On the geography of emerging industry technological networks: the breadth and depth of patented innovations

On the geography of emerging industry technological networks: the breadth and depth of patented... Abstract We study an emerging industry’s technological network in which the patented technologies of the industry are connected to the geographic locations of the inventors. Our research context is the global wind turbine industry. The network maps the geography of the industry’s patented technologies over time. It shows the locations’ patenting activities in different technologies, centered around the core classes of electricity and aerodynamics. These activities shape the locations’ positions in the global patent network and indicate their relative importance to the industry’s innovations. We note that the network and thus the relative positions of the locations within it are constantly changing. We examine how the existing patented knowledge stock at a location affects its position in this technological network. Further, we analyze how locations that enter the global technology network attain centrality. 1. Introduction Regional development is a complex phenomenon involving a variety of actors and linkages (Coe et al., 2008). While industries influence regional development (Storper and Walker, 1989), global production networks and regional relational networks interact (Coe et al., 2004; Yeung and Coe, 2015) feeding into the process of regional growth. Growth at the level of firm, region or nation does not always imply rising innovation activity there (Awate et al., 2012). However, a high level of innovation activity positively correlates with growth. Further, going beyond the absolute levels, a rising trend of innovation activity indicates the growth focus of the relevant entity. In this paper, we study the innovation-led growth of locations. Innovation activity at the level of a location encapsulates the activities of micro entities such as firms, and within them, of people. The aggregation of the innovation activities of these entities defines the innovativeness of a location. The technologies that the local firms and their scientists and engineers work on form the technological signature of that location. As these actors search for knowledge, their search spans technological as well as geographic dimensions, which are non-orthogonal. In fact, a location’s technological linkages with other locations are exploited by firms and their inventors in their knowledge search strategies to generate innovations (Bell and Zaheer, 2007; Funk, 2014; Hannigan et al., 2015; Paruchuri and Awate, 2016). The technological network connections of a location thus facilitate its coupling to the industry’s innovation network (Hannigan et al., 2015) and the global production network, thereby driving its growth (Coe et al., 2004). An industry’s technological network of locations maps the technologies that a certain location directly connects to (i.e., works in) and the strength of those connections. Further, in its entirety, the network models the core and peripheral technologies associated with the industry, the relative position of the locations in this technological network and avenues for their nonlocal technological searches. A central position of a location in such a network is the outcome of a higher amount of innovation activity in a number of technologies relative to other locations in the network, commonly measured by degree centrality (Turkina et al., 2016). However, as the literatures on economic geography, strategy, technology management and sociology note, such a network is far from constant. Be it the ever-changing geography of innovation (Storper and Walker, 1989), the Red Queen effect in firms’ technological competition (Banker et al., 2013), industry lifecycles (Klepper, 1996) or the ecology of technological change (Podolny and Stuart, 1995), all point to the dynamics of technology-based networks. Thus, a central innovative location with a strong technological coupling to the industry’s innovation network is likely to experience its coupling weaken over time if it does not reinforce it through sustained innovation activity. Over time, it may find itself relegated to a peripheral position in relation to other locations. How might a location and the micro entities present there keep up as well as catch up with others in the changing technological network? How might a location leverage its extant innovations to define its position in the network? We empirically explore these questions using patented technological innovations created at a location. A key element of a patent innovation network is the importance given to the individual inventor. Thus, our study addresses the ‘need for approaches that recognize the significance of innovative networks that extend beyond firms and, in particular, those associated with the movement of knowledgeable individuals’ (Coe and Bunnell, 2003, 437). We situate our study in an emerging/growing industry, as the technological dynamics in such industries are stronger than in mature industries. The emerging industry studied in this paper is the global wind power industry. The industry came into existence in the early 1980s after the leading industrial nations of the world suffered the oil crises of the 1970s. Since then, the