88 (6.65)	3.69 (2.77)
	Non-directional	12.62 (5.46)	4.69 (2.71)
Minimum number of contacts	Mutual	2	0
	Non-directional	4	0
Maximum number of contacts	Mutual	25	11
	Non-directional	25	12
Isolated nodes^e	Mutual	2	5
	Non-directional	0	1
Degree of reciprocity (%)^f		57.23	89.23
Note: Standard deviation in brackets.
^aIn a mutual network, a tie between two nodes exists only if reciprocated. This measure is based on an asymmetric binary matrix.
^bThe density of a binary network is the total number of ties divided by the total number of possible ties. The measure is based on the symmetry of the original matrix.
^cFor a valued network, the indicator is the total of all values divided by the number of possible ties. In this case, the density gives the average value.
^dThe average number of other nodes to which a node is directly connected.
^eIn a directed graph, two vertices are in the same weak component if there is a semi-path connecting them. Two vertices, x and y, are in the same strong component if there is a path connecting x to y and a path connecting y to x.
^fWeak reciprocity: percentage of symmetric pairs, no value data.
∗Significant at 5% level.

Figure 1. (a) Information and (b) knowledge network

Table 2 also shows that in both networks, mutual density, measured on only reciprocated contacts, is lower than non-directional density, which also includes unreciprocated contacts. Moreover, the degree of reciprocity shows that knowledge exchanges are almost exclusively restricted to where ties are reciprocated; for information exchanges this is always lower. In fact, reciprocity entails high stability and trustworthiness, and is commonly associated with knowledge sharing (Schrader, 1991). This finding is in contrast to the widespread belief, usually supported only by anecdotal evidence, that local production systems are communities characterized by dense, social interactions based on reciprocity among local actors.

The heterogeneity in the relational capability of both networks, as measured by the number of contacts, is another finding that contradicts the traditional wisdom about local production systems. In the cluster under analysis, there are actors that maintain a much higher number of contacts than others, indicating a very heterogeneous access to informational resources in the local knowledge system.

Overall, the evidence presented above confirms that the structure of the information network is different from that of the knowledge network in terms of connectivity. The results also provide preliminary evidence supporting the argument that knowledge is shared within relatively smaller groups of actors with respect to information. They suggest that dense interactions (i.e. high relational capital) at local level can be observed only by assuming non-reciprocity among interacting agents.

The next step in the analysis is to investigate how the networks are structured (e.g. fragmented, polarized). This is an important issue since the partition of networks in either few or many communities (e.g. cliques, sub-groups, partitions, core–periphery structure, etc.) may affect the extent to which knowledge and information circulate within the cluster.

5.2 The Structure of the Knowledge Network

To explore how knowledge and information are shared among the local community and whether there is a dominant actor, or group of actors, we compute the degree of variability (i.e. heterogeneity) of the power and betweenness indices (see the appendix for definitions) and their degree of centralization (i.e. concentration). The degree of variability measures to what extent actors differ in terms of their abilities to produce, acquire and share knowledge, whereas the degree of centralization detects the emergence of leaders in the network. Table 3 shows that both networks have a moderate degree of heterogeneity (the mean value is higher than the standard deviation), with higher variability in the knowledge network (57.5) than in the information network (33.6). Similarly, the knowledge network is slightly more concentrated (15.7 vs. 10.35), although in both networks the degree of centralization is moderate. As far as the betweenness is concerned, we find much higher variation (the mean value is lower than the standard deviation), and a slightly, bust still moderate, concentration. Results are reversed for the two networks, pointing to the presence of more actors acting as bridges in the information network.

Table 3. Concentration and heterogeneity in the information and knowledge networks

	Power^a		Betweenness^b
	Information network	Knowledge network	Information network	Knowledge network
Degree of heterogeneity^c	33.6	57.5	163	121
Concentration (%)	10.35	15.78	20.69	15.74
^aBeta parameter: 0.100000, matrix symmetrized and dichotomized.
^bNormalized betweenness; matrix dichotomized not standardized.
^cStandard deviation divided by mean times 100.

The above results suggest that knowledge and information are not appropriated and controlled by one single actor. This means that the two networks do not resemble a typical star configuration in which a central actor controls all the communication flows. In principle, this is good news since diversity engenders learning. The moderate degree of heterogeneity, along with the low degree of concentration, suggests that some firms may be more influential in informal exchanges. This is supported by the in/out-degree and betweenness centrality indices presented in Table 4, which shows that some actors (e.g. Rovellotti and Brigatti) have achieved a position over the mean.

