Portal de Periódicos da CAPES

Artigo Acesso aberto Produção Nacional Revisado por pares

BIBTEX RIS

Spatial and non‐spatial proximity in university–industry collaboration: Mutual reinforcement and decreasing effects

2020; Elsevier BV; Volume: 13; Issue: 4 Linguagem: Inglês

10.1111/rsp3.12312

1757-7802

Emerson Gomes dos Santos, Renato García, Veneziano Araújo, Suelene Mascarini, Ariana Ribeiro Costa,

Entrepreneurship Studies and Influences

The role of geographical proximity in fostering innovation is widely recognized, and local flows of information and knowledge-sharing play a critical role in interactive learning. The aim of this paper is to evaluate synergistic effects and non-linearities in spatial and non-spatial proximity in university–industry collaborations. Previous studies have evaluated this issue in scientific collaboration. Here, we examine the role of geographical and cognitive proximity in university-industry collaboration using data from collaborative projects between academic research groups and firms in Brazil (4,342 collaborations involving 3,063 firms and 1,738 research groups in most scientific fields). Our results reveal synergistic effects that mutually reinforce the benefits of geographical and cognitive proximity and reveal non-linearity in the impacts of geographical and cognitive proximity in university-industry collaborations. The contribution of our research is its examination of synergistic effects and non-linearities in university-industry collaborations that include information exchange and knowledge-sharing between firms and their academic partners. El papel de la proximidad geográfica en el fomento de la innovación está ampliamente reconocido, y los flujos locales de información y diseminación de conocimientos juegan un papel fundamental en el aprendizaje interactivo. El objetivo de este artículo es evaluar los efectos sinérgicos y las no linealidades en la proximidad espacial y no espacial en las colaboraciones universidad-industria. Estudios anteriores han evaluado esta cuestión como una colaboración científica. Aquí, se examina el papel de las proximidades geográficas y cognitivas en la colaboración universidad-industria, a partir de datos de proyectos de colaboración entre grupos de investigación académica y empresas de Brasil (4.342 colaboraciones en las que participaron 3.063 empresas y 1.738 grupos de investigación en la mayoría de los campos científicos). Nuestros resultados revelan efectos sinérgicos que refuerzan mutuamente los beneficios de la proximidad geográfica y cognitiva y revelan una no linealidad en los impactos de la proximidad geográfica y cognitiva en las colaboraciones universidad-industria. La contribución de esta investigación es su examen de los efectos sinérgicos y las no linealidades en las colaboraciones universidad-industria que incluyen el intercambio de información y conocimientos entre las empresas y sus socios académicos. イノベーションの醸成における地理的近接性の役割は広く認識されているが、インタラクティブ・ラーニングにおいては情報と知識共有の地域的フローが重要な役割を担っている。本稿の目的は、産学共同研究における空間的および非空間的な近接性の相乗効果と非線形性を評価することである。既存研究では、科学的共同研究におけるこの問題を評価している。今回は、ブラジルにおける大学研究グループと企業の共同プロジェクト (大部分の科学分野における3,063の企業と1,738の研究グループを含む4,342のコラボレーション)のデータを用いて、産学共同における地理的および認知的な近接性の役割を検討した。結果から、地理的および認知的な近接性の利点が相互に強化される相乗効果があること、そして産学協力における地理的および認知的な近接性の影響が非線形性であることを明らかにする。本稿は、企業とその研究パートナーの情報交換と知識共有を含む産学協力における相乗効果と非線形性を検討し、研究に寄与するものである。 The early literature on this topic considered geographical proximity a major factor that facilitates economic interaction (Gertler, 2007; Storper & Venables, 2004). Research has studied geographical proximity in university-industry collaboration and found that it can affect the likelihood of linking academic research to industrial R&D (D'Este & Iammarino, 2010; Garcia, Araujo, Mascarini, Santos, & Costa, 2015; Laursen, Reichstein, & Salter, 2011). By incorporating geographical and non-spatial dimensions of proximity, recent studies have sought to evaluate how the dimensions of proximity among agents can foster collaboration between universities and firms. Both spatial and non-spatial proximity play important roles in partner selection and the establishment of collaborative networks in scientific collaborations (Cao, Derudder, & Peng, 2019; Capello & Caragliu, 2018; Fernández, Ferrándiz, & León, 2016; He, Wu, & Zhang, 2019; Werker, Korzinov, & Cunningham, 2019), in the creation of regional patent partnerships (Marrocu, Paci, & Usai, 2013), and in other types of regional partnership projects (Marek, Titze, Fuhrmeister, & Blum, 2017). Previous studies identified the main benefits of proximity for collaboration (Arundel & Geuna, 2004; D'Este & Iammarino, 2010; Laursen et al., 2011) and emphasized the role of local knowledge flows in fostering interactive learning among agents, with positive effects on innovation (Brekke, 2020; Fitjar & Gjelsvik, 2018). However, these studies used formal channels of university-industry interaction, such as patents, technology licensing and spin-offs, to measure the effects of collaboration between academic research and industrial R&D. In this study, we examine the synergistic effects and non-linearity in spatial and non-spatial proximity in university-industry collaborations. We also investigate the relation between geographical and cognitive proximity. Data on university-business collaboration offer several