What drives the formation of ‘valuable’ university–industry linkages?

Research Policy 38 (2009) 906–921

Contents lists available at ScienceDirect

Research Policy journal homepage: www.elsevier.com/locate/respol

What drives the formation of ‘valuable’ university–industry linkages? Insights from the wine industry Elisa Giuliani a,∗ , Valeria Arza b a b

Robert Schuman Centre for Advanced Studies European University Institute Via delle Fontanelle, 10 I - 50014 San Domenico di Fiesole, Italy Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET), Centro de Estudios para la Transformación (CENIT), Callao 796 6to piso, (1023) Buenos Aires, Argentina

a r t i c l e

i n f o

Article history: Received 5 July 2006 Received in revised form 20 February 2009 Accepted 23 February 2009 Available online 24 March 2009 JEL classification: O33 O31 L66

a b s t r a c t Most of the literature on university–industry (U–I) linkages assumes that these linkages are beneficial per se. We question this assumption, suggesting that not all such linkages are equally helpful. In this paper, we explore the factors driving the formation of ‘valuable U–I linkages’, conceived as those linkages between universities and firms that have a higher potential to diffuse knowledge to other firms in their regional economy. Our empirical strategy combines case-study methodology with econometric techniques using data from two wine clusters in Chile and in Italy. The firm’s knowledge base is found to be a key driver of ‘valuable’ U–I linkages. We conclude that selectivity should be encouraged among policy makers endeavouring to promote U–I linkages. © 2009 Elsevier B.V. All rights reserved.

Keywords: University–industry linkages Knowledge diffusion Wine Chile Italy

1. Introduction Universities are increasingly considered to be central actors in the economic development processes of countries and regions. In recent times, their direct involvement with industry has increased, and policies have been designed to promote university–industry (U–I) networking. However, this endeavour is raising concerns about the costs and time consumed by U–I networking that may be detrimental to university research. It is suggested first, that university research has a value per se, independent of whether or not it is connected directly with industry, because it keeps alive curiosity-led investigation, which is a cultural value worth transmitting to succeeding generations. Second, it is also suggested that universities whose links with industry are too intensive become interested in more industry-driven, short-term, problem-solving research—which might undermine researchers’ intellectual freedom in the definition of research agendas and the way that research results are used. This tension has sparked a debate on whether or

∗ Corresponding author. Current address: DEA, Facoltà di Economia, University of Pisa, Via Ridolfi 10, 56124 Pisa, Italy. Tel.: +39 050 2216280; fax: +39 050 541403. E-mail addresses: [email protected] (E. Giuliani), [email protected] (V. Arza). 0048-7333/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.respol.2009.02.006

not U–I linkages should be promoted, which has important implications for policy-making (see for example Poyago-Theotoky et al., 2002). We believe that both benefits and costs are involved in U–I relations and thus acknowledge the relevance of this debate. However, our critique is constructive and we do not attempt to classify U–I linkages as an aspect to be either necessarily supported or limited. Rather, we argue that some U–I linkages are more ‘valuable’ than others in terms of their higher potential to diffuse ‘knowledge’,1 thus generating positive effects on the economy. In the spirit of the debate over U–I linkages, we would argue that, from a policymaking perspective, it would be desirable, therefore, to support only the creation of valuable linkages. But, what are the factors that favour the formation of ‘valuable’ U–I linkages? We consider this an important but under-explored question; the literature tends to focus simply on what affects the formation of such linkages, assuming that they will per se have a beneficial effect.

1 By ‘knowledge’ we mean here “a fluid mix of framed experience, values, contextual information, and expert insight that provides a framework for evaluating and incorporating new experience and information. It originates and is applied in the minds of knowers” (Davenport and Prusak, 2000: 5). Therefore, interaction between agents increases the potential of knowledge diffusion.

E. Giuliani, V. Arza / Research Policy 38 (2009) 906–921

More specifically, our interest here is to explore the factors that influence the formation of linkages between universities, and firms that are more likely to diffuse their knowledge to other firms located in the same regional cluster (which we consider makes the U–I linkage valuable). To do so, we use an original dataset of wine producers in two wine-producing areas, in Chile and Italy. We analyse the data using econometric techniques that allow us first, to estimate the probability of forming U–I linkages and second, to estimate the degree of knowledge diffusion from any linkages formed, among the regional clusters of firms. The paper is organised as follows: Section 2 presents the conceptual framework and elaborates our research hypotheses. Section 3 discusses the research context and Section 4 describes the methodology. Section 5 presents our empirical results and Section 6 concludes.

2. Conceptual framework 2.1. Fortifying U–I linkages—an open debate For those interested in economic development, the role that universities can play in enhancing regional and national innovations systems is a matter of great and increasing interest. Several studies have suggested that universities can be central players in economic systems (Charles, 2003; Cooke, 2001; Dasgupta and David, 1994; Kitagawa, 2004; Lundvall, 1992; Nelson, 1993, 2004; Salter and Martin, 2001). Also, scholars have promoted the idea that universities should go beyond their traditional teaching and research activities, and undertake a ‘third mission’, aimed at more direct interaction with and contribution to industry (Etzkowitz and Leydesdorff, 2000; Slaughter and Leslie, 1997). A growing number of studies documents the existence and the drivers of linkages between universities and industry (Anselin et al., 2000; Arundel and Geuna, 2004; Bonaccorsi and Piccaluga, 1994; Bruno and Orsenigo, 2003; Cohen et al., 2002; Fontana et al., 2006; Fritsch and Schwirten, 1999; Geuna, 2001; Gregorio and Shane, 2003; Hall et al., 2003; Kaufmann and Todtling, 2001; Link and Scott, 2003; Meyer-Krahmer and Schmoch, 1998; Mowery et al., 2001; Santoro and Chakrabarti, 1999; Slaughter et al., 2002; Tornquist and Kallsen, 1994; Van Looy et al., 2003; Velho and Saez, 2002). These studies highlight the number of different ways facilitating the formation of U–I linkages, including employment by industry of university graduates, informal meetings and joint research programmes involving both parties, industry commissioned consultancy not involving original research, licensing of university patents, university purchase of industry developed prototypes, etc. (e.g. Cohen et al., 2002; D’Este and Patel, 2007; Meyer-Krahmer and Schmoch, 1998; Schartinger et al., 2002; Slaughter et al., 2002). Many of these studies attempt to identify the firm, industry and university characteristics that affect the probability of U–I linkages being formed (e.g. Anselin et al., 2000; Arundel and Geuna, 2004; Bruno and Orsenigo, 2003; Cohen et al., 2002; Fontana et al., 2006; Jaffe, 1989; Lee, 1996; Santoro and Chakrabarti, 1999; Tornquist and Kallsen, 1994), while others analyse the extent to which they are beneficial to firms’ innovative performance (Kaufmann and Todtling, 2001). Enthusiastic views on the importance of U–I linkages are counterbalanced by the concerns voiced over: (i) the goals of public research; and (ii) the appropriation and use of research outputs. These views usually support the historical role of universities as generators of public knowledge. In term of the goals of public research, the main issue is whether it should be directed to solving concrete problems in industry. This presumes that such a focus undermines researchers’ intellectual freedom in the definition of research agendas and the way the results of their research are

907

used and made public (Blumenthal et al., 1997; Louis et al., 2001; Nelson, 2004; Tapper and Salter, 1995), which are seen as harmful to the long-term creative potential of universities. In this view, U–I linkages are seen as time-consuming and distracting/costly and potentially detrimental to university research (Crespo and Dridi, 2007; Slaughter and Leslie, 1997). In effect, it is argued that universities with very intensive links to industry may become more interested in short-term, consultancy-based research, rather than long-term fundamental investigation. Some authors also highlight that the alternative sources of funds for universities based on their relationships with industry, reduce government responsibility to support university research, which may penalise basic science (Lee, 1996) or may bias research agendas toward more profitable applied research activities. In terms of the appropriation and use of knowledge, problems may arise because companies usually want exclusive rights on their inventions, while a public university’s priority should be wide diffusion of the knowledge it creates (Jelinek and Markham, 2007; Pickering et al., 1999; Santoro and Chakrabarti, 1999; Slaughter et al., 2002). In some countries, this tension has sparked intense reactions against policies promoting U–I linkages (e.g. De Sousa Santos, 2006; Llomovatte et al., 2006). This paper contributes to the debate in contending that, for the reasons outlined above, the formation of U–I linkages implies an important opportunity-cost for universities. This opportunity-cost affects university researchers and, indirectly, can have an impact on the whole economic system—for example, in terms of reduced generation of public knowledge or, in the case of developing countries, lines of research being diverted away from development policy goals. From a policy-making perspective, therefore, in our view it would be better to foster U–I linkages with a higher potential for generating knowledge spillovers (i.e. ‘valuable’ linkages), and reduce the generation of potentially less valuable ones. This leads to our research question of what factors favour the formation of ‘valuable’ U–I linkages, which we think is a very important yet under-explored area. To the best of our knowledge, there have been no published attempts to identify the factors that affect the quality of U–I links; the literature tends to focus on what it is that affects the formation of U–I linkages, and assumes that they will per se have a beneficial effect. 2.2. The drivers of valuable U–I linkages: a model In this section, we develop a set of research hypotheses about the factors that could affect: the formation of U–I linkages (Section 2.2.1); and the transfer of knowledge by firms with U–I linkages, to other firms in their regional clusters, i.e. to assess the value of such linkages (Section 2.2.2). 2.2.1. Factors affecting the formation of U–I linkages In line with much of the recent literature, we consider firmand university-level factors that may be associated with a higher tendency toward U–I linkages: 2.2.1.1. Firms’ knowledge bases. Several studies have explored how firm characteristics affect the formation of different types of linkages to universities and public research organisations (Arundel and Geuna, 2004; Cohen et al., 2002; Fontana et al., 2003; Laursen and Salter, 2004). Our main interest here is on a key dimension of firms’ internal characteristics: the knowledge base (KB), defined here as the “set of information inputs, knowledge and capabilities that inventors draw on when looking for innovative solutions” (Dosi, 1988: p. 1126). Knowledge is seen as residing in skilled knowledge workers in firms and as being accrued and generated through their experimentation efforts, to both exploit and explore new ways to solve problems (Nelson and Winter, 1982).