Table 4. Out-degree and betweenness centrality indexes

	Nodes	Information network		Knowledge network
		Out-degree	Betweenness	Out-degree	Betweenness
Nodes with the highest centrality	Rovellotti	52	12.542	22	20.585
	Brigatti	48	22.93	29	19.812
	Prolo	43	1.789	10	5.017
	Mazzoni	29	6.651	7	0.067
	Dessilani	25	8.589	7	1.684
	Miru	24	5.689	19	17.963
Nodes with the lowest centrality	Le Piane	4	0.114	7	3.363
	Zanetta	4	0	3	0
	Turetta	4	0	2	0
	Bianchi	4	0.393	0	0
	Barbaglia	2	0	9	0.525
	Ca Nova	2	0.077	3	0

Given the heterogeneity detected and the emergence of some leaders, it is interesting to assess whether a core–periphery configuration emerges out of this group of more central actors. A core–periphery structure intuitively consists of two groups of actors: on the one side, those that are tightly connected to each other (i.e. the core) and on the other side, those that are loosely connected to the core and also are not connected to each other (the periphery).11 This specific structure of network is particularly relevant because it represents the degree of cognitive polarization of the local area relating it to specific actors’ features.

Testing this hypothesis implies verifying to what extent the two networks (i.e. information and knowledge) reproduce an ideal core–periphery configuration (Borgatti & Everett, 1999; Everett & Borgatti, 1999). Taking into account the densities within groups, we focus the analysis on the knowledge network, which is characterized by such a configuration, while information flows appear rather dispersed over both groups (Table 5). In the knowledge network, we observe intense core-to-core interactions (the valued density is 2.35) compared with periphery-to-periphery connections (0.22). The core-to-periphery (0.37) and periphery-to-core (0.42) linkages, as expected, are rather sparse, although it is worth noting that a certain amount of knowledge does circulate between the two groups, especially from the periphery to the core. It is also significant that the density within the core community is several times higher than the network average density (0.34, see Table 2), confirming the existence of a cohesive core. Consequently, the remaining part of the network is close to a periphery structure with a lower than average density (0.22).

Table 5. Core–periphery structure: densities within groups^a

	Information network (all ties)^b		Knowledge network
			All ties^b		Strong ties only^c
	Core (11)^d	Periphery (15)	Core (5)	Periphery (21)	Core (4)	Periphery (22)
Core	1.41	0.84	2.35	0.37	1	0.17
Periphery	0.37	0.18	0.42	0.22	0.15	0.03
^aThe core–periphery structure has been obtained using a genetic algorithm implemented by UCINET VI. In order to test the robustness of this structure we have run the Tabu search algorithm, implemented through the same software, which provides similar results. The initial and final fitness coefficients are, respectively, 0.247 and 0.593 for the information network and 0.4 and 0.529 for the knowledge network.
^bThe density of a valued network is given by the total of all tie values divided by the number of all possible ties. In this case the density gives the average value.
^cComputed on a dichotomous matrix of the original data set. The maximum value of the density indicator in this case is equal to 1. Strong ties identify relations that provide the key contribution to solve technical problems (see the appendix for details).
^dNumber of firms in parenthesis.

To better understand the relative importance of linkages among the actors in the core and in the periphery, we examine the sub-group of the knowledge network consisting of only strong ties, providing key advice to solve technical problems (see the appendix for details). Figure 2 shows that the core of strong ties, with the exception of one, coincides with the original core; also, core nodes are among those with the highest out-degree index (measured by the size of the node). Evidence of cohesion in the core is further supported by comparing densities within the two groups (i.e. core vs. periphery). Table 5 shows that density of core-to-core interactions is equal to 1, while that of periphery-to-periphery interactions is close to 0 (0.03). Intra-periphery knowledge flows almost disappear when only strong ties are examined, indicating that they are mainly based on weak and unstable linkages. Consequently, periphery actors cannot be regarded as an interlinked knowledge community, although they maintain a few strong ties with the core. This confirms that the core is a cohesive group of highly interconnected actors who share knowledge among themselves rather than with peripheral actors.

Figure 2. Core–periphery structure of the knowledge network (strong ties only).

Given the emergence of a core–periphery configuration in the knowledge network, it is interesting to link the actors’ different position in the network to their structural and cognitive characteristics. This is the aim of the next section.