advantages over formal channel collaboration data because they involve a wider range of forms of collaboration and because such joint projects represent more widespread ways of technology transfer (Agrawal & Henderson, 2002; Banal-Estañol, Jofre-Bonet, & Lawson, 2015). We examine this topic with respect to the Brazilian context. Similar to many developed and developing countries, universities in Brazil play an increasingly important role in fostering innovation, contributing significantly to the participation of Brazilian firms in international markets (Albuquerque, Suzigan, Kruss, & Lee, 2015; Suzigan, Albuquerque, Garcia, & Rapini, 2009). However, it is important to consider countries with less-developed institutional and economic environments, in which university-industry collaboration differs in its nature and its main determinants from that in developed countries (Moraes-Silva, Furtado, & Vonortas, 2018). Despite the differences in countries' institutional and economic characteristics, we believe that our results enhance the general understanding of the importance of spatial and non-spatial proximity in university-industry collaboration. This paper makes three main contributions. First, we address questions related to proximity in the context of university-industry collaboration. There is a lack of literature on the main characteristics of university-industry collaboration, particularly the factors that stimulate collaboration between academic researchers and industrial R&D staff (Perkmann et al., 2013). Second, our analysis relies on data on linkages such as collaborative projects, whereas previous studies mainly used formal channels of knowledge transfer, such as patents, licenses and spin-offs. Data on collaboration offer several advantages over data on formal channels of interaction because collaborative projects with industry are both more widespread and more important for technology transfer (Agrawal & Henderson, 2002; Banal-Estañol et al., 2015; D'Este & Patel, 2007). Third, we present new empirical evidence on the role of spatial and non-spatial proximity for universities and industry in Brazil. This empirical evidence fills an important gap in the literature by increasing our understanding of the characteristics of and factors that affect university-industry collaboration in developing countries (Barletta, Yoguel, Pereira, & Rodríguez, 2017; Mascarenhas, Ferreira, & Marques, 2018; Suzigan et al., 2009). The paper has four sections. Section 2 presents the conceptual background regarding the role of spatial and non-spatial proximity in fostering university-industry engagement. Section 3 provides a brief description of the data set and the main methodological questions. Section 4 presents the results and discusses the primary evidence regarding the synergies and non-linearities of proximity for university-industry collaboration. Section 5 presents concluding remarks and discusses the limitations of our study and policy implications. Proximity plays an important role in knowledge-sharing and interactive learning. Previous studies have revealed the substantial benefits of geographical proximity for knowledge-sharing, with impacts on regional innovation. Knowledge spillovers from clustering are geographically bounded, and benefits are larger among nearby firms. In addition, geographical proximity facilitates knowledge-sharing and interactive learning by generating positive externalities and by creating mechanisms that stimulate interaction, such as face-to-face contact and frequent interaction (Marrocu et al., 2013; Storper & Venables, 2004). To increase the understanding of how knowledge is exchanged, it have developed a comprehensive theoretical framework that distinguishes several different but interdependent non-spatial proximity dimensions, such as social, cognitive, institutional and organizational dimensions (Boschma, 2005; Knoben & Oerlemans, 2006). The importance of non-spatial forms of proximity has emerged in response to the argument that geographical space is the only setting in which knowledge disseminates (Breschi & Lissoni, 2001; Capello, 2009). Particular attention has been paid to the cognitive dimension of proximity as an explanation of knowledge diffusion. Cognitive proximity is typically defined as the degree of knowledge shared between organizations (Garcia, Araujo, Mascarini, Santos, & Costa, 2018; Marek et al., 2017). Such proximity facilitates knowledge-sharing and strengthens the absorptive capacity of firms by enabling agents to identify, acquire, understand and exploit knowledge available from others (Boschma, 2005; Knoben & Oerlemans, 2006). Cognitive proximity is related to the knowledge base of others (Cunningham & Werker, 2012; Nooteboom, van Haverbeke, Duysters, Gilsing, & van den Oord, 2007). Different cognitive bases and different absorptive capacities are necessary to identify, interpret and explore new knowledge (Cohen & Levinthal, 1990). That is, there are similarities in the way the world is perceived, interpreted, understood, and evaluated by different agents. With shared technological experience, cognitive proximity also encompasses technological proximity because the cognitive dimension involves the intermediate tools, devices and knowledge required to create products and services as well as the agents' knowledge