908

E. Giuliani, V. Arza / Research Policy 38 (2009) 906–921

We believe that this an important variable explaining the formation of U–I linkages, because firms with stronger knowledge bases through their enhanced absorptive capacity (Cohen and Levinthal, 1990), have better capabilities for searching and exploiting valuable external knowledge, one source being universities. This view is corroborated by several studies that show that firms with higher R&D intensity have more university collaborations (Arundel and Geuna, 2004; Fontana et al., 2003; Schartinger et al., 2002). Hence, we can formulate a first hypothesis: Hypothesis 1. The stronger the firm’s knowledge base, the higher will be the probability that the firm forms linkages with universities and other public research organisations. 2.2.1.2. Scientific quality of university departments. The propensity to form U–I linkages also depends on the characteristics of university departments. Mowery and Sampat (2004), for example, suggest that public research organisations vary in structure, size and strategy and, therefore, should not be considered homogeneous entities. Here, we consider a key aspect of university departments: their scientific quality. Several scholars have explored the degree to which the scientific quality of universities influences the formation of U–I linkages and obtained contradictory results. On the one hand, some studies find that top tier universities or departments establish more U–I linkages than those with less high quality scientific records. For instance, in a seminal study, Mansfield and Lee (1996) look at whether the quality of universities affects the likelihood of their being linked to and supported by industry. By asking firms to cite academic researchers whose work, in the previous decade, contributed most importantly to their companies’ new products and processes, they show that the top four US universities have a significant number of linkages to industry. In a more recent study, Bruno and Orsenigo (2003) analyse the determinants of industrial funding of universities in Italy and find that it is the quality of university research and staff, rather than the strength of industry demand that is the main driver. Hence, they conclude that, in the context analysed, U–I linkages are not very intensive because Italian universities are characterised by low-quality research and, thus, have little to offer to industry. Other studies also find that the propensity to form linkages with universities is associated with the quality of the universities (e.g. Tornquist and Kallsen, 1994) or the quality of researchers (e.g. D’Este and Fontana, 2007). On the other hand, some studies find that lower academic scientific quality is associated with higher U–I linkages. For instance, even Mansfield and Lee (1996, p. 1055), observed that second tier universities had significant interactions with industry, but that these interactions were of a different nature to those observed in top tier universities and involved mainly applied R&D and with nearby firms. This result is in line with D’Este and Patel (2007), which uses UK data and finds that poorly rated university departments seem to engage in a wide range of U–I interactions in the case of applied disciplines. They see this as being due first, to the fact that lower quality departments may have more limited access to public funding (especially in systems where public research funds are allocated on the basis of research assessment exercises), making industry funding a necessity. And second, those lower quality departments probably have a higher proportion of researchers with limited interest in or ability to undertake ‘blue sky’ research, and who prefer engagement in less ambitious and more problem-solving research for industry. Other scholars reach similar conclusions: for example, in developing countries some have raised concerns that the research skills of university scholars are being undermined due to funding pressures, which drive universities to provide consultancy services to industry (e.g. Arocena and Sutz, 2005; Kruss, 2006; Vega-Jurado et al., 2007).

On the basis of these opposing findings we can formulate the following hypotheses: Hypothesis 2a. University departments with higher scientific rating have a higher probability of linking with industry. Hypothesis 2b. University departments with lower scientific rating have a higher probability of linking with industry. 2.2.2. Factors affecting the diffusion of knowledge by firms with U–I linkages We have referred to the fact that U–I linkages are more valuable when there is a higher potential for knowledge diffusion (Section 1). In other words, we consider a linkage to be valuable when the firm that is collaborating with a university actively helps to diffuse the knowledge to other firms in its regional cluster. These interfirm linkages are in the form of direct and indirect knowledge ties established by the firm informally with other firms in a regional cluster, for example, to solve a technical problem (von Hippel, 1987). U–I linkages will have much less value if the firms involved are ‘dead ends’ in the U–I knowledge pipeline, that is, despite their links with a university, these firms do not diffuse the knowledge to other firms in the cluster (see also Bell and Giuliani, 2007).2 It is important, therefore, to explore whether firms that have established linkages with universities also contribute to the diffusion of knowledge to other firms. Previous research has shown that there is a positive relationship between the propensity of firms to establish knowledge linkages to other firms and the relative strength of firms’ knowledge bases in their regional cluster – which is also consistent with Cohen and Levinthal (1990). For example, Giuliani (2007), using micro-level data on fine wines producers, in three wine clusters located in Italy and Chile, shows that firms with stronger knowledge bases transfer more innovation-related knowledge to other firms in the cluster. Similarly, Steiner and Ploder (2008), using network data on the region of Styria in Austria, find that firms with significant R&D capacities are more central in the region’s R&D networks than other actors. One reason for this is that these firms may be perceived by other firms as local ‘technological leaders’, leading to their being more frequently sought out as sources of innovationrelated advice and knowledge, than firms with weaker knowledge bases. Several studies point to the existence of a positive association between firms’ technological capabilities or R&D activities, and their knowledge linkages (among many others, Dantas, 2006; Fischer and Varga, 2002). In this paper, we are interested in the group of firms that have connections with universities. In particular, we want to identify the drivers that increase the potential for knowledge diffusion of this specific group of firms. The literature provides various insights. On the one hand, a number of works show that firms that make intensive use of external knowledge are likely to focus their internal capabilities on only certain areas of advanced knowledge, supplementing other areas with external sources (including universities) (e.g. Dussauge et al., 2000; Miotti and Sachwald, 2003; Velho and Saez, 2002; Veugelers and Cassiman, 2005). This behaviour may impinge on their potential for knowledge diffusion: it is possible that the type of knowledge that these firms are able to transfer is too specialised to be of general interest to other firms in the nearby area. On the other hand, some suggest that, in certain industries, firms need to develop internal capabilities in a variety of areas of knowledge, even if they mainly rely on external sources for some

2 We make the reasonable assumption that among the firms that are connected to universities, those that have knowledge linkages to other firms in their regional cluster have more potential to diffuse university-generated knowledge than those that have little or no linkages at all.

E. Giuliani, V. Arza / Research Policy 38 (2009) 906–921

of this knowledge. This may be because it may be difficult for the firm to control the timing of knowledge or technology transfer from external sources (Brusoni et al., 2001). In these cases, U–I linkages do not lead to over-specialisation by firms in certain technological areas. We build on this second view and argue that, in linking with universities, firms do not seek to complement their knowledge bases but rather to strengthen their capabilities. This is particularly the case in the wine industry, the context of our study, where firms internally process the knowledge acquired from universities and exploit it in their innovative or production processes, leading to the generation of new skills and capabilities. Our view, therefore, is that there is no reason to assume that U–I linkages would specialise firms’ knowledge bases to the extent that their knowledge would not be of interest to other firms in the cluster. Rather, we suggest that the effect of firms’ knowledge bases on the potential of knowledge diffusion for the group of firms with U–I linkages does not differ from the positive effect found in the literature for firms in general. On this basis, we propose the following research hypothesis: Hypothesis 3. The stronger the knowledge bases of firms with linkages to universities, the higher their potential to diffuse knowledge to the cluster. 3. The context 3.1. The wine industry The wine industry, the context of the present study, has recently undergone radical worldwide transformations, both in market characteristics and wine production (Archibugi, 2007). While a comprehensive discussion of the wine industry is beyond the scope of this paper, we refer briefly to the most significant changes in the industry since the 1990. First, the traditional Mediterranean wine-producing countries (i.e. Italy, France, Spain and Portugal) have been challenged by producers in other continents: Australia, South Africa, the USA, Argentina and Chile among others, which have become major producers and exporters of wine (for statistics, see Anderson and Norman, 2001; Smith, 2007). Second, the total volume of wine produced is declining, but the quality of production is increasing, which suggests lower consumption of wine but with a higher unit price (Archibugi, 2007). As Archibugi (2007 p. 125) suggests “the era in which a litre of honest un-bottled table wine was cheaper than a litre of gasoline still exists in Italian, French and Spanish towns, but is drawing to a close”. Thus, the intrinsic quality of wine has improved in absolute terms with respect to the past—although there are differences across market segments. One aspect that makes the wine industry an appropriate context for testing our research hypotheses is the importance of university research for introducing improvements in the industry. As reported by Paul (1996), scientists have played a key innovative role in the wine industry since the 1860s’ phylloxera outbreak,3 when “science played a large role in the constitution of the devastated vineyards, and especially in the preservation of quality wines from vinifera vines” (Paul, 1996 pp. 10–11). Most of the advancements in agronomic, chemical or engineering research on the wine industry are based on applied science, carried out in public research organisations or universities and leading to directly applicable solutions for the industry. For example, there are great benefits from discover-

3 Phylloxera is a pest that attacks certain types of vines, particularly the vitis vinifera, and is endemic in Europe. In the second half of the 19th century, it destroyed most of Europe’s vineyards, most notably in France. Current European vines are grown on a pest resistant rootstock, which was introduced following research in universities.

909

ies such as the relationship between the incidence of certain viral pathogens and the decrease in grape yields, a greater understanding of the relationships between different types of yeasts and the sensory attributes of wines, or even about advice on irrigation in the face of climate change.4 Hence, universities are seen as the source of skills and knowledge, with which firms can engage in joint research activities.5 University researchers also commonly offer formal or informal technical advice to the wineries on matters such as how to recognise and treat a particular pest, or how to analyse a soil. Such U–I interactions are widely documented in the literature (Aylward, 2003; Giuliani et al., 2008; Loubere, 1990; McDermott et al., 2007; Morrison and Rabellotti, 2007; Smith, 2007). 3.2. The two wine clusters This study is focused on two countries: Italy and Chile. It looks at U–I linkages between national universities (and/or public research organisations) and wine producers located within a regional wine cluster—Bolgheri/Val di Cornia in Italy and Valle de Colchagua in Chile. The characteristics of these two wine clusters have been described elsewhere by one of the authors of this paper (see, e.g. Bell and Giuliani, 2007; Giuliani, 2007). Here, we would briefly point to the fact that, in spite of being located in two very different countries, these clusters share some fundamental characteristics. They are of similar size (in both cases areas of about 50 km from north to south), and both are rural areas densely populated by fine wine producers and grape growers, with other firms in the wine production value chain – upstream and downstream – located outside the cluster territory, making the vertical division of labour within the clusters rather shallow. Each cluster has a business association, whose primary aim is the promotion of wines and the marketing of the local wine route, with other actors (universities, suppliers, clients, etc.) located outside the cluster boundaries. In both cases, the clusters have a long history of wine production, but have experienced economic growth only since the 1980s, when worldwide consumption of high quality wines began to grow. 4. Methodology 4.1. Data collection Our study is based on micro-level firm data collected in the two wine clusters, through face-to-face interviews with skilled workers (i.e. oenologists and agronomists) in charge of the production process at plant level in the firms. The surveys were carried out in 2002 and addressed to the entire population of fine wine producers in the two clusters; in all 41 in Bolgheri/Val di Cornia (Italy) and 32 in Valle de Colchagua (Chile) were interviewed, making a total of 73 firms.6 Apart from general background and contextual information, the interviews were also aimed at gleaning information to enable the development of quantitative indicators in three key areas: (i) firms’ knowledge linkages with national universities or other public research organisations. This information was collected through a roster that included a number of specific institutions,

4 Examples given in personal interviews with Chilean and Italian university researchers. 5 A recent initiative to formally bring together problem-solving in the industry with scientific research in the universities has been launched by the Chilean consortium VINNOVA (http://www.vinnova.cl/EN/), whose website reports examples of collaborative projects that have been mutually advantageous to the industry and the universities. 6 Grape-growers are not included in this study.