5.3 The Core and the Periphery of the Knowledge Network

In this section, we investigate the existence of a relation between an actor's position in the knowledge network and its connections with other actors internal and external to the cluster, its structural features, its performance indicators and its knowledge bases. To interpret the findings presented in Table 6, we also draw on qualitative evidence collected through in-depth interviews with technicians, entrepreneurs and key informants from the locally operating technical and research institutions (e.g. extension services, universities). In these interviews, we have explored both the motives driving the connections between firms and the different actors and the participation of the different cognitive groups of firms (i.e. core, periphery and isolated periphery) in the learning activity in the wine cluster.

Table 6. Are firms in the core knowledge network different from firms in the periphery?

Knowledge network position (no. of firms)	Firms structural characteristics (average value)
	Wine production (l)	Sales (000[euro])	Sales to large retailers (as % of all distribution channels)		Premium and super-premium wine (as % of the total production)
Core (4)	27,680	112.4	0		19.7
Periphery (8)	37,935	718.6	1.1		20.8
Isolated periphery (14)	38,500	261.5	6.6		30
	Performance indicators (average value)
	Annual growth rate (%)		Export performance (export/sales) (%)	Innovative performance^a
Core (4)	8.9		7.0	60.0
Periphery (8)	16.0		13.5	64.7
Isolated periphery (14)	17.5		30.0	75.0
	Knowledge base indicators (average value)
	% of graduates in technical fields		External oenologists (%)	Experiments (%)	Investment in new technologies (000[euro] per ha)
Core (4)	40.0		20.0	60.0	2613.2
Periphery (8)	64.7		60.0	58.8	4431.7
Isolated periphery (14)	50.0		75.0	75.0	12301.5
	Openness	Local connectivity
	% of firms with linkages outside the local system	% of firms with linkages with local business associations	% of firms with linkages with extension services	Betweenness	Out-degree centrality
Core (4)	40.0	20.0	80.0	10.8	17.2
Periphery (8)	52.9	35.2	76.0	5.7	8.1
Isolated periphery (14)	50.0	50.0	25.0	0.05	0.5
^a% of firms introducing process innovation.

Table 6 presents information about the structural characteristics, the performance and the knowledge base of the different groups of firms, identified on the basis of their position in the local knowledge network.12

Taking into account firms belonging to the overall periphery of the network, they are larger both in terms of volume and value of production; besides on average, they sell more to supermarkets13 than firms in the core, and they produce higher-priced wines.14 In terms of performance, peripheral firms have better indicators with respect to both the rate of growth and the export performance, indicating that they are more internationally connected than the core firms. Moreover, their performance is also better in terms of innovation.

In terms of knowledge base, we find that firms in the overall periphery are better endowed, in terms of knowledge assets than those at the core. In the cluster under investigation, therefore, peripheral firms are cognitively distant from the firms in the core, which build a large part of their technological and informational assets on local sources. These peripheral firms maintain stronger connections than the core with external sources of knowledge and contribute very little to the intra-cluster learning system. The relational indicators confirm that the overall periphery is the best connected to external sources of knowledge. When this finding was discussed with some key informants, they agreed that for problem-solving activities, peripheral firms rely either on their internal human resources or on direct contacts with external oenologists,15 rather than on local contacts with other firms or the extension services. Among peripheral firms, most external contacts are with firms located in the South of Piedmont, which is far more advanced and export-oriented than the cluster under investigation (see Section 3).

Concerning the small firms in the core, they are mainly connected to external sources of knowledge through the main local extension agency, Vignaioli Piemontesi, a key player in the wine system at regional level, which acts as gatekeeper in the cluster (Morrison & Rabellotti, 2007; Section 3). As shown in Table 6 and widely confirmed by our in-depth interviews, the core firms interact intensively with the local extension agency in many different ways. Many experiments and innovations undertaken by small local firms, particularly those in the core, are encouraged and often implemented with the assistance of the association's technicians. Vignaioli Piemontesi is a source of up-to-date technical knowledge because it participates directly in many research projects financed by Regione Piemonte in collaboration with scientific institutions such as universities and other research centres.

To sum up, the above evidence is indicative of a knowledge system that is centred on a weak core in terms of knowledge endowments, surrounded by a periphery of more knowledgeable actors, connected less strongly with the local area than with external knowledge sources. The emergence of this peculiar structure suggests that the process of knowledge socialization, which consists of exchanging ideas, technical advice and problem-solving activities, does not take place at level of the whole cluster, but it is restricted to few bounded communities. In the wine local system under investigation, the strongest community in terms of density of interactions is the weakest as far as knowledge resources are concerned. This implies that relying on the local knowledge basis, this community may be locked into a declining learning path. However, the local extension agency acting as a technological gatekeeper plays a key role in connecting this weak core to external sources of knowledge.