industry has witnessed the establishment of a dominant design, steady output growth, shakeouts, changing policy regimes and technological discontinuities, all of which are characteristic features of emerging industries (Awate et al., 2012). We use patented innovation data of the wind power industry from the United States Patent and Trademark Office (USPTO) database. Our unit of analysis is a location, which aggregates inventor locations, noted on patents, within a 30-mile radius. Specifically, we construct measures of breadth and depth of patented innovation at each location. Breadth measures the number of active patent technology classes in a location and their importance to the industry. Depth measures the amount of patenting in the set of local technology classes. We find that breadth and depth of patented innovations produce curvilinear effects on a location’s future degree centrality in the industry's technological network. Further, we find that breadth produces a larger effect than depth suggesting that location’s centrality in an emerging-industry technological network is far more sensitive to breadth than depth. We also find that, among the locations entering the emerging industry, the established innovative locations are more likely to increase depth of patented innovation than breadth in the focal industry’s technologies. In contrast, the emergent innovative locations are more likely to increase breadth of patented innovation than depth. We argue that the emergent locations’ breadth focus helps them to rapidly improve their centrality in the technological network of locations. The remaining paper is organized as follows. We review the literature in economic geography and technology management to build our hypotheses in Section 2. We explain the development of the global wind power industry in Section 3, followed by details of the data and methodology in Section 4. We then present the estimations and the results in Section 5 and discuss the findings, implications and limitations in Section 6. 2. Geographic industrialization and innovation In the dominant model of geographic industrialization and innovation, Storper and Walker (1989) proposed that dynamic sectors are capable of generating their own resources thereby granting the leading firms ‘substantial freedom to develop where they are, or to locate where they please … innovators in a fast growing industry can ignore the traditional locational calculus’ (p. 75). Their model predicts four stages of geographic industrialization, namely localization, clustering, dispersal and the rise of new industrial centers. While that is the general pattern, the growth of a location in a particular industry is driven by the complex interactions of multiple entities, both local and global, that are embedded in a variety of networks (Coe et al., 2008). These networks contain different relations such as vertical buyer–supplier connections, horizontal competitive ties, institutional connections, interfirm and intrafirm linkages, personal diaspora connections as well as technological linkages (Turkina et al., 2016). These various networks are responsible for a region’s strategic coupling with global production networks thereby facilitating the region’s value creation, enhancement and value capture resulting in regional growth (Coe et al., 2004; Yeung and Coe, 2015). Given that our focus is on the innovation-led growth of locations, we study the technological linkages of locations. These develop as firms and inventors conduct technological searches and generate technological recombinations. While technological relatedness can only be judged ex post by an outsider to a focal innovation, the insider firm makes constant judgments about the underlying technologies and their combinative capabilities (Kogut and Zander, 1992). Thus, when the innovative activities of all the firms situated at a location are aggregated, one sees the realized industry-specific technological relatedness, which is discovered by the firms and the inventors situated at that location. Extending this exercise to include all of the industry-specific innovations reveals a technological network of locations for the industry, which shows the core and peripheral industry technologies and the locations’ connections to them. Going a step further, these connections can be weighted by the number of micro actors such as firms and their inventors situated at a location. A location that connects to a number of technologies, and connects strongly due to a large population of firms and inventors there, has high degree centrality and is very likely to be a technological hotspot of the industry (Pouder and St. John, 1996). Such a central location is tightly coupled to the industry’s global innovation network due to its high innovation output (Hannigan et al., 2015). Further, the network also shows how two locations are related by common technological knowledge. As the geographic scope of innovation widens in the modern knowledge economy, firms are increasingly expanding their knowledge networks to tap into distant knowledge sources (Cantwell and Mudambi, 2005). As firms and their inventors expand their technological search to include distant locations and develop extra-regional linkages, knowledge flows through a variety of conduits such as intrafirm, interfirm, institutional