910

E. Giuliani, V. Arza / Research Policy 38 (2009) 906–921

e.g. particular universities, and other public research organisations, which had emerged as being relevant during the pilot fieldwork. Firms were asked: (1) Could you mark, among the actors included in the roster, those that have transferred relevant technical knowledge to this firm? (2) Could you mark, among the actors included in the roster, those with whom this firm has collaborated in research projects during the last two years? Respondents were able to add any other relevant universities or public research institutions not included in the roster; (ii) firms’ knowledge linkages with other firms in their regional clusters. The collection of relational data was based on a roster study and on the following two questions: (1) “If you are in a critical situation and need technical advice, to which of the local firms mentioned in the roster do you turn?” and (2) “Which of the following firms do you think have benefited from technical support from your firm?”. These questions capture the transfer of inter-firm knowledge directed to the solution of a technical problem or to the transfer of technical know-how, á la von Hippel (1987). This is because this paper focuses on technological innovation. Also, as a consequence of the limited division of labour within clusters, as discussed in Section 3.2, only horizontal linkages among fine wine producers are mapped by this survey. Wine producers reported vertical linkages, e.g. with suppliers of machinery, enzymes, chemicals, etc., all of which are external to the cluster. Vertical linkages with grape-growers within a cluster were not relevant as channels of knowledge, as growers do not appear to play a critical role in the processes of innovation in the two regional clusters, and therefore are not included in this study. This also applies to the local business associations, which do not play a role in the process of technological innovation. This focus on horizontal knowledge linkages is consistent with other studies that highlight their importance in innovation (among many others, see Lissoni, 2001; von Hippel, 1987). (iii) a set of variables for firm characteristics that allow the operationalisation of key concepts (e.g. firm knowledge base). The questionnaire included enquiries about: (1) the number of technical employees with a graduate degree (BSc, MSc, MPhil, Dhil) in a technical discipline; (2) level of higher education obtained (BSc, MSc, MPhil, DPhil); (3) number of months of employment in wine industry firms, specifying whether domestic or foreign; (4) experimentation activity—type and area of the production process. We also collected information about university departments with links to at least one firm in the cluster.7 In particular, we collected information about: (i) the quantity and quality of scientific research of the universities, gathered through the ISI Web of Knowledge Science Citation Index Expanded—1954 to present;8 and (ii) geographical distance to the industry cluster. For the former type of data, we searched on the names of individual departments in each university, whose discipline was connected to agronomy or oenology, for example, the University of Florence search included the Interfaculty Department of Vegetal Biology, the Department of Agronomic Engineering, the Department of Soil and Plant Nutrition, the Department of Agrarian Biotechnologies, and so on (a full

7 In the Chilean case, these turned out to be the University of Chile, the Catholic University in Santiago, the University of Talca and the Instituto Nacional de Investigacion Agropecuaria (INIA) in Santiago (La Platina). In the Italian case, they were the Universities of Pisa, Florence and Milan, and the Istituto Vitivinicolo di Conegliano Veneto and the Agenzia Regionale per lo Sviluppo e l’Innovazione nel settore Agroforestale (ARSIA). 8 Although it misses other relevant research outputs, use of publications and citations in ISI journals as measures of output and impact provides comprehensive and consistent metrics for all researchers.

list of the departments searched, for each university, is provided in Appendix A).9 The selection criteria were designed to exclude publications about humans or animals or specifically non-relevant vegetal species (e.g. potatoes, tomatoes, beans, etc.). 4.2. Operationalisation of variables 4.2.1. Dependent variables The paper identifies factors that affect the formation of ‘valuable’ U–I linkages. As discussed in Section 4.4, our research question requires the estimation of models for two dependent variables: the measure of U–I linkages and the measure of firms’ knowledge diffusion to other firms in the regional cluster. The two variables were built as follows: (i) UI LINK measures the existence of a linkage between the firm (i) and a university (u). It is a binary variable that takes the value of 1 if firm (i) has at least one linkage with the university (u). This variable includes two types of linkages: those formed for joint research and those formed through the transfer of technical knowledge from university to industry (see questions in Section 4.1 (Point i)). (ii) DIFFUSION measures the degree to which a firm diffuses innovation-related knowledge to other firms in the regional cluster. We use a standard measure of social network centrality, known as out-closeness (Freeman, 1979; Wasserman and Faust, 1994), calculated from network data obtained through the responses to the questions reported in Section 4.1 (Point ii).10 This indicator is based on directed network data and takes account only of out-going linkages, thus measuring the degree to which knowledge is transferred from firm (i) to other firms. Out-closeness is slightly more sophisticated than other measures of centrality as it takes account not only of the direct number of linkages established with other firms (e.g. for degree centrality), but also of the extent to which the firm manages to ‘reach’ other firms in the cluster to which it is not directly connected. To understand out-closeness, we need to think about knowledge diffusion as a multi-step process in which a distinct piece of knowledge goes from firm (i) to its direct contacts (or ‘alters’) and from the alters to other firms (Rogers, 1983). According to how many direct connections firm (i) has, and also to the connections that its alters establish, firm (i) can be more or less close to multiple other firms. A firm has high outcloseness if it can reach the rest of the firms in its cluster in a minimum number of steps. In technical terms, out-closeness is measured as the reciprocal of the sum of the shortest paths (or geodesics) between firm (i) and the other firms in the network. The measure has been normalised as data come from networks of two different sizes, based on dividing this value by the maximum value of closeness obtained for the specific network. Isolated nodes take the value 0. The indicator was computed by UCINET (Borgatti et al., 2002) and details on its measurement can be found in Wasserman and Faust (1994). We use this measure as an indicator of knowledge ‘diffusion’ because it captures, not only knowledge transfer through direct linkages with alters, but also the degree to which the knowledge transferred to the alters can potentially reach other firms through a

9 Double checks were performed by searching on the researchers’ names listed on each selected department’s web page. 10 Relational data collected through the questions are expressed in matrix form. The matrix is composed of n rows and n columns, corresponding to the number n of firms in each cluster. Each cell in the matrix represents the existence of knowledge being transferred from firm i in the row, to firm j in the column.

E. Giuliani, V. Arza / Research Policy 38 (2009) 906–921

limited number of steps. This latter is a critical aspect because the higher the number of steps involved in the diffusion process, the greater the downgrading of the original knowledge transferred.11

911

Table 1a Statistics on Principal Component Analysis of the components of firms’ knowledge bases: correlation matrices. Chile

4.2.2. Independent variables 4.2.2.1. Firm knowledge base (KB). Hypotheses 1 and 3 test the impact of the firm’s KB on its likelihood to be connected to a university (Hypothesis 1) and to diffuse knowledge to other firms in the cluster (Hypothesis 3). We are aware of the endogeneity problems generated by similar variables – e.g. R&D – in estimating the effects on the formation of linkages or on different types of innovative output (e.g. Cassiman and Veugelers, 2002; Veugelers and Cassiman, 2005). Here, we operationalise the firm’s knowledge base using stock variables rather than flows, however, which minimises any endogeneity effects. Consistent with previous studies using the same set of data (e.g. Giuliani, 2007; Giuliani and Bell, 2005), KB is a composite indicator of three dimensions of the firm’s knowledge base: (i) formal training of human resources; (ii) experience of human resources in the field; and (iii) firm’s experimentation intensity. While (i) and (ii) refer to the human resources at the time of the interviews (2002), (iii) takes account of experimentation activities in the two previous years because the pilot fieldwork showed that the time span of two years gave a good idea of the intensity of experimentation carried out by a wine producer. These variables are defined as follows: (i) Formal training of human resources (HR): represents the cognitive backgrounds of each firm’s skilled knowledge workers based on their education (degree) level. We assume that the higher the level of education the stronger the cognitive background of the firm. On this assumption we weight each skilled knowledge worker according to the degree obtained: HR = 0.8 × First Degree + 0.05 × Masters + 0.15 × Doctorate A weight of 0.8 is applied to the number of graduates employed in the firm, including those with higher levels of specialisation. We add a further 0.05 for employees with a masters degree and a further 0.15 for employees with a PhD.12 We only take account of degrees and higher levels of specialisation in technical and scientific fields related to the activity of wine production (i.e. agronomy, chemistry, etc.) (ii) Human resource experience (MONTHS): represents the working experience of the above types of employees, in temporal terms. Time is indicative that accumulation of knowledge has occurred via ‘learning by doing’. The variable is the result of a weighted mean of months of work of each skilled knowledge worker in the country and abroad: MONTHS = 0.4 × number of months (national) +0.6 × number of months (international)

11 By definition, all connections in the network imply some degree of knowledge transmission. However, the knowledge transferred by firms connected to universities, to other firms in their cluster, may not necessarily have originated from within the university. It could have come from internal firm exploration or other types of connections. However, among firms that maintain linkages to universities, it is reasonable to suppose that those with higher out-closeness have more potential to diffuse university-generated knowledge. 12 These weights are set ad hoc. We have carried out sensitivity analysis trying different weights and this did not change the results, also because the number of human resources with Masters and PhD is limited and unlikely to change the overall measure of firm ‘knowledge base’.

HR MONTHS EXPE

0.79 0.45

Italy MONTHS

HR

MONTHS

0.54

0.72 0.61

0.46

Table 1b Statistics on Principal Component Analysis of the components of firms’ knowledge bases: main statistics on factor analysis. Chile

HR MONTHS EXPE

Italy

Factor Loadings

Uniqueness

Factor Loadings

Uniqueness

0.89 0.92 0.75

0.21 0.15 0.43

0.92 0.85 0.80

0.16 0.27 0.37

We give a higher weight to time spent working abroad because the diversity of the professional environment might stimulate more active learning behaviour and a result in a steeper learning curve. We consider only learning experiences in the wine industry. (iii) Experimentation effort (EXPE): is a proxy for knowledge creation efforts. This is measured on a scale from 0 to 4, according to the number of areas in which experimentation occurs: for example, if a firm experiments in all the production phases – from the introduction of different clones or varieties in the vineyard ‘terroir’, management of irrigation and vine training systems, and fermentation techniques, enzymes and yeast analysis to analysis of the ageing period – the firm will score 4 for experimentation intensity while a firm that conducts no in-house experimentation will score zero. The different areas of experimentation were determined after extensive interviews with industry experts during the pilot fieldwork for this research.13 Although these variables measure different aspects of firms’ knowledge bases, they are highly correlated—especially HR and MONTHS (>0.7 in both countries, see Table 1a). This high correlation of the variables points the need for the construction of a composite indicator of firms’ knowledge bases using factor analysis. This technique is used to discover underlying dimensions (or factors) that largely or entirely explain the behaviour of a larger number of observed variables. We use Principal Component Analysis (PCA) to extract the factors, since our aim is to summarise the behaviour of the three variables described above in a single variable, which we can interpret as being representative of the firm’s knowledge base.14 We did separate PCA for Chile and Italy. A single factor was extracted in each case, representing 73% of data variation in each case. We calculated factor scores to define the variable firm Knowledge Base (KB). These scores are the weighted average of the

13 This involved some 40 interviews with experts including university researchers, oenologists and agronomists (both consultants and full-time employees) in small and large wineries, experts in business associations, wine consortia, and other organisations. The information was complemented by data from specialist wine industry publications. 14 Factor analysis requires ratio or interval data, not available for our ordinal EXPE. The restrictive assumption we make is that distances between ordinal values are sensitive to measurement of intensity; we apply the same restriction to all respondents.