base. Similarities in the knowledge base of different agents facilitate collaboration and network formation (Cao et al., 2019). The formation of regional networks of scientific collaboration has been assessed through different measures, such as co-patenting and formal collaborations among agents. Collaboration can also be affected by different types of non-spatial proximity. The role of different dimensions of proximity in the formation of such networks is an increasingly important topic in economic geography (Cao et al., 2019; Capello & Caragliu, 2018; He et al., 2019; Werker et al., 2019). The dimensions of proximity can affect the likelihood of collaboration, which can foster knowledge-sharing and support learning processes (Autant-Bernard & LeSage, 2011; Ponds, van Oort, & Frenken, 2007). Recent studies have sought to evaluate proximity with respect to scientific collaboration. Empirical studies show that as geographical proximity decreases, cognitive and technological proximity become more relevant. This relationship indicates the existence of complementarity and synergistic effects among the different dimensions of proximity (Cao et al., 2019; Capello & Caragliu, 2018; Werker et al., 2019). Previous studies have noted that geographical proximity and cognitive proximity are interchangeable (Garcia et al., 2018). On the one hand, cognitive proximity between agents can overcome barriers to collaboration related to long geographical distances. Long-distance collaborations can occur within a non-local community of professionals, in which agents can exchange information and share complementary knowledge. On the other hand, in specialized clusters, the combined presence of geographical and cognitive proximity can strengthen information exchange and knowledge-sharing. In this context, these proximities are complementary (Broekel & Boschma, 2012; Lazzeretti & Capone, 2016). However, the varying empirical evidence indicates the need to increase our understanding of potential synergistic effects and complementarity between geographical and cognitive proximity. Some research, most of which has addressed scientific collaborations, has noted the existence and the relevance of synergic effects between different dimensions of proximity (Capello & Caragliu, 2018). Geographical proximity seems to compensate for a lack of cognitive proximity between collaborating scientific researchers. However, when geographical distance increases, so do the requirements of cognitive proximity between agents (Cao et al., 2019; Capello & Caragliu, 2018; Werker et al., 2019). This assumption leads us to our first hypothesis. Hypothesis 1.There are synergistic effects in the relation between geographical and cognitive distance. In addition, other studies have observed non-linearity in the main impacts of cognitive distance with respect to scientific collaboration. That is, an increase in geographical and cognitive proximity implies that scientific collaboration increases but at decreasing rates (Capello & Caragliu, 2018). This result leads us to our second hypothesis. Hypothesis 2.There is no linearity in the relation between geographical and cognitive distance. Therefore, it is important to test both for synergies and for non-linear effects in the context of university-industry engagement, particularly using data from joint collaborative projects between academic research groups and firms. Data were collected from the Brazilian Ministry of Science and Technology using the Brazilian National Council for Scientific and Technological Development (CNPq) Directory of Research Groups of the Lattes platform. The Research Group Directory (DGP) is the main source of data on university-industry linkages in Brazil (Garcia, Araujo, & Mascarini, 2013; Rapini et al., 2009; Suzigan et al., 2009). The database encompasses a large set of academic research groups in Brazil and covers their main features, such as scientific field, number of researchers, research performance and specific information on collaborations. We use data from 2010. To analyse the regional university-industry collaboration network in Brazil, data were aggregated by mesoregion (equivalent to European NUTS 2). In addition, to better represent and explain these networks, we had to consider the continental dimensions of the country and the substantial regional concentration of the collaborating firms (Garcia et al., 2015). Therefore, we selected 36 mesoregions with more than 1.5 million inhabitants in 2010. This cut-off represents 63.2% of the Brazilian population distributed in 42% of the municipalities, where 4,342 collaborations were considered (61.4%). The dependent variable was the number of collaborations between each pair of regions, considering the sum of the collaborations that occurred between the firms in each of the 36 regions with academic research groups in each region. Therefore, there were 1,296 possibilities for regional collaboration (units of analysis) from the 36 regions. A total of 391 (30.2%) of the regional networks had at least one collaboration between firms and academic research groups, most of them in the same region (2,710 collaborations, or 62.4%). Regarding regional differences, regional structural differences that may affect interest in collaboration are considered. The more regional collaborations that occur, the