6. Conclusions

In recent years, scholars have suggested that informal contacts are relevant and effective channels for sharing information and knowledge in districts and clusters but few empirical studies have provided detailed and convincing evidence on this issue. This article contributes at filling this gap by providing empirical evidence at firm level on informal contacts in a small Italian wine cluster in order to understand the structural differences of information and knowledge networks and how firms’ specific characteristics (e.g. size, performance, knowledge base) may affect their structural position in the knowledge network as well as their extra-cluster linkages.

Our empirical findings suggest that the conceptual distinction between information and knowledge is indeed relevant in the analysis of informal networks in local production systems. In the cluster of “Colline Novaresi”, the investigated networks of informal contacts show different structures depending on what they convey: on the one hand, there is a dense information network in which actors are linked by non-reciprocal ties; on the other hand, the knowledge network is rather sparse, with a core of actors connected through mutual ties. Thus, this study provides new insights into some controversial issues. The local system analysed is not populated by a homogeneous community of entrepreneurs and technicians, sharing technical advice and generic information. On the contrary, knowledge flows, defined as technical advice, are restricted to a tightly connected community of local producers, while information is easily accessible by almost everyone. This implies that face-to-face contacts are limited in their scope and mainly serve to know what is produced and who sells it. Knowledge, however, far from being a local public good, is actually a club good, whose membership is restricted and not simply regulated by geographical proximity.

Our finding is in line with a recently increasing body of literature in which it has been shown that knowledge is unevenly distributed in clusters and that networks (of knowledge, information, business relationships, input exchanges, management ties, etc.) in clusters differ a great deal in terms of their production and diffusion mechanisms (Bell, 2005; Boschma & Ter Wal, 2007; Dahl & Pedersen, 2004; Giuliani, 2007; Giuliani & Bell, 2005; Kauffeld-Monz & Fritsch, 2007; Lissoni & Pagani, 2003; Morrison, 2008). Overlooking these differences may lead to a biased analysis and misleading implications for policy. In particular, cluster policy initiatives aimed at facilitating the access to local know-how should take into account the existence of specific communities of firms/technicians and eventually select them as policy target rather than the geographical cluster as a whole. Moreover, knowing the nature, size, cohesiveness and relations among the different networks populating a cluster may provide useful insights on how to design better services for potential users.

Furthermore, this study provides interesting findings on who participate in the local knowledge network. We find that the core of the network, sharing a great amount of local knowledge, is populated by a group of wineries that are on average smaller in size, less innovative and less open to external knowledge than the periphery. We suggest that these features (i.e. size, innovative behaviour) explain the cognitive closeness of these firms. Their gateway to external sources of knowledge is represented by their intensive connections with a local extension agency, which is a key player in the regional system of innovation for the wine industry (Morrison & Rabellotti, 2007).

Conversely, the remaining firms (the periphery), which are on average larger, more innovative and open to external sources of knowledge, can access knowledge inputs either through their outside linkages, mainly to the dynamic wine system of the Southern Piedmont, or by their better-developed internal resources. These peripheral firms are not interested in tapping into local knowledge and therefore they contribute very little to intra-cluster learning. These latter findings contrast with the cases discussed in some recent studies on networks in clusters (Boschma & Ter Wal, 2007), but are in line with other cases, for instance the Colline Pisane wine cluster discussed in Giuliani (2007), which appears to be a typical example of declining wine cluster. In other words, our results, along the lines of some existing evidence, suggests that intra-cluster knowledge connectivity is not per se associated with positive knowledge endowments or with better economic and technological performances, rather it depends on the context-specific characteristics of the cluster under analysis.16

From our results it further descends that larger and more knowledgeable firms (relative to other cluster's members) do not necessarily act as knowledge gatekeepers (Graf, 2007; Morrison, 2008). These firms in fact look for knowledge assets, which, if not available inside the cluster, are searched elsewhere. Overall, this latter evidence provides insights into the theoretical argument claiming that global pipelines “when they are too strong they could threaten the long term existence of a cluster. Strong external linkages could then provoke segmentation among the members of a cluster, reducing its coherence and threatening its long-term future” (Bathelt et al., 2004, p. 48).