912

E. Giuliani, V. Arza / Research Policy 38 (2009) 906–921

Table 2 Descriptive statistics about firms and universities. Chile Firm level variables Number of Firms Experience (MONTHS)a Human resources (HR)a Experimentation (EXPE)a No experimentation Low experimentation High experimentation

Italy

Count Avg. Avg. Avg. Proportion Proportion Proportion

32 67.79 1.71 1.59 0.28 0.50 0.22

41 28.66*** 0.45*** 1.10*** 0.54*** 0.29*** 0.17

AGE

Years (Avg.)

25.44

25.46

Type of firms Independent (ORG 1) Group: vertically integrated (ORG 2) Group: vertically desintegrated (ORG 3)

Proportion Proportion Proportion

0.22 0.13 0.66

0.07*** 0.00*** 0.93***

Foreign firms (OWN) Exports over sales (EXPORT) Employees (FSIZE) Connection to the cluster (DIFFUSION)a

Proportion Avg.% Avg. Avg.

0.19 60.94 55.59 10.43

0.02*** 32.07*** 7.98*** 6.58***

University level variables Number of Universities Scientific Quality (UQUAL)a University Scale (USCALE) Geographical distance to universities for connected firms (GEOD)

Count Avg. Publications (Avg.) km (Avg.)

4 0.47 36.75 169.60

5 0.76*** 116.20*** 146.84

a ***

See methodology for definition. Difference is significant at 1%.

standardised version of the three variables that define our KB indicator (HR, MONTHS, and EXPE).15 Factor loadings and uniqueness are reported in Table 1b. 4.2.2.2. Scientific quality (UQUAL). We took the average number of citations received by publications in wine-related fields from the university departments listed in Appendix A, as an indicator of university scientific quality. To measure UQUAL, we normalised the average number of citations per publication by the number of years since publication to control for the fact that older publications are likely to have more citations through the effect of time as opposed to quality. We considered the ISI publications of researchers in the departments, for the period 2000–2002—that is, the years when U–I linkages were formed according to the responses to the questionnaires. This implies that the data include citations to earlier publications by authors employed in the departments in those years, which can be justified by the fact that a particular researcher has a value, in terms of departmental scientific quality, which is captured by the average number of citations she/he receives in the course of her/his career, not just in the recent past. 4.2.3. Control variables The estimations use a set of control variables from the literature that are associated with our dependent variables. As firm-level variables, we consider: • firm size, measured by the number of employees in 2002 (FSIZE); • firm age, measured as the number of years since operations started, to 2002 (AGE); • firm ownership, which is a dummy variable indicating whether the firm is foreign - (1) or domestic-owned (0) (OWN); • type of organisation—the firms in the clusters have three types of organisational structures: ORG1 refers to firms (foreign or domestic) that are independently owned and that perform all production phases in the cluster in which the headquarters is also

15 We use the Bartlett method to calculate weights, which are based on factor loadings.

located; ORG 2 is firms that are part of a domestic group and perform all phases of the production process within the cluster; and ORG3 is firms that are part of a domestic group, but perform only part of the production process, usually grape-growing, within the cluster; • share of exports (EXPORT), which is measured as the percentage of firm sales going to export. For the universities, we control for size although we do not have information on number of employees (or any other direct measure for size). Instead, we use a very rough measure of scale, based on total number of publications in wine-related fields appeared in the ISI Web of Knowledge Science Citation Index for all the departments of each university that are indicated in Appendix A (USCALE). This is a cumulative variable that takes account of the volume of research undertaken in the past by the university.16 We try also to control for geographic distance of the university from the wine industry because geographic proximity to industry has been shown to favour interaction and the generation of knowledge flows (e.g. Abramovsky et al., 2007; Anselin et al., 2000; Arundel and Geuna, 2004; Jaffe, 1989). We measure this as the distance (in km) between the cluster (based on the location of its main village), and the town or city where the university is located. However, as we only have information on universities with at least one connection to the cluster, we cannot capture the real influence of geography, because not all potential universities are included. As discussed below, our results remain the same if we include or exclude this control for distance. 4.3. Main characteristics of each wine cluster Table 2 presents the mean values of all variables used in the estimations to characterise each cluster. It highlights some dif-

16 Note that this variable is highly correlated with number of researchers with an ISI publication, in the universities considered here (Pearson coefficient: for Chile 0.98; for Italy 0.99). Hence, we can reasonably argue that number of publications reflects number of researchers.

E. Giuliani, V. Arza / Research Policy 38 (2009) 906–921

ferences between Chilean and Italian firms; in particular, in the Chilean cluster the proportion of foreign-owned firms is significantly higher than in Italy; and Chilean firms are larger, export more, have higher indicators for KB—especially in the quality of their skilled human resources (HR, MONTHS). Also, on average, Chilean firms show significantly higher values for DIFFUSION than Italian firms. In terms of universities’ characteristics, universities in Italy are bigger and of higher scientific quality (UQUAL), than those in Chile. This in part is as expected given the indicators used to measure quantity and quality. It is likely that researchers in Italian universities have more opportunity to publish internationally, given their location in Europe and especially compared to a South American country university. Finally, the average distance between the regional cluster and its universities is 170 km in Chile and 147 km in Italy. In Chile, three out of the four universities included in the study are in the capital, thus dispersion in distance is small: the coefficient of variation is only 7%. In contrast, the Italian universities are much more dispersed, which implies a higher variance in geographical distance (coefficient of variation is 79%).

4.4. Econometric estimation As already mentioned, in this paper we explore two types of effects: first, the factors affecting the probability of formation of a U–I linkage (UI LINK) and, second, the factors affecting the degree of connectivity of the firms linked to universities (DIFFUSION). Since UI LINK and DIFFUSION may not be independent events, we estimate a two-stage Heckman model, which estimates in a single model the probability of a U–I linkage and the degree of knowledge diffusion of the linking firms. In the first step, a Probit model estimates the likelihood of connecting to a university (Eq. (1)). This allows us to estimate the expected residuals of the truncated observations of DIFFUSION (i.e. only for firms linked to universities) used in the second step. The second step is a correct specification of an Ordinary Least Squares (OLS) estimation (Eq. (2)) controlling for the fact that UI LINK and DIFFUSION might not be independent events. We estimate robust and cluster standard errors to account for heteroschedasticity and to control for correlation of errors for observations belonging to the same university. UI LINK = [KB, UQUAL, USCALE, GEOD, FSIZE, OWN, AGE, ORG2, ORG3] (1) DIFFUSION = [KB, FSIZE, OWN, AGE, ORG2, ORG3, ; EXPORT] (2) This method requires identification of the variables that have a significant effect on the first step (Eq. (1)), but do not affect the second step (Eq. (2)), which we achieve through the inclusion of three variables for university characteristics. This should not affect the degree of firm links within the cluster, but will affect the likelihood of connection to a university. We do separate estimations for Chile and Italy. As will be seen that the correlation of residuals in the first and second steps (), is not significant, which means estimation of the Heckman model is unnecessary and separate Probit and OLS can be consistently estimated and should be preferred because it is a more efficient strategy. As expected, the results from these two procedures do not vary significantly nor do they change our conclusions regarding the explanatory variables. We report all our results in the section that follows.

913

5. Empirical results 5.1. Descriptive statistics Table 3 reports the descriptive statistics for the differences between firms with and without established linkages to universities. It shows that in both Chile and in Italy, about two-thirds of the firms in the cluster maintain at least one link with a university (23 out of 32 firms in Chile and 27 out of 41 firms in Italy).17 The comparison is done systematically for both countries. Table 3a reports the mean values (or proportions) of firmlevel characteristics, according to whether or not firms have established U–I linkages, and indicating where differences are statistically significant. The first striking result is that differences across firms, with or without U–I linkages, are much more marked in Chile than in Italy. In Chile, firms with U–I linkages have stronger knowledge bases, reflected in the training and experience of their human resources and intensity of experimentation; they are also significantly younger and larger, tend to be independently owned and export a larger share of their sales. In Italy, firms with links to a university do not differ significantly from those with no links. The only significant difference among Italian firms is related to experimentation: those firms that do conduct experimentation activities are more likely to have a linkage with a university than not. However, there are no differences across firms with high levels of experimentation intensity. In terms of university characteristics, Table 3b shows that in Chile, U–I linkages are more likely to be with larger universities and universities with high scientific quality, while in Italy, the universities with the most linkages with industry appear to be those with the poorest scientific record. In Italy, university size does not seem to have an effect on firm linkages. In both countries, it seems that the universities that are more geographically proximate to the cluster, on average have more linkages with industry. In the context of this study, it is interesting that Chilean firms with university links also have higher DIFFUSION values (14.49 for firms with U–I linkages vs. 7.65 for firms without U–I linkages), while this is not the case in Italy (DIFFUSION is around 7 in both cases). Based on the discussion in Section 1, it could be argued that firms in the Chilean cases are more likely than firms in the Italian case to establish ‘valuable’ linkages. In other words, in the Chilean case, firms that establish linkages with universities have relatively high potential to diffuse the knowledge to the rest of the cluster, while in the firms in the Italian case that establish connections with universities are not necessarily well-connected within the cluster. This finding is confirmed by the correlation analysis for DIFFUSION and UI LINK, as well as by the partial correlation between these variables when controlling for firms’ knowledge base: for Chile the correlation coefficient is significant while for Italy it is not.18 Section 5.2 explores the drivers of ‘valuable’ U–I linkages. We identify the main drivers of U–I linkages for both clusters (Hypotheses 1, 2a and 2b) and also the drivers of knowledge diffusion to the industrial cluster for firms with UI linkages (Hypothesis

17 Note that these data should be interpreted with caution because they include only universities with at least one firm linkage, i.e., although our data include the population of wine producers in each cluster, the list of universities does not include the whole population of national universities. Thus, UI LINK = 0 refers to the absence of a U–I linkage between firm (i) in the cluster, and the universities identified by the remaining sample firms (j). An individual firm may have a link with only one university rather than the whole set of universities included in the study. 18 The correlation coefficient for DIFFUSION and UI LINK is 0.46 (at 1% significance) for Chile and 0.04 (and not significant) for Italy. If we control for KB, the correlation coefficient for Chile is 0.25 (at 1% significance) and 0.06 (and not significant) for Italy.