smaller the gap between the regional productive structure (GDP per capita) and the capacity to apply the scientific knowledge (patents by employees) of the regions (Capello & Caragliu, 2018). The regional productive structure can also influence collaborations because urban and diversified regions have more appropriate conditions for knowledge exchange and innovation (Duranton & Puga, 2000; Storper & Venables, 2004). Table 1 presents the variables used to evaluate the hypotheses regarding synergistic effects and non-linearity. Table 2 presents the descriptive statistics of the main variables. Our dependent variable is numerical data. In examining such data, it is common to use the Poisson distribution, which assumes that the mean is equal to the variance. However, this approach was not valid in our study, in which the variance and the mean of the regional collaborations were 432.14 and 3.35, respectively (Table 2). In addition, because in nearly 70% of the region pairs there are no collaborations between firms and universities, there are a large number of zeros. Zero inflated negative binomial distribution (ZINB) applies in cases in which there is joint overdispersion and an excessive number of zeros (Cameron & Trivedi, 2013; Lawless, 1987). The negative binomial approach is frequently used in economic geography research (Capello & Caragliu, 2018; Marek et al., 2017). It involves a two-stage process: a negative binomial regression, which estimates the number of collaborations between a pair of regions, and a logistic regression, which estimates the likelihood of no collaboration. The difference between this approach and the ZINB model is that it considers two sources of zeros, including true zero, which is part of the distribution (Fernández et al., 2016). Tests are used to determine whether to employ a negative binomial model or a ZINB model (Vuong, 1989). In the z-Vuong test, the null hypothesis is that both models can be used without a large difference in the results. The ZINB model is the best model if z-Vuong > 0; otherwise, the zero-free inflated model is the best. Empirical verification of the drivers of the number of regional collaborations was performed for five models. To address possible data heteroscedasticity, all of the models were estimated considering robust standard error. Model (1) considers only regional characteristics and differences. Model (2) also includes geographical and cognitive distances. Models (3), (4) and (5) are used to evaluate Hypothesis 1 regarding potential synergistic effects between geographical and cognitive proximities and Hypothesis 2 regarding non-linearity in the relationship between the two types of distance. The results are presented in Table 3, which shows that the tests for overdispersion are significant for all estimates, ensuring the most appropriate use of the negative binomial distribution. That is, we can confirm the validity of using the negative binomial model inflated with zeros. In addition, the Vuong test indicates the need to use a zero-inflated model instead of only the negative binomial. Model (1) includes only the control variables. Its results confirm the importance of the selected characteristics in the other models because all variables present significance and the expected sign. The selected regional characteristics are relevant. Regions with higher absorptive capacity (abscap), a larger number of firms (n_firm) and a higher number of publications (sciart) have more regional collaboration between university and industry. Regarding the differences between regions, regions with distinct economic (gdp_diff) and technological (pat_diff) profiles present fewer collaborations. Model (2) includes geographical and cognitive distances between the collaborating regions. The estimated coefficients are negative and significant, indicating that increases in both geographical and cognitive distance reduce university-industry collaboration between regions. This result is consistent with that of previous studies (Capello & Caragliu, 2018; Garcia et al., 2018). Regarding the controls, the coefficients are significant, and the results are similar to those of the initial model. Model (3) evaluates the potential synergistic effects between geographical and cognitive distance. To this end, we include in the basic model the interaction term between geographical and cognitive distance (geo*cogn). The results of the original model are maintained with the exception that the coefficient of technological difference between regions is not significant (pat_diff). Regarding potential synergistic effects between geographical and cognitive distance, the results indicate that the interaction term between distances (geo*cogn) is positive and significant, and the individual linear terms (geo and cogn) have negative and significant coefficients, similar to the outcomes of Models (1) and (2). The positive and significant interaction term indicates that the effects of cognitive distance decrease as geographical distance increases. Our results reveal synergistic behaviour between geographical and cognitive distance, which indicates that a large distance in one type of proximity reduces the negative impact of the other type of proximity. Cognitive proximity between collaborative firms and their academic partners can compensate for the