This study has a number of limitations that we hope will encourage further research on this topic. The empirical analysis is based on a small number of firms in a peculiar wine cluster, so concerns may be raised about the universality of the results. As partial answers to these potential drawbacks, we stress that the firms included in the research are not drawn from a sample but they do constitute the population of the wineries in the cluster and besides, through the survey, we have collected unique relational data, which can be barely found in already existing larger surveys or secondary databases. Nonetheless, in order to strengthen the robustness of our conclusions, more empirical evidence from case studies, covering different sectors and geographical areas, is highly needed.

Moreover on a theoretical ground, this study relies on a definition of knowledge and information that can be regarded as rough, and partly blurring. We acknowledge this limitation and we believe that a richer conceptualization (on this point see, among others, Asheim and Gertler (2005)) would be helpful for a deeper understanding of the process of interactive learning and knowledge diffusion. However, on an empirical ground the operationalization of somehow vague, abstract and complex concepts into observable and measurable indicators necessarily requires for simplification.

To conclude, we suggest that some issues, only marginally touched in this study, will deserve more attention in the literature. Further research should explore themes like the role of external relations, given that network analyses in clusters have so far been focused mainly on local linkages, while little is still known on the structural characteristics of external networks. Another interesting question to address is related to external actors in clusters: are they likely to have different ability to access to local knowledge networks? Finally, the dynamics of the networks is still little explored; hence, an effort in collecting longitudinal data on inter-firm collaboration within clusters is highly welcome.

Acknowledgements

We wish to thank Elisa Giuliani for letting us access the questionnaire elaborated for her PhD thesis at SPRU, University of Sussex, and for her comments. We acknowledge the collaboration in the interviews of Ombretta Cabrio and Marco Vuturo. Thanks go to two anonymous referees, Mauro Lombardi, Mari Sako, Michael Storper and participants at seminars held at Cespri, Università Bocconi, and DRUID, Copenhagen Business School for their comments. Finally, a special thanks goes to the people interviewed, who gave us their time and knowledge besides frequent opportunities for wine tasting. Financing from PRIN “Capabilities dinamiche tra organizzazione di impresa e sistemi locali di produzione”, IRES Piemonte and Fondazione CRT—Progetto Alfieri is gratefully acknowledged.

Notes

The literature has proposed many different, sometimes overlapping, terms and definitions of geographical agglomeration of economic activities. While acknowledging the relevance of this debate (see Paniccia (2002) for a review), the analysis of the peculiarity of these different forms of agglomerations goes beyond the aim of this work.

Most of these approaches refer either explicitly or implicitly to Polanyi's (1962) conceptualization of knowledge.

We acknowledge that there are many types of knowledge which may be exchanged in informal networks as it is suggested in Asheim and Gertler (2005), who distinguish among synthetic, analytical and symbolic knowledge. Nevertheless, in the empirical exercise undertaken in this article the definition of knowledge is simplified for making it operational.

The attribution of these appellations depends on strict regulations that establish the production area, the grape varieties that can be used in a particular regional blend, the vine yield, the wine/grape yield, the alcoholic content, production and ageing methods, together with a specification of which kind of information can be put on the wine label (Odorici & Corrado, 2004). Piedmont produces 11 DOCG (Denominazione di Origine Controllata e Garantita) wines (over 32 in all Italy) and 45 DOC (Denominazione di Origine Controllata) (over 311 in all Italy), which account for almost 80% of the overall regional production and 15% of Italian production of appellation wines. This makes Piedmont the second biggest Italian producer of DOC and DOCG wines after Tuscany.

This area, known as the “wine route”, has its main nodes in four villages: Sizzano, Ghemme, Fara and Boca located in the province of Novara. The area includes 25 municipalities.

Assoenologi (1998) and http://www.provincia.novara.it/Agricoltura/TerritorioAgricolo/vite.php

The Chamber of Commerce database includes 71 firms. We have excluded grape growers, farmers producing only for self-consumption and farmers for whom wine production is not the core business. The selection has been refined, along with the criteria indicated above, with the assistance of the representatives of the two largest local associations of farmers (Coldiretti and Vignaioli Piemontesi). The total number of firms satisfying our criteria is 32, from which four producers are excluded because they had ceased their activity and two because they were not keen to answer the questionnaire.

For those few cases in which the two figures were different, we interviewed both the owner and the technician.

Since an exhaustive list of organizations and firms located outside the local area was not available, we gave respondents an incomplete list (open roster), asking them to add any other organization they have contacted. For the firms, we asked them to indicate the number of firms they had been in touch with, their location and the type of contact.