914

E. Giuliani, V. Arza / Research Policy 38 (2009) 906–921

Table 3 Descriptive statistics of variables included in the models. Indicator

Chile

Italy

UI LINK = 0

UI LINK = 1

UI LINK = 0

UI LINK = 1

Number of firms Number of Observationsa

Count (%) Count (%)

9 (28%) 76 (58%)

23 (72%) 52 (42%)

14 (34%) 160 (78%)

27 (66%) 45 (22%)

(a) Firm level variables Experience (MONTHS)b Human resources (HR)b

Avg. Avg.

45.67 1.22

100.12*** 2.44***

30.20 0.47

1.07 0.42 0.49 0.09

2.37*** 0.08*** 0.52 0.40***

1.06 0.57 0.25 0.18

−0.36 32.34

0.53*** 15.35***

0.01 26.01

−0.04 23.51

23.17 0.40

Experimentation (EXPE)b No experimentation Low experimentation High experimentation

Avg. Proportion Proportion Proportion

Knowledge Base (KB) AGE

Avg. Years (Avg.)

Type of firms Independent (ORG 1) Group: vertically integrated (ORG 2) Group: vertically desintegrated (ORG 3)

Proportion Proportion Proportion

0.09 0.12 0.79

0.40*** 0.13 0.46***

0.07

0.09

0.93

0.91

Foreign firms (OWN) Exports over sales (EXPORT) Employees (FSIZE) Knowledge diffusion in the cluster (DIFFUSION)b

Proportion Avg.% Avg. Avg.

0.13 47.36 38.67 7.65

0.27** 80.79*** 80.33*** 14.49***

0.03 30.91 8.88 6.44

0.02 36.22 4.76 7.05

0.40 34.20 174.07

4 0.58*** 40.48** 169.60**

0.79 121.90 200.60

(b) University level variables Number of PRO Scientific Quality (UQUAL)a University Scale (USCALE) Geographical Distance (GEOD)

Count Avg. Publications (Avg.) km (Avg.)

1.22 0.42* 0.44** 0.13

5 0.66*** 95.93 146.84**

a Measured on the basis of all possible connections that could have been formed between the population of cluster firms and the universities included in the study. That makes 128 (i.e. 32 times 4) in Chile and 205 (41 times 5) in Italy. b See methodology for definition. * Difference is significant at 10%. ** Difference is significant at 5%. *** Difference is significant at 1%.

Table 4 Econometric results on the probability of U–I linkages. Chile (obs 128)

Italy (obs 205)

Model 1 1st stage Heckman

Model 2 Probit

Model 3 1st stage Heckman

Model 4 Probit

Model 1 1st stage Heckman

Model 2 Probit

Model 3 1st stage Heckman

Model 4 Probit

KB

0.486*** [0.175]

0.489*** [0.176]

0.443*** [0.168]

0.446*** [0.168]

0.115 [0.148]

0.114 [0.149]

0.119 [0.145]

0.118 [0.147]

UQUAL

6.827*** [0.836]

6.811*** [0.772]

3.781*** [1.091]

3.783*** [1.098]

−3.364*** [0.364]

−3.412*** [0.074]

−2.608*** [0.146]

−2.537*** [0.209]

USCALE

0.220*** [0.031]

0.218*** [0.029]

−0.008*** [0.003]

−0.009*** [0.003]

0.003*** [0.000]

0.003*** [0.000]

0.002*** [0.001]

0.002*** [0.001]

GEOD

0.391*** [0.056]

0.390*** [0.053]

0.001 [0.001]

0.001*** [0.000]

FSIZE

0.008** [0.003]

0.008** [0.003]

0.007** [0.003]

0.007*** [0.003]

−0.030** [0.012]

−0.029** [0.012]

−0.030** [0.012]

−0.029** [0.012]

AGE

−0.015*** [0.004]

−0.015*** [0.004]

−0.016*** [0.004]

−0.016*** [0.003]

0.002 [0.003]

0.002 [0.003]

0.001 [0.003]

0.002 [0.003]

ORG2

−0.487 [0.776]

−0.502 [0.802]

−0.38 [0.606]

−0.389 [0.618]

ORG3

−1.304* [0.762]

−1.307* [0.768]

−1.160** [0.586]

−1.161** [0.585]

−0.2 [0.329]

−0.202 [0.324]

−0.19 [0.324]

−0.203 [0.317]

OWN

0.075 [0.132]

0.054 [0.172]

0.052 [0.129]

0.038 [0.150]

−0.102 [0.649]

−0.101 [0.643]

−0.107 [0.648]

−0.095 [0.627]

Constant Term

−78.181*** [10.551]

−77.979*** [10.049]

−1.063 [0.654]

−1.034 [0.666]

1.460** [0.671]

1.473*** [0.301]

1.189*** [0.371]

1.173*** [0.382]

LL Pseudo R2  (p-value) Overall correct selection

−178.59***

−44.9*** 0.4576

−182.3***

−50.61*** 0.4146

−229.56***

−96.66*** 0.1041

−229.96

−97.297*** 0.0982

* ** ***

1.38 (0.23) 83%

Difference is significant at 10%. Difference is significant at 5%. Difference is significant at 1%.

84%

2.06 (0.15) 80%

81%

0.03 (0.86) 79%

79%

0.33 (0.56) 78%

79%

E. Giuliani, V. Arza / Research Policy 38 (2009) 906–921

915

Table 5 Econometric results on the intensity of knowledge diffusion within the cluster. Chile (obs 128) Model 3 2nd stage Heckman KB

FSIZE

AGE

ORG3

EXPORTS

LL R2 * ** ***

11.281*** [0.375]

−0.005* [0.002]

−0.009 [0.042]

2.809*** [0.317] −0.040** [0.012]

−0.038 [0.239]

−0.069 [0.241] 0.018* [0.008]

0.03 [0.032]

0.032 [0.030]

11.153*** [0.429] −3.479 [3.225]

5.096*** [0.572]

5.199*** [0.491]

−1.007 [0.731] 0.133 [2.483]

−2.711 [2.431] 4.556*** [0.962]

4.504** [0.980]

0.111*** [0.017]

0.109*** [0.017]

10.711*** [1.819]

−182.3***

−0.518 [0.773]

11.215*** [1.998] −0.001 [0.008]

0.081** [0.041]

8.600* [2.960] −0.977* [0.514]

0.009 [2.315] 8.662*** [0.605]

0.077*** [0.009]

no U–I link U–I link

2.735*** [0.360]

−0.098 [3.909]

no U–I link U–I link

Constant

−0.009 [0.038]

no U–I link U–I link

3.057** [0.552]

Model 5 OLS 3.661*** [0.102]

−0.018** [0.003]

no U–I link U–I link

OWNER

−0.005*** [0.002]

no U–I link U–I link

Model 3 2nd stage Heckman

0.014** [0.003]

no U–I link U–I link

ORG2

3.149*** [0.534]

no U–I link U–I link

Model 5 OLS 3.923*** [0.558]

no U–I link U–I link

Italy (obs 205)

0.084 [0.044] 7.053*** [0.669]

5.736 [7.604]

3.41 [4.950]

−229.96 0.9162

0.7248

Difference is significant at 10%. Difference is significant at 5%. Difference is significant at 1%.

3). The results will allow some speculation about why U–I linkages in the Chilean case are ‘valuable’ compared to those in the Italian case. 5.2. Results of econometric estimations Table 4 presents Eq. (1) (U–I linkages) estimated separately for each country. Model 1 reports the 1st stage of the twostage Heckman model. These results are comparable to results from Model 2 (Probit model). Models 3 and 4 respectively are the 1st stage Heckman and Probit model estimations, when GEOD is dropped from the analysis.19 Below we read the

19 As explained in the methodology, GEOD does not fully capture the effect of geographical distance because only connected universities are included in our sample. Thus our results for GEOD are spurious. In Models 1 and 2 we find that a greater distance from the university seems to increase the probability of a link with industry, which goes against the indications in the literature. Also, GEOD is highly correlated to the other variables for university characteristics. The Pearson correlation coefficient for GEOD and UQUAL is −0.68, and for GEOD and USCALE is −0.99 for Chile, while GEOD and UQUAL is 0.68 for Italy. We decided, therefore, to drop GEOD from

results from Model 4, our preferred model for Eq. (1) on U–I linkages. Regarding what determines the probability of a U–I linkage, the econometric analysis corroborates the descriptive analysis reported in Table 3: Chilean firms with strong knowledge bases (KB) have more links with universities compared to similar firms in Italy. In Italian firms, KB is not a significant variable in explaining the probability of formation of a U–I linkage. Therefore, Hypothesis 1 is validated in the case of Chile, but is rejected in the case of Italy. For Hypotheses 2a and 2b, our results reflect the ambivalence in the literature. On the one hand, we find positive and significant results for Chile, indicating that the higher the university’s scientific quality, the more likely will be a link with industry, which is in line with Hypothesis 2a. On the other hand, we find the reverse results for Italy: the coefficient for UQUAL is negative and significant,

the analysis to avoid problems of multicollinearity. The differences in the results for USCALE and UQUAL between Models 1 and 2, and 3–4 may be explained by the multicollinearity between GEOD and the other university variables in Models 1 and 2.