challenges of interaction over large geographical distances. Additionally, partnerships between firms and universities located in the same or spatially close regions can take advantage of geographical proximity to establish partnerships between agents that do not share the same common knowledge base. In these cases, overall mechanisms related to geographical proximity, such as frequent contact and face-to-face interaction, can strengthen collaboration between cognitively distant agents. Regarding Hypothesis 2, Model (4) evaluates the potential non-linearity of the effects of both geographical and cognitive distance on university-industry collaboration between regions. The results indicate that the two dimensions of proximity present similar overall behaviour, with negative and significant linear terms (geo and cogn) and positive and significant quadratic terms (geo2 and cogn2). Collaboration increases when partners are geographically and cognitively close. In addition, non-linearity occurs between both geographical and cognitive distance and collaboration, which supports Hypothesis 2. The effect on the number of collaborations of the two terms (linear and quadratic) of geographic and cognitive distance presents the same overall result. This finding not only reveals the negative effect of distance and positive values for the quadratic term but also strongly demonstrates the negative effect of decreases in distance. These results confirm our theoretical expectation, suggesting the existence of a gain limit with both geographical and cognitive proximity. That is, there is an optimal level of distance for university-industry collaboration. In addition, otherresults remain significant for the terms used in Model (2) except for economic difference (gdp_diff), which loses significance. Finally, Model (5) is estimated to evaluate synergy against non-linearity. The results are similar to those reported above, confirming the existence of synergistic effects between geographical and cognitive proximity and non-linearity between university-industry collaboration and both geographical and cognitive proximity. Both the spatial and non-spatial dimensions of proximity have clear effects on university-industry engagement in regions. Previous studies found similar results, but most of these studies addressed scientific collaboration (Cao et al., 2019; Capello & Caragliu, 2018; Fernández et al., 2016; Werker et al., 2019). Regarding collaboration between universities and industry, both dimensions play an important role in the establishment of collaborative projects between firms and their academic partners. Geographical proximity strengthens interaction through typical mechanisms of economic geography, such as face-to-face contact and frequent interaction. Cognitive proximity facilitates information exchange and knowledge-sharing between industrial scientists and engineers and their academic partners because they share similar knowledge bases. The synergistic effects between geographical and cognitive proximity on university-industry engagement strengthen this result. Partners that share a similar knowledge base tend to deepen their collaboration by increasing the breadth and scope of their collaborative projects. The strengthening of the linkages between firms and their academic partners reinforces collaboration between the participants, with positive effects on knowledge exchange and interactive learning. The synergistic effects can mutually reinforce the effects of spatial and non-spatial dimensions of proximity on collaboration. Firms tend to collaborate with academic partners who are located in nearby regions and who possess a similar knowledge base. The same result was found in analyses of scientific collaborations (Cao et al., 2019; Capello & Caragliu, 2018) and of university-industry engagement (Fitjar & Gjelsvik, 2018; Garcia et al., 2018; Laursen et al., 2011). Collaboration between firms and universities in specialized clusters illustrates this effect. In specialized clusters, benefits related to geographical proximity between firms and their academic partners can be reinforced by the existence of similar knowledge bases between collaboration participants. Technological overlap between two co-located partners increases the likelihood of their engagement in collaborative projects. This outcome indicates that local firms tend to link more with technologically similar partners, and geographical proximity tends to be a driver of network formation (Broekel & Boschma, 2012; Gordon & Kourtit, 2020; Kaygalak & Reid, 2016). However, cognitive proximity can also overcome the costs and barriers of long-distance collaboration because a similar knowledge base between partners can stimulate knowledge-sharing and interactive learning. In fact, long-distance collaborations may occur within knowledge communities, where meetings, conventions, and conferences become powerful tools that help create temporary spatial proximity for agents with similar knowledge bases (Cao et al., 2019; Maskell, Bathelt, & Malmberg, 2006). Our findings also reveal the existence of non-linearity between university-industry engagement and the proximity dimensions. The squares of both proximity dimensions reflect a non-linear effect of spatial and non-spatial proximity on university-industry collaboration. This