For a brief introduction to the measures and indicators used, see the appendix.

Borgatti and Everett (1999, p. 377) define the idea of core–periphery in a discrete model as follows: “the core periphery model consists of two classes of nodes, namely a cohesive sub-graph (the core) in which actors are connected to each other in some maximal sense, and a class of actors that are more loosely connected to the cohesive sub-graph but lack of any maximal cohesion with the core”.

Due to the scarceness of more fine-grained data, we acknowledge that some of the indicators used in the analysis may be considered weak from a statistical point of view. Nevertheless as far as the firm-specific indicators are concerned, the questions in our survey are in line with the standard surveys carried out at the more aggregated level (e.g. national census, CIS).

In general, only larger producers are able to sell to large retailers such as supermarkets because the requirements in terms of quality standards, size and continuity of orders are difficult to satisfy for smaller firms.

Bottles sold at an ex-producer price higher than [euro]8.

This is confirmed by evidence reported in the last section of Table 6. For example, 60% of firms in the periphery and 75% of those in the isolated periphery employ an external oenologist as a consultant.

This is in line with Ahuja (2000, p. 251, emphasis added) who claimed that “the impact of different network attributes and positions can only be understood relative to a particular context”.

See Wasserman and Faust (1994) for a comprehensive review of measures. Indicators have been computed using UCINET VI and PAJECK.

For details, see Bonacich (1987).

Appendix

The structural properties of local networks are analysed using the following indicators:17

1. Overall network density is the ratio between the total number of ties and the total number of possible ties. It is a measure of connectivity and shows the presence of some relational capital in the local system:

   where g are nodes and L identifies all lines in a graph. The density ranges from 0 (L = 0, no lines present) to 1 (all possible lines present).

2. Actor centrality degree indicates the number of nodes to which each node is directly connected. The higher the degree, the more actors access to knowledge:

   where d _i identifies the number of lines incident to it. In directional networks we distinguish between in-degree centrality (i.e. the number of in-going ties), which measures the flow of information or knowledge each actor receives, and out-degree centrality (i.e. the number of out-going ties) measuring the flows generated by each actor.

3. Average degree is a synthetic measure obtained by averaging the centrality degree:

4. Strength of ties provides a measure of the importance (in terms of frequency and quality) and efficiency of communication flows. Respondents were asked to rank their linkages in terms of frequency on a 1–3 scale (respectively, 1 in a year, month, week or more) for their information linkages, and in terms of relevance and contribution to resolving technical problems on a 1–3 scale (respectively, low, medium, high) for their knowledge linkages. Strong (weak) knowledge linkages imply key (not useful) advice on technical problems, while strong (weak) information ties identify frequent (occasional) contacts among local actors.

To investigate the presence of a core–periphery structure the following measures are used:

1. Bonacich power index 18 is a version of centrality degree. The centrality of actors depends on how many connections they develop, and how many connections their neighbours have. The more connections the actors have, and the more connections their connected actors have, the lower their power. In our context, power is an indication of an actor's ability to be a source of knowledge or information and to diffuse these flows within the local system.

2. Betweenness degree of centrality identifies those actors on the shortest path between two other actors, i.e. “in the middle”, and therefore able to control or impede the communication flow to the detriment of others. In other words, this measure reflects aspects of arbitrage:

   where i is distinct from j and k. The index minimum is 0 and the maximum is (g−1)(g−2)/2.

3. Centralization index of power and betweenness: “expresses the degree of inequality or variance in the network as a percentage of that of a perfect star network of the same size” (Hanneman, 2001, p. 65). Higher centralization (i.e. concentration) for both indexes means that one actor is the leader of the communication network:

   where C _j stands for C _D or C _B , C _j (n∗) is the highest observed value.

4. Heterogeneity index is equivalent to the coefficient of variation (standard deviation divided by mean times 100). Higher variance implies higher inequality, that is actors have different abilities to produce, acquire and control flows of information and knowledge:

   where C¯ _j is the mean actor degree index.

To analyse the linkages with organizations and external networks three indicators are used:

1. The indicator of openness measures the external relations of firms, to either firms or organizations located outside the local area. It is equal to 1 if the interviewee has contacts with firms located outside the Province of Novara, 0 otherwise.

2. The indicator of collaborations with extension services is a dummy variable equal to 1 if the interviewee has contacts with the local extension service, 0 otherwise.

The indicator of relations with local business associations is a dummy variable equal to 1 if the interviewee has contacts with local wine consortia and business associations, 0 otherwise.