916

E. Giuliani, V. Arza / Research Policy 38 (2009) 906–921

indicating that the lower the university’s scientific record, the more are its links with industry. Hence, we find support for Hypothesis 2b. Among our control variables, both firm size and the university scale are significant factors affecting the creation of U–I linkages in Chile and in Italy. According to Model 4, in Chile, large firms are more prone to connect to universities, while small universities are more likely to be connected to industry. In Italy we find the opposite results: small firms and large universities are more likely to establish linkages. Table 5 presents the results from Eq. (2) (Diffusion) estimated using the Heckman model (Model 3, 2nd stage)20 and an OLS model (Model 5). Results for Model 5, which is our preferred model for Eq. (2) (Diffusion), are described below For knowledge diffusion within the cluster, our findings are consistent for both clusters that firms with stronger KB are associated with higher levels of DIFFUSION. These results are valid for all firms, regardless of whether or not they are connected to a university. This supports Hypothesis 3 for both clusters: the stronger the knowledge base of firms with linkages with universities, the higher will be its level of knowledge diffusion to other cluster firms. This means that, among firms that connect to universities, those with a stronger knowledge base will be better diffusers of knowledge – including university-generated knowledge – to other firms.21 In terms of our control variables, it should be noted that the determinants of knowledge diffusion to the cluster vary for the group of firms connected to a university vis à vis the other group. For the group of firms with U–I linkages, we observe that small, foreign, and export-oriented firms in Chile show a higher potential to diffuse knowledge within the cluster; while in Italy size is not significant, but ownership and export orientation affect connection to the cluster in the same direction as in Chile. 5.3. Robustness checks As already mentioned, firm KB is a composite indicator of three firm characteristics: formal training of human resources (HR), working experience of skilled human resources (MONTHS), and experimentation (EXPE). In order to confirm that this composite indicator really accounts for the firm’s knowledge base, we run robustness checks by estimating Models 4 and 5 using the individual components rather than the composite KB indicator. Since all three elements are highly correlated (see Table 1a), we need to take account of multicollinearity. Therefore, we include them separately. We first build an indicator for skills22 and then include dummy variables to account for different levels of experimentation.23 The results of these robustness checks are presented in Tables B1 and B2 in Appendix B. We estimate two specifications of Model 4 (in Table B1) and two specifications of Model 5 (in Table B2). These specifications include the variables for skills and experimentation separately, rather than KB (knowledge base) as in Models 4 and 5 in Section 5.2. The results in Table B1 are consistent with those in Table 4 and show that, in Chile, the probability of linking to a univer-

20 We report the 2nd stage Heckman model based on a 1st stage estimation without GEOD as the independent variable. Were GEOD to be included in the 1st stage, the 2nd stage results would be almost unchanged. For this reason we do not report them, but they are available upon request to the authors. 21 We also tested for the squared effect of KB but results were not significant. 22 This variable is constructed from the Bartlett factor scores extracted from MONTHS and HR, our two indicators of workforce skill. 23 We include two dichotomous variables for experimentation: low [EXPE = (1 or 2)] or high [EXPE = (3 or 4)]. The baseline corresponds to no experimentation EXPE = 0.

sity is higher for firms with a skilled workforce, which is not the case in Italy. Similarly, in Chile, firms involved in intensive experimentation have a higher probability of connecting to a university than firms that do experiment very little or not at all, whereas in Italy, firms with low levels of experimentation are more likely to connect to a university. The results in Table B2 are consistent with those in Table 5. In both Chile and Italy, the more skilled the workforce, the more likely that firms (both connected or not to a university) will diffuse knowledge to other cluster firms. Similarly, we find that in both clusters experimentation positively affects the degree to which firms diffuse knowledge to other cluster firms, regardless of whether they have links to a university.

6. Conclusions This paper contributes to the debate on whether or not U–I linkages should be promoted to enhance innovation and development. We address particularly the debate over the beneficial effects of U–I linkages and propose an alternative and novel view. We suggest that some U–I linkages will inevitably be more ‘valuable’ than others, based on the different potential for knowledge diffusion of firms that establish U–I linkages. Thus, we framed a research question related to exploring what are the factors that influence the formation of ‘valuable’ U–I linkages. Based on an original dataset of wine producers in two regional wine clusters, in Chile and Italy, we estimate both the probability U–I linkages being formed and the potential for knowledge diffusion within the regional cluster. We use Heckman two-stage models, Probit and OLS models. The results are interesting and somewhat unexpected. In Chile, we find that the probability of forming U–I linkages increases with the strength of the firms’ knowledge base and the scientific quality of universities. Also, the potential to diffuse knowledge to other wine producers in the regional cluster is higher for firms with stronger knowledge bases, for both Chilean firms with university links and those with no links. In the Italian case the results are different. We find that the probability of forming a U–I link decreases with the scientific quality of the university, and that the firm’s knowledge base is not significant in the formation of U–I linkages—meaning that firms with weak knowledge bases are as likely to establish linkages with universities as firms with strong knowledge bases. However, as in the case of Chile, the firm’s knowledge base is a significant and positive determinant of the potential for knowledge diffusion to the cluster, regardless of whether the firm is or is not connected to a university. In sum, in both cases, the stronger the knowledge base of the firms that establish links with universities, the more likely that those linkages will be ‘valuable’ in the sense of increased potential for diffusing university-produced knowledge. Our results suggests that, in Italy, universities forge a number of linkages with firms that are ‘dead-ends’ in the knowledge pipeline—as in the case of firms with weak knowledge bases, that tap into university knowledge, but do not transfer the knowledge to other firms in their regional cluster. This is consistent with the descriptive analysis in Section 5.1, which presents the Italian case as one where linkages are less ‘valuable’ than in the Chilean cluster. The econometric results provide some justification for this conclusion and provoke speculation that the patterns of U–I linkages in these countries are different. In Chile, linkages seem to be somewhat more ‘selective’, with the ‘best’ firms connecting to the ‘best’ universities, therefore giving rise to ‘valuable’ U–I linkages. In Italy, however, U–I linkages seem to be more ‘pervasive’, irrespective of the internal qualities of the universities or firms, with the result that there are many more ‘non-valuable’

E. Giuliani, V. Arza / Research Policy 38 (2009) 906–921

U–I linkages—for example, in the case of linkages with ‘dead-end’ firms. These findings are important for conceptual, methodological and policy-making reasons. In terms of conceptual advancement, this paper is original as it proposes a new perspective on the U–I linkages literature, and suggests more research attention on the factors driving the formation of ‘valuable’ linkages, rather than to U–I linkages per se. As discussed in Section 1, we believe that, on the one hand, while it is important that university-generated knowledge is diffused – directly or indirectly – as much as possible within the industry, on the other hand, there is an opportunity cost involved in researchers forming linkages with industry. This paper also makes a contribution in combining econometric analysis and case-study methodology. This is a relatively rare approach seldom in the literature, case-study methodology usually being associated with qualitative, story-telling analysis while econometric approaches are usually applied to secondary data. In this case, the paper benefits from the qualitative insights of case-study research and the rigour of quantitative analysis. In terms of policy-making, we believe that this paper is informative and interesting. First, it suggests less rhetorical and pessimistic investigation of Latin American countries than is usually found in the literature. Also, although among European countries Italy may not be the best example, this study suggests that the Chilean model of selective linkage formation seems to be more reasonable, and more justifiable from a policy-perspective, than the pervasive model of linkages found in the Italian case. Although we have no direct evidence on which to establish such a claim, it is possible that in Chile, this model might be the result of policies adopted during the 1990s with respect to U–I linkages. By this we mean Chile’s competitive bidding schemes to finance innovative projects and new research ventures by firms and universities that decide to collaborate in strategic industries (e.g. Programas de Fomento (PROFO) promoted CORFO, the national agency for industrial development).24 Given the fact that the allocation of funds under these programmes is typically based on the quality of applicants and on the potential economic impacts of proposed projects (Echevarria et al., 1996), it is possible that such schemes might have instilled a competitive mentality in Chile which stimulates the best actors to collaborate.25 This attitude is generally absent in Italy, where selection is often not based on meritocracy or the quality of the actors (Margottini, 2008). Also, the absence of a systematic and fully fledged funding scheme to connect universities to agro-industry and the wine industry in particular, might be responsible for the formation of the many informal ties between universities and ‘dead-end’ firms. Anecdotal evidence suggests that the formation of such ties is often driven by the need either for universities to find land or vineyards suitable for experimentations, or by the needs of wine producers with a lack of skilled labour, to access ‘free’ advice on specific matters. On the basis of these speculations, our recommendation for policy is that the promotion of U–I linkages should be selective, and that selection should be based on factors that enhance knowledge diffusion at the regional or national level. Ideally, ‘good’ universities

24 Natural resources are central in Chile’s development strategy; thus, funding schemes have especially targeted agriculture among manufacturing in general, and are particularly generous in the area of wine production. 25 Although aimed at overcoming the constraints in small firms (Benavente and Crespi, 2003), anecdotal evidence suggests that competitive bidding schemes have led to the selection of the best candidates among universities and firms, both because they may belong to oligarchic elites with higher lobbying power and because the weakest actors often lack the resources to undertake the administrative load involved in an application.

917

should have a small number of links, to ‘good’ firms with internal knowledge bases that allow the absorption and improvement of university-generated knowledge and its diffusion to other firms in the economy. University researchers should reduce the number of linkages to dead-end firms, leaving more time for them to improve their research skills, generate public knowledge and maintain autonomy in research agendas. Based on this reasoning, we would suggest that, especially in developing countries, ‘weaker’ universities should be supported to improve their internal scientific qualities, rather than being pushed to becoming problem-solvers for industry. This study has certain empirical and methodological limitations. The first is it is a single industry study, which means that the generalisation of results is bounded by the specificities of the wine industry. However, we consider the wine industry to be a suitable context to explore the importance of university research in promoting technological change in industry, which is widely documented in the literature. Second, in this study we use a narrow definition of U–I linkages that includes only collaborative research and technical knowledge transfer between universities and firms. Similarly, we operationalise the concept of ‘valuable’ U–I linkages, focusing merely on the potential for knowledge diffusion to other firms in a regional cluster. This limitation is driven by the availability of a unique dataset mapping inter-firm knowledge linkages at cluster level (for a discussion on the limitations of these data see Giuliani, 2007), and other scholars might want to develop alternative measures for ‘valuable’ U–I linkages. The third limitation is that our measure of university scientific quality includes only published papers included in the ISI database, which implies that we miss other relevant research outputs, such as books, patents and publications in journals not listed in the ISI database. The fourth limitation relates to the lack of some control variables for university characteristics; this should be taken seriously when interpreting the results. For example, we do not control for the university’s financing strategy, and universities that rely on industry as a major source of funding may be more open to the formation of U–I linkages than universities that are fully financed by government. However, this concern is mitigated by the fact that, while there are some differences across countries, there is no significant variability among different universities within the same country, so it is plausible to believe that this variable would not affect our results substantially. However, further studies might also want to consider additional or different university-related variables. Finally, lack of any analysis of the performance or growth of the two clusters is another limitation. This aspect is beyond the scope of the present study, but raises interesting questions for future research and thinking. We know that both clusters have grown quite dynamically since the end of the 1980s; thus, it would be useful explore the consequences of the observed patterns of U–I linkage formation in the longer term. For example, it would interesting to investigate whether the ‘selective’ process of U–I linkage formation leads to faster growth rates in terms of the overall performance of cluster firms than the ‘pervasive’ model, or whether the ‘selective’ model enhances more skewness in the distribution of firm performance (i.e. those connected to universities grow faster than the rest) than in the ‘pervasive’ model. These are intriguing questions which we leave open for future research.