as well as personal ties (Soda and Zaheer, 2012; Lorenzen and Mudambi, 2013; Scalera et al., 2018), with the least common denominator being technological commonality among the locations. A central location in the network is thus likely to guide the nonlocal searches of a number of industry firms and inventors, thereby shaping the industry’s technological trajectory. However, as the industries evolve, the relative importance of the technologies changes. The firms change their location strategies and the locations experience a reorganization of their local assets (Martin and Sunley, 2011). Although manufacturing activities at a location may undergo drastic structural changes due to this reorganization, knowledge-related activities are sticky and a location’s innovation capabilities are more persistent (Sturgeon et al., 2008; Hannigan et al., 2015). In addition to the local knowledge stock, a location’s technological connections to other locations bring in new knowledge that finds innovative recombinations with the local knowledge, and assimilates with the local knowledge base. The combination of local and nonlocal knowledge reinforces a location’s innovation capabilities (Hannigan et al., 2015). We focus on the features of the patented knowledge stock of a location, specifically the breadth and depth of patented innovations. We then hypothesize their effects on the location’s future centrality in the industry’s technological network of locations. The participation of locations that are currently central in an industry’s global innovation network and those that are recent entrants, differ in systematic ways. A key contribution of our study is to develop theory and examine empirically the contrasting logics that drive these differences. Our analysis is based on the rich longitudinal data on technology classes, inventor locations and co-inventor linkages contained in the comprehensive population of wind turbine patents. We use it to build the associated global innovation network and analyze the logics that drive a location’s centrality within it. 2.1. Breadth and depth of patented innovation Patents are the manifestations of the underlying innovation activity. While innovative knowledge, including commercially valuable knowledge, exists in many forms, the most common form studied by the technology management literature consists of patents (Narin et al., 1987). Patents are considered as a proxy of firm innovation capabilities (Griliches, 1990; Jaffe et al., 1993) and have been shown to track very well with other measures of innovation like R&D expenditures and new product announcements (Hagedoorn and Cloodt, 2003). The process of generating patented innovations is often modeled using a knowledge production function (Griliches, 1979; Jaffe, 1989). An important input to this function is the existing knowledge stock. The knowledge production function highlights that new patented knowledge builds on the existing stock of patented knowledge (Porter and Stern, 2000). As early as the seminal work of Thomas Kuhn (1962), knowledge is shown to advance through two distinct processes of ‘exploration of new possibilities and exploitation of old certainties’ (March, 1991, 71). Exploration involves searching across disciplines and integrating diverse bodies of knowledge. In contrast, exploitation involves execution and refinement of existing knowledge, learning-by-doing (Arrow, 1962), and specialization. Integration and specialization are thus two distinct yet correlated processes (Mudambi et al., 2012). The integration process can be measured by the diversity of the knowledge building blocks that are recombined. In a like manner, the specialization process can be gauged in terms of the extent of specific and focused prior knowledge extant and used. In this sense, integration and specialization may be operationalized by the terms ‘breadth’ and ‘depth’, respectively. Breadth captures the number of discrete knowledge elements involved in creating the patented innovation as well as their importance to the focal industry. That is, it measures the technological scope of the patented innovation (Bierly and Chakrabarti, 1996; Katila and Ahuja, 2002; Lorenzen and Mudambi, 2013). Knowledge in more fields helps in identifying related knowledge elements that may exist in other technological areas. In addition to the raw number, the importance of these knowledge elements to the industry also factors into the definition of breadth. An industry’s core technologies when recombined with other technologies may find more direct application to the focal industry (Tushman and Murmann, 1998). Such exploratory recombinations are known to generate path-breaking innovations (Schumpeter, 1934; Penrose, 1959; Nelson and Winter, 1982). Depth, on the other hand, captures how well a certain technology is known. Depth develops as R&D activities continue in certain technologies. It often involves exploitative learning by repeating organizational routines (Levinthal and March, 1981). Over time, through learning-by-doing, deep knowledge of individual technologies and their interconnections is developed. For a firm, this deep technological acumen forms the basis of its core competencies making it a specialist in these technological areas (Hamel and Prahalad, 1994). For a location, the patented innovations