finding indicates that collaboration decreases with an increase in geographic and cognitive distance but at decreasing rates, which means that there are decreasing effects of both spatial and non-spatial dimensions of proximity. The non-linear relation between geographical proximity and university-industry collaboration suggests an inverted U-shaped relation for geographical proximity and collaboration. Collaboration decreases with an increase in geographical distance, indicating the strong role of localized knowledge spillovers. Local collaboration between universities and industry can benefit from frequent interaction and face-to-face contact, which is helpful in transferring tacit knowledge (Fitjar & Gjelsvik, 2018; Garcia et al., 2015). The costs of long-distance collaboration increase with geographical distance. However, after a certain level of geographical distance, the effect of proximity on collaboration becomes less important, which explains the decreasing effects of the increase in geographical distance on collaboration. Regarding the decreasing effects of cognitive proximity on collaboration, our results confirm our theoretical expectation that cognitive proximity plays a crucial role in fostering collaboration between partners. When partners are cognitively close, they share similar knowledge bases and take advantage of the proximity to establish collaborative projects (Nooteboom et al., 2007). However, this advantage decreases when interregional cognitive proximity increases, indicating increasing returns to cognitive proximity at decreasing rates. This finding is in line with empirical results for scientific collaboration (Cao et al., 2019; Capello & Caragliu, 2018) and with theoretical expectations (Boschma, 2005; Nooteboom et al., 2007) of cognitive lock-in generated by the excessive cognitive proximity among local actors. In specialized clusters, long-term development requires a certain degree of diversification among agents and local organizations. Firms and their academic partners must be sufficiently different to benefit from one another's knowledge and sufficiently similar to comprehend one another (Cunningham & Werker, 2012). Finally, drivers of university-industry engagement found in several developed countries also affect collaboration in developing countries (Barletta et al., 2017; Kaygalak & Reid, 2016; Suzigan et al., 2009). More collaboration is particularly fruitful for innovation. However, it can also be useful to university scholars in producing academic outputs and technology transfer in both developed and developing countries, even with strong differences in institutional contexts (Garcia et al., 2018; Moraes-Silva et al., 2018). The findings of synergistic effects and non-linearity between spatial and non-spatial dimensions of proximity are not new. Empirical studies found synergistic effects between geographical and cognitive proximities, but they examined scientific collaboration (Cao et al., 2019; Capello & Caragliu, 2018; Werker et al., 2019). Other studies found non-linearity between spatial and non-spatial dimensions of proximity, also for scientific collaboration (Capello & Caragliu, 2018). The novelty of this paper is its investigation of these issues with respect to university-industry collaboration, which includes a substantial exchange of information and thorough knowledge-sharing. Our empirical findings are based on data on university-industry collaborative projects rather than on data on formal channels of knowledge transfer, such as patents, licences, or spin-offs. university-industry collaboration is also characterized by different concerns, motivations, and barriers between partners. Industrial scientists want to add to their corporate profit stream the monies derived from the rights to use private knowledge, whereas academic researchers are typically concerned with increasing the stock of public knowledge (Perkmann et al., 2013). We found synergistic effects of geographical and cognitive proximity. When firms and their academic partners are physically close and share a similar knowledge base, the benefits of collaboration are intensified. Additionally, non-linearity between the dimensions of proximity and collaboration means that the benefits of collaboration increase when geographical proximity increases but at decreasing rates. We also contribute to the literature by investigating a developing country. Although our empirical study was applied to Brazil, we believe that the main results are sufficiently general to be applied to other contexts for two reasons. First, similar to other developed and developing countries, universities in Brazil have been playing an increasingly important role in fostering innovation (Albuquerque et al., 2015; Suzigan et al., 2009). In the context of new knowledge-intensive technologies, which are often associated with so-called Industry 4.0, firms have been forced to intensify their search for new sources of scientific and technological knowledge. Universities have been increasingly used by firms as privileged sources of access to such new knowledge. Second, similar to developed countries, universities in Brazil are viewed as actors that can play a wider and systemic role in regional development by building bridges among participants in regional innovation systems (Asheim, Isaksen, & Trippl, 2019; Brekke, 2020). Even in less developed regions such as Brazil, universities can be active actors in the exchange of information and knowledge-sharing among local firms, which can foster regional development (Barletta et al., 2017; Garcia et al., 2015; Kaygalak & Reid, 2016). Geographical and cognitive proximity can support the formation of local networks of actors and local knowledge-sharing, which could represent a strong mechanism for enhancing regional economic growth. However, it is also important to recognize that this investigation of university-industry collaboration in Brazil has several limitations regarding the generalization of its results. The requirements for developing countries to update their national innovation systems and the role of universities in supporting these systems differ from those of developed countries (Albuquerque et al., 2015; Suzigan et al., 2009). The presence of less developed institutional and economic environments in developing countries suggests that university-industry engagement may have a different nature and determinants than in developed countries (Moraes-Silva et al., 2018). For example, an important share of university-industry collaboration depends on the diffusion and adaptation of foreign technology (Albuquerque et al., 2015). In addition, because of low private R&D expenditure, firms can use academic research as a substitute for investment, which can encourage academic research to focus on applied research. These differences in the institutional context can have an important influence on university-industry collaboration in Brazil, with effects on the patterns of the spatial distribution of collaboration projects. For example, there is a strong concentration of collaborative projects between firms and academic partners in the southern part of the country, where the main R&D expenditure occurs and where high-performing universities are geographically concentrated (Garcia et al., 2015). Finally, our results have several policy implications. One such implication is that policy-makers should design policies to strengthen university-industry linkages. However, policy-makers should consider the role of geographical proximity as a tool to foster interactive learning between the collaborative firm and its academic partners. Policies should create mechanisms that enable collaborating firms to benefit from the externalities that emerge from geographical agglomeration. Economic geography-related mechanisms, such as frequent contact and face-to-face interaction, must be strengthened by policy initiatives. The synergistic effects between geographical and cognitive proximity with respect to university-industry collaboration are in agreement with this implication. Policy-makers should consider such effects and create programmes that stimulate collaborative projects between firms and universities in which the partners are located in nearby regions and share a similar knowledge base. Policy measures that strengthen local capabilities in specialized clusters are important. Local universities can be an important source of innovation for firms and can represent a link between local manufacturers and external knowledge. However, for partners that are geographically distant from one another, policy should mitigate the barriers and costs of long-distance interactions by searching for partners that share a similar knowledge base and by reinforcing cognitive proximity between the collaborative firm and its academic partner. In this type of collaborative project, partners can benefit from mutual understanding due to their similar knowledge base. A common knowledge base can help partners create mechanisms that enable them to benefit from temporary knowledge proximity, for example, events such as scientific and technological meetings and conferences supported by long-distance communication technologies. However, policy-makers should also consider the importance of dissimilar capabilities, which play an important role in creating new knowledge combinations and in avoiding regional lock-in. Policy recommendations should also consider non-linearity between university-industry collaboration and the dimensions of proximity. Policy-makers should consider geographical proximity because it affects typical factors of economic geography, such as frequent interaction, face-to-face contact, and local knowledge spillovers. However, because there is non-linearity between the benefits of collaboration and geographical distance, policy measures should disregard the typical factors of economic geography in the case of long-distance collaboration. After a certain geographical distance, these benefits lose relevance. Similarly, policy recommendations should consider cognitive proximity. Policy-makers should prioritize collaborative projects between cognitively close universities and firms because cognitive proximity reinforces interaction between physically close agents and can overcome the costs and barriers of long-distance interaction. However, collaborative projects involving agents that do not share similar knowledge bases must also be considered because they can represent a new source of knowledge and help avoid regional lock-in. This work was supported by Fapesp under Grant no. 2012/23.370-5; and CNPq under Grant no. 473.705/2013-3.