Acknowledgments The authors are grateful to Rikard Stankiewicz, Aldo Geuna, Finn Valentine, David Mowery, Paul David, Pablo D’Este Cukierman, Ed Steinmueller, Luke Haywood, three anonymous referees and to the

918

E. Giuliani, V. Arza / Research Policy 38 (2009) 906–921

members of the European Forum Seminars (2004–2005) at the Robert Schuman Centre for Advanced Studies (European University Institute), for helpful comments on previous versions of this paper. The collection of data for this research benefited from the support of Marcelo Lorca Navarro, Cristian Diaz Bravo, Erica Nardi, Elena Bartoli, Giovanni Mattii and Roberto Pelletti. We also acknowledge financial support from by the European Jean Monnet Fellowship Scheme (European University Institute) and the UK Economic and Social Research Council (PTA-026270644, PTA-030200201739). The usual disclaimers apply. Appendix A. List of university departments and other public research organisations 1. Chile a. University of Talca: Facultad Ciencias Agrarias (Dep. Horticultura, Dep. Produccion Agricola y Centros Tecnologicos), Instituto de Biología Vegetal y Biotecnología. b. University of Chile (Santiago): Facultad Ciencias agronòmicas (Dep. industria y enologia, Dep. de ingenieria y suelos, Dep. de recursos, Dep. de sanidad vegetal, Dep. de producción agricola). c. Catholic University (Santiago): Facultad Ciencias agronòmicas y Facultad Ciencias Biologicas (Dep. de sanidad vegetal, Dep. de producción agricola; Dep. de ciencias vegetales,

Dep. de fruticultura & enologia, dep. Dep. chem & bioproc engn; Dep. Ecologia, Dep. de omol & enol, Dep. de ciencias forestales). d. INIA (Santiago): – 2. Italy a. University of Pisa: Facoltà di Agraria (Dip. di Coltivazione e Difesa delle Specie Legnose, Dip. di Agronomia e Gestione dell’Agroecosistema, Dip. di Biologia delle Piante Agrarie, Dip. di Chimica e Biotecnologie Agrarie). b. University of Florence: Facoltà di Agraria (Dip. interfacoltà di biologia vegetale, Laboratori di botanica agraria e forestale, Dip. di ingegneria agraria e forestale, Dip. di ortoflorofrutticoltura, Dip. di scienze del suolo e nutrizione della pianta, Dip. di scienze agronomiche e gestione del territorio agroforestale, Dip. di biotecnologie agrarie, Dip. di scienze e tecnologie ambientali forestali.) c. University of Milan: Facoltà di Agraria (Consorzio A&Q, Istituto di idraulica agraria, Istituto di ingegneria agraria, Istituto di patologia vegetale, Dip. di produzione vegetale, Dip. di scienze biomolecolari e biotecnologie, Dip. di scienze e tecnologie alimentari e microbiologiche, Dip. di scienze molecolari agroalimentari) d. Agenzia Regionale per lo Sviluppo e l’Innovazione del settore Agricolo e Forestale (ARSIA). e. Research Institute of Conegliano Veneto.

Appendix B. Robustness checks Table B1 Probit models on the probability of U–I linkages, using the components of knowledge base as explanatory variables. Chile (obs 128) Model 4 Probit with SKILLS SKILLSa

Italy (obs 205) Model 4 Probit with EXPE

0.284** [0.112]

Model 4 Probit with SKILLS

Model 4 Probit with EXPE

0.014 [0.166]

Low EXPEb

0.684* [0.360]

0.667*** [0.217]

High EXPEb

1.638*** [0.511]

0.134 [0.421]

UQUAL

3.696*** [1.044]

3.987*** [1.118]

−2.525*** [0.202]

−2.660*** [0.191]

USCALE

−0.009*** [0.003]

−0.009*** [0.003]

0.002*** [0.001]

0.002*** [0.001]

FSIZE

0.009*** [0.003]

0.008*** [0.002]

−0.024** [0.012]

−0.031** [0.012]

AGE

−0.017*** [0.003]

−0.017*** [0.003]

0.002 [0.004]

0.004 [0.004]

ORG2

−0.298 [0.616]

−0.852* [0.507]

ORG3

−1.269** [0.545]

−1.080* [0.578]

−0.319 [0.383]

−0.508** [0.234]

OWN

0.032 [0.131]

−0.163 [0.133]

−0.096 [0.588]

0.224 [0.554]

Constant Term

−0.951 [0.612]

−1.787** [0.903]

1.231*** [0.413]

1.205*** [0.198]

LL Pseudo R2 Overall correct selection

−52.13*** 0.3971 79%

−49.49*** 0.4392 79%

−97.60*** 0.0953 78%

−93.59*** 0.1325 78%

a b * ** ***

This variable is the score of the principal components between MONTHS and HR (see Section 5.3). These are dummy variables. Low EXPE corresponds to values 1 or 2 of EXPE and High EXPE corresponds to values 3 or 4 of EXPE (see Section 5.3). Difference is significant at 10%. Difference is significant at 5%. Difference is significant at 1%.

E. Giuliani, V. Arza / Research Policy 38 (2009) 906–921

919

Table B2 OLS models on intensity of knowledge diffusion within the cluster, using the components of knowledge base as explanatory variables. Chile (obs 128) Model 5 OLS with SKILLS SKILLSa

Low EXPEb

High EXPE 3

FSIZE

AGE

ORG2

ORG3

OWNER

EXPORTS

Constant

R2 a b * ** ***

Italy (obs 205) Model 5 OLS with EXPE

Model 5 OLS with SKILLS

no U–I link

3.861** [0.744]

2.549*** [0.133]

U–I link

3.511*** [0.392]

2.267** [0.571]

Model 5 OLS with EXPE

no U–I link

4.071* [1.541]

2.716*** [0.524]

U–I link

10.040*** [1.449]

3.285** [0.883]

no U–I link

5.479*** [0.429]

13.369*** [0.331]

U–I link

6.987** [1.780]

8.622* [4.025]

no U–I link

0.016** [0.004]

0.028*** [0.004]

−0.007 [0.013]

−0.084*** [0.011]

U–I link

−0.010*** [0.002]

0.020*** [0.003]

−0.031 [0.224]

−0.315 [0.242]

no U–I link

−0.021*** [0.003]

−0.040*** [0.004]

0.01 [0.007]

0.043** [0.009]

U–I link

0.018 [0.027]

−0.005 [0.039]

0.026 [0.025]

0.052 [0.048]

no U–I link

1.004 [3.772]

−1.524 [5.413]

0 [0.000]

0 [0.000]

U–I link

12.582*** [0.300]

9.805*** [1.086]

0 [0.000]

0 [0.000]

no U–I link

−3.819 [2.881]

−3.24 [4.145]

−3.419** [0.744]

5.166*** [0.505]

U–I link

5.183*** [0.289]

4.025** [0.798]

−0.866 [2.456]

−0.201 [4.974]

no U–I link

−3.39 [2.244]

−3.128 [3.752]

8.612*** [0.513]

11.247*** [0.502]

U–I link

4.698*** [0.665]

6.073** [1.092]

10.731*** [1.701]

13.683*** [1.721]

no U–I link

0.086*** [0.008]

0.066** [0.015]

0.025** [0.008]

0.004 [0.008]

U–I link

0.117*** [0.010]

0.037 [0.023]

0.098* [0.039]

0.077 [0.059]

no U–I link

8.136* [2.607]

5.387 [4.306]

8.376*** [0.732]

−2.242*** [0.452]

U–I link

−1.302 [0.604]

−2.86 [1.641]

3.786 [4.759]

1.795 [7.005]

0.924

0.899

0.701

0.742

This variable is the score of the principal components between MONTHS and HR (see Section 5.3). Dummy variables created from EXPE. Low EXPE corresponds to values 1 or 2 and High EXPE to values 3 or 4. Difference is significant at 10%. Difference is significant at 5%. Difference is significant at 1%.

References Abramovsky, L., Harrison, R., Simpson, H., 2007. University Research and the Location of Business R&D. The Economic Journal 117 (519), 114–141. Anderson, K., Norman, D., 2001. Global Wine Production, Consumption and Trade, 1961 to 1999: A Statistical Compendium. Centre for International Economic Studies, Adelaide. Anselin, L., Varga, A., Acs, Z., 2000. Geographical spillovers and university research: a spatial econometric perspective. Growth and Change 31 (4),501–515. Archibugi, D., 2007. Introduction to the special issue on knowledge and innovation in the globalising world wine industry. International Journal of Technology and Globalisation 3 (2–3), 125, 26. Arocena, R., Sutz, J., 2005. Latin American Universities: from an original revolution to an uncertain transition. Higher Education 50 (4), 573–592.

Arundel, A., Geuna, A., 2004. Proximity and the use of public science by innovative European Firms. Economics of Innovation & New Technology 13 (6), 559–580. Aylward, D.K., 2003. A documentary of innovation support among new world wine industries. Journal of Wine Research 14 (1), 31–43. Bell, M., Giuliani, E., 2007. Catching up in the global wine industry: innovation systems, cluster knowledge networks and firm-level capabilities in Italy and Chile. International Journal of Technology and Globalisation 3 (2–3),197, 223. Benavente, J.M., Crespi, G., 2003. The Impact of an Associative Strategy (the Profo Program) on Small and Medium Enterprises in Chile. SPRU Electronic Working Paper Series, SPRU. Blumenthal, D., Campbell, E.G., Anderson, M.S., Causino, N., Louis, K.S., 1997. Withholding research results in academic life science—evidence from a national survey of faculty. JAMA—Journal of the American Medical Association 277 (15), 1224–1228.