are embodied in the firms and their inventors situated there (Almeida and Kogut, 1999). The literature on economic geography discusses such effective aggregation of the innovation activities of the micro entities at the local and regional level. For example, the prominent model of geographic industrialization by Storper and Walker (1989) assigns regional evolution to firms’ location strategies. Likewise, the literature on regional renewal and adaptive cycles describes how the micro entities shape location-level macro structures (Martin and Sunley, 2007, 2011) and how firms’ innovation activities change the regional innovation trajectory (Zhu et al., 2017). The literature on spatial knowledge creation places firms and their inventors at the center of the evolutionary dynamic. It extends the micro-level theories of individuals and firms to explain innovations at the aggregate level of a location (Maskell and Malmberg, 2007). The local technological milieu as well as the nonlocal pipelines of a location are indeed maintained and exploited by the micro actors (Bathelt et al., 2004; Lorenzen and Mudambi, 2013). As the knowledge elements of individual firms aggregate to form the local knowledge base, agglomeration effects and synergies increase the possible permutations and combinations of these knowledge elements that an individual firm can achieve. The aggregation of breadth and depth at the location level may therefore be more effective in generating innovative recombinations than firm-specific breadth and depth alone. As the micro actors conduct their technological searches in generating patented innovations, they use the breadth and depth of patented innovation already existing at a location. Locations that contain and support a large number of technologies, including those fundamental to the industry, exhibit a high breadth of patented innovation. The breadth of such locations may help firms identify exploratory innovative recombinations across technology fields. Such economies of technological scope may enable locations ‘to reap the intangible benefits of learning’ (p. 470) from many different high value-added local activities and facilitate a broad spectrum of production and entrepreneurial endeavors (Coe et al., 2004). On the other hand, the locations that contain and support a great concentration of certain technologies exhibit high depth in those technologies. It corresponds to the economies of scale achieved through ‘highly localized concentrations of specific knowledge, skills and expertise’ (p. 470) that attract firms to use the extant technology base at the location to identify niche and exploitative advances (Coe et al., 2004). However, a sole focus on either exploration (breadth) or exploitation (depth) is detrimental to innovation activity (March, 1991). Too much depth in certain technologies may lock in the current technological trajectory. As core competencies of the firms become core rigidities (Leonard-Barton, 1995), the location may experience an inward focus, groupthink and disconnect from key external stakeholders (Kher, 2010). It causes spatial myopia as the boundedly rational innovation actors select technological solutions that are closer to their existing knowledge base. Their search tends to be local not just technologically but also spatially (Maskell and Malmberg, 2007). With local labor primarily immobile, increase in the technological depth of a location and decrease in the variety may be natural consequences of the location’s evolution unless corrected by building global pipelines to distant knowledge sources and trans-local actors to access diverse knowledge (Coe et al., 2004; Lorenzen and Mudambi, 2013; Yeung and Coe, 2015; Turkina et al., 2016). Absent knowledge diversity, the local innovative specialization may continue beyond the point where buyers consider such advances to be valuable or worth paying for (Christensen 1993; Scalera et al., 2018). Further, returns to even the most valuable existing competencies eventually begin to decline (Mudambi and Swift, 2014). As the technological homogeneity in the local environment increases, firms in the technological hotspots find themselves less competitive than the non-hotspot firms (Pouder and St. John, 1996). Excessive breadth with comparatively less depth may also pose problems. A sole focus on integrating across various technological domains creates an excessive reliance on distant knowledge and increases the uncertainty of innovative outcomes (Cyert and March, 1963; Katila and Ahuja, 2002). The limits to the new knowledge, brought in by high breadth, are shown to exist in several contexts. For example, increases in a firm’s innovation breadth through structural hole spanning in its interfirm alliance network brings in new technological knowledge, but reduces interfirm collaboration and decreases its ultimate innovation performance (Ahuja, 2000). Likewise, a firm’s breadth in geographic space has been shown to have a negative curvilinear effect on innovation quality (Lahiri, 2010). Leveraging increases in innovation breadth also requires more cognitive and organizational bandwidth (Narula, 2014). Such bandwidth builds over time through intensive knowledge sharing between actors working in particular technologies. Without this, the search for recombinations solely based on breadth is often infeasible. Thus, while breadth and depth positively affect a location’s innovation activity, an excessive amount of either of the two has a negative effect. Further, they also have an interactive effect. A location with higher breadth has more high-breadth firms working in a variety of industry technologies. As a result, a higher number of distinct knowledge elements may be available locally. However, that may not translate into more recombinations unless these firms (or other local firms) have high depth in those technologies. The sheer mass of specialist knowledge in specific technological fields generates challenges for those who attempt to integrate it (Grant, 1996). If the location has predominantly low depth firms, they may find integrating these diverse knowledge elements beyond their cognitive capabilities (Cohen and Levinthal, 1990). Knowledge integration and specialization are thus correlated processes. The knowledge bases of the innovation leaders in most industries tend to be both broad and deep (Awate et al., 2012). We thus hypothesize the following: H1: A location’s breadth of patented innovation in terms of technology classes has a negative curvilinear effect on its degree centrality in the industry’s global technological network. H2: A location’s depth of patented innovation in the technology classes currently present has a negative curvilinear effect on its degree centrality in the industry’s global technological network. H3: A location’s breadth and depth of patented innovation have a positive joint effect on its degree centrality in the industry’s global technological network. As we have argued, breadth requires a location to support several industries whereas depth requires a location to support specialist firms operating in certain industries. The trade-off between depth and breadth brings us to the well-known debate about the relative importance of Marshall–Arrow–Romer (MAR) externalities and Jacobian externalities in industrial agglomeration. According to the MAR model, knowledge is industry-specific (Marshall, 1890; Arrow, 1962; Romer, 1986; Van der Panne, 2004) so that knowledge spillovers arise between firms within the same industry. These externalities are enhanced when a location specializes or builds depth in particular technologies. Jacobs (1969), on the other hand, noted that knowledge spills over across industries, resulting in novel recombinations, so that the key to innovative strength is industry diversity or technological variety. Innovations in general are new combinations of extant ideas (Schumpeter, 1934). From a technological standpoint, it implies that new industries arise when knowledge from several industries is usefully recombined to generate revolutionary products or services (Kodama, 1992). Such industry-creating knowledge recombinations include biomedicine, nanotechnology and new media (Feldman and Lendel, 2010) as well as digital photography (Tripsas, 2009). Accordingly, emerging industries are found to benefit more from Jacobian externalities of technological variety than MAR externalities of technological specialization (Henderson et al., 1995; Boschma and Frenken, 2011; Neffke et al., 2011). Additionally, from a regional standpoint, Storper and Walker (1989) argued that ‘industries create regional resources and not the other way around … Firms and sectors generate their own input histories, and those of their chosen regions, at the same time.’ (p. 96). Criticizing classic location theory which suggested that industries develop where the best locational conditions are, Storper and Walker proposed that the conditions do not exist per se but are created first and often as an outcome of industry activities. They argued that new industries form away from the established industrial centers as they thrive not on specialist inputs but modifiable generic resources. The new locations with such generic resources may be peripheral to the existing industrial nexus and may be present in both advanced and emerging economies, but are unlikely to be in the poorest underdeveloped countries (Mudambi and Santangelo, 2016). That is, these locations must contain enough generic resources providing them with the ‘potential to engage in global innovation networks’ (Mudambi and Santangelo, 2016, 3). The generic resources may be infrastructure, institutions, universities and supplier networks, but more importantly for this paper, a skilled labor force that can undertake standardized activities in established, mature technologies. These technologies also offer a significant amount of standardized and codified knowledge through patented innovations. The standard knowledge of mature technologies when recombined may find innovative applications in the new industries. Firms may recombine such generic knowledge of mature technologies available locally with firm-specific knowledge to rebundle the existing technologies and find a new combination that generates growth (Bathelt and Boggs, 2003). Thus, knowledge about a number of existing technologies, including those core to the new industry (which we call the breadth of patented innovation) is more effective for emerging industry patented innovations than specialization in particular technologies (which we call the depth of patented innovation). We thus hypothesize the following: H