920

E. Giuliani, V. Arza / Research Policy 38 (2009) 906–921

Bonaccorsi, A., Piccaluga, A., 1994. A theoretical framework for the evaluation of university–industry relationships. R&D Management 24 (3), 229–247. Borgatti, S.P., Everett, M.G., Freeman, L.C., 2002. Ucinet 6 for Windows. Analytic Technologies, Harvard. Bruno, G.S.F., Orsenigo, L., 2003. Variables influencing industrial funding of academic research in Italy: an empirical analysis. International Journal of Technology Management 26 (2–4), 277–302. Brusoni, S., Prencipe, A., Pavitt, K., 2001. Knowledge specialization, organization coupling, and the boundries of the firm: why do firms know more than they make? Administrative Science Quarterly 46 (4), 597–625. Cassiman, B., Veugelers, R., 2002. R&D cooperation and spillovers: some empirical evidence from Belgium. The American Economic Review 92 (4). Charles, P., 2003. Universities and territorial development: reshaping the regional role of UK Universities. Local Economy 18 (1), 7–20. Cohen, W.M., Levinthal, D.A., 1990. Absorptive-capacity—a new perspective on learning and innovation. Administrative Science Quarterly 35 (1), 128–152. Cohen, W.M., Nelson, R.R., Walsh, J.P., 2002. Links and impacts: the influence of public research on industrial R&D. Management Science 48 (1), 1–23. Cooke, P., 2001. Regional innovation systems, clusters and the knowledge economy. Industrial and Corporate Change 10 (4), 945–974. Crespo, M., Dridi, H., 2007. Intensification of university–industry relationships and its impact on academic research. Higher Education 54 (1), 61–84. D’Este, P., Fontana, R., 2007. What drives the emergence of entrepreneurial academics? A study on collaborative research partnerships in the UK. Research Evaluation 16 (4), 257–270. D’Este, P., Patel, P., 2007. University–industry linkages in the UK: What are the factors underlying the variety of interactions with industry? Research Policy 36 (9), 1295–1313. Dantas, E., 2006. The development of knowledge networks in latecomer innovation systems: the case of Petrobras in the Brazilian offshore oil industry, Brighton. DPhil Thesis. SPRU, University of Sussex. Dasgupta, P., David, P., 1994. Toward a new economics of science. Research Policy 23 (5), 487–521. Davenport, T.H., Prusak, L., 2000. Working knowledge: how organizations manage what they know. Ubiquity 1 (24). De Sousa Santos, B., 2006. La Universidad Popular Del Siglo Xxi. Universidad Nacional Mayor de San Marcos. Fondo Editorial de la Facultad de Ciencias Sociales, Lima. Dosi, G., 1988. Sources, procedures and microeconomic effects of innovation. Journal of Economic Literature 26 (3), 1120–1171. Dussauge, P., Garrette, B., Mitchell, W., 2000. Learning from competing partners: outcomes and durations of scale and link alliances in Europe, North America and Asia. Strategic Management Journal 21 (2), 99–126. Echevarria, R.G., Trigo, E.J., Byerlee, D., 1996. Institutional change and effective financing of agricultural research in Latin America. World Bank Papers. Washington, DC. Etzkowitz, H., Leydesdorff, L., 2000. The dynamics of innovation: from national systems and “Mode 2” to a triple helix of university–industry–government relations. Research Policy 29 (2), 109–123. Fischer, M.M., Varga, A., 2002. Technological innovation and interfirm cooperation: an exploratory analysis using survey data from manufacturing firms in the metropolitan region of Vienna. International Journal of Technology Management 24 (7–8), 724–742. Fontana, R., Geuna, A., Matt, M., 2003. Firm size and openness: the driving forces of university–industry collaboration. SPRU Electronic Working Papers. p. 103. Fontana, R., Geuna, A., Matt, M., 2006. Factors affecting university–industry R&D projects: the importance of searching, screening and signalling. Research Policy 35 (2), 309–323. Freeman, L.C., 1979. Centrality in social networks: conceptual clarification. Social Networks 1, 215–239. Fritsch, M., Schwirten, C., 1999. Enterprise-university cooperation and the role of public research institutions in regional innovation systems. Industry and Innovation 6 (1), 69–83. Geuna, A., 2001. The changing rationale for European university research funding: are there negative unintended consequences? Journal of Economic Issues 35 (3), 607–632. Giuliani, E., 2007. The selective nature of knowledge networks in clusters: evidence from the wine industry. Journal of Economic Geography 7, 139–168. Giuliani, E., Bell, M., 2005. The micro-determinants of meso-level learning and innovation: evidence from a chilean wine cluster. Research Policy 34 (1), 47–68. Giuliani, E., Morrison, A., Pietrobelli, C., Rabellotti, R., 2008. Why do researchers collaborate with industry? An analysis of the wine sector in Chile, South Africa and Italy Cespri Working Paper 217. CESPRI Working Paper 217. Bocconi University, Milan. Downloadable from: http://www.cespri.unibocconi.it/ folder.php?vedi=4242&tbn=albero&id folder=1917. Gregorio, D.D., Shane, S., 2003. Why do some universities generate more start-ups than others? Research Policy 32 (2), 209–227. Hall, B.H., Link, A.N., Scott, J.T., 2003. Universities as research partners. The Review of Economics and Statistics 85 (2), 485–491. Jaffe, A.B., 1989. Real effects of academic research. American Economic Review 79 (5), 957–970. Jelinek, M., Markham, S., 2007. Industry-university IP relations: integrating perspectives and policy solutions. IEEE Transactions on Engineering Management 54 (2), 257–267.

Kaufmann, A., Todtling, F., 2001. Science-industry interaction in the process of innovation: the importance of boundary-crossing between systems. Research Policy 30 (5), 791–804. Kitagawa, F., 2004. Universities and regional advantage: higher education and innovation policies in English regions. European Planning Studies 12 (6), 835–852. Kruss, G., 2006. Working partnerships: the challenge of creating mutual benefit for academics and industry. Perspectives in Education 24 (3), 13. Laursen, K., Salter, A., 2004. Searching high and low: what types of firms use universities as a source of innovation? Research Policy 33 (8), 1201–1215. Lee, Y.S., 1996. ‘Technology Transfer’ and the research university: a search for the boundaries of university–industry collaboration. Research Policy 25 (6), 843–863. Link, A.N., Scott, J.T., 2003. Us Science Parks: The diffusion of an innovation and its effects on the academic missions of universities. International Journal of Industrial Organization 21 (9), 1323–1325. Lissoni, F., 2001. Knowledge codification and the geography of innovation: the case of Brescia Mechanical Cluster. Research Policy 30, 1479–1500. Llomovatte, S., Juarros, F., Naidorf, J., Guelman, A., 2006. La Vinculación Universidad˜ y Empresa: Miradas Críticas Desde La Universidad Pública, Buenos Aires: Mino Dávila editores. Loubere, L.A., 1990. The Wine Revolution in France. Princeton University Press, Princeton, New Jersey. Louis, K.S., Jones, L.M., Anderson, M.S., Blumenthal, D., Campbell, E.G., 2001. Entrepreneurship, secrecy, and productivity: a comparison of clinical and nonclinical life sciences faculty. Journal of Technology Transfer 26 (3), 233. Lundvall, B.-Å., 1992. National Systems of Innovation Towards a Theory of Innovation and Interactive Learning. Pinter Publishers. Distributed exclusively in the USA and Canada by St. Martin’s Press, London, New York. Mansfield, E., Lee, J.Y., 1996. The Modern University: contributor to industrial innovation and recipient of industrial R&D support. Research Policy 25 (7), 1047–1058. Margottini, L., 2008. Italy: a plea for ‘Transparent’ funding. Science 320 (5878), 861–961. McDermott, G., Corredoira, R., Kruse, G., 2007. Public-private networks as sources of knowledge and upgrading capabilities: a parametric stroll through argentine vineyards. Academy of Management Best Papers Proceedings. Meyer-Krahmer, F., Schmoch, U., 1998. Science-based technologies: university–industry interactions in four fields. Research Policy 27 (8), 835–851. Miotti, L., Sachwald, F., 2003. Co-operative R&D: why and with whom? An integrated framework of analysis. Research Policy 32 (8), 1481–1499. Morrison, A., Rabellotti, R., 2007. The role of research in wine: the emergence of a regional research area in an italian wine production system. International Journal of Technology and Globalisation 3 (2–3), 155–178. Mowery, D.C., Nelson, R., Sampat, B., Ziedonis, A., 2001. The growth of patenting and licensing by U.S. Universities: an assessment of the effects of the Bayh-Dole Act of 1980. Research Policy 30(1), 99–119. Mowery, D.C., Sampat, B.N., 2004. Universities in national innovation systems. In: Fagerberg, J., Mowery, D.C., Nelson, R.R. (Eds.), The Oxford Handbook of Innovation. Oxford University Press, Oxford. Nelson, R.R. (Ed.), 1993. National Innovation Systems. Oxford University Press, New York. Nelson, R.R., 2004. The market economy, and the scientific commons. Research Policy 33 (3), 455–471. Nelson, R.R., Winter, S.G., 1982. An Evolutionary Theory of Economic Change. Belknap Press of Harvard University Press, Cambridge, Mass. Paul, H.W., 1996. Science. In: Vine and Wine in Modern France. Cambridge University Press, Cambridge. Pickering, J.F., Matthews, D.N., Wilson, C., Kirkland, J., 1999. The university:industry interface in the generation of intellectual property. Higher Education Quarterly 53 (1), 6–28. Poyago-Theotoky, J., Beath, J., Siegel, D.S., 2002. Universities and fundamental research: reflections on the growth of university–industry partnerships. Oxford Review of Economic Policy 18 (1), 10–21. Rogers, E.M., 1983. Diffusion of Innovations. The Free Press, New York. Salter, A.J., Martin, B.R., 2001. The economic benefits of publicly funded basic research: a critical review. Research Policy 30 (3), 509–532. Santoro, M.D., Chakrabarti, A.K., 1999. Building industry-university research centers: some strategic considerations. International Journal of Management Reviews 1 (3), 225–244. Schartinger, D., Rammer, C., Fischer, M.M., Frohlich, J., 2002. Knowledge interactions between universities and industry in Austria: sectoral patterns and determinants. Research Policy 31 (3), 303–328. Slaughter, S., Campbell, T., Holleman, M., Morgan, E., 2002. The “Traffic” in graduate students: graduate students as tokens of exchange between academe and industry. Science, Technology, & Human Values 27 (2), 282–313. Slaughter, S., Leslie, L., 1997. Academic Capitalism: Politics, Policies and the Entrepreneurial University. Johns Hopkins University Press. Smith, K., 2007. Technological and economic dynamics of the world wine industry: an introduction. International Journal of Technology and Globalisation 3 (2–3), 127–137. Steiner, M., Ploder, M., 2008. Structure and strategy within heterogeneity: multiple dimensions of regional networking. Regional Studies 46 (2), 793–815. Tapper, E.R., Salter, B.G., 1995. The changing idea of university autonomy. Studies in Higher Education 20 (1).

E. Giuliani, V. Arza / Research Policy 38 (2009) 906–921 Tornquist, K.M., Kallsen, L.A., 1994. Out of the Ivory tower: characteristics of institutions meeting the research needs of industry. The Journal of Higher Education 65 (5), 523–539. Van Looy, B., Debackere, K., Andries, P., 2003. Policies to stimulate regional innovation capabilities via university–industry collaboration: an analysis and an assessment. R&D Management 33 (2), 209–229. Vega-Jurado, J., Fernández de Lucio, I. and Huanca-López, R., 2007. La Relación Universidad-Empresa En América Latina:¿ Apropiación Incorrecta De Modelos Foráneos? Journal of Technology Management & Innovation 2 (2).

921

Velho, L., Saez, W.T., 2002. R&D in the public and private sector in brazil: complements or substitute? UNU/INTECH Discussion Papers, 1564–8370. Veugelers, R., Cassiman, B., 2005. R&D cooperation between firms and universities. some empirical evidence from Belgian manufacturing. International Journal of Industrial Organization 23 (5), 355–379. von Hippel, E., 1987. Cooperation between rivals: informal know-how trading. Research Policy 16, 291–302. Wasserman, S., Faust, K., 1994. Social Network Analysis. Methods and Applications. Cambridge University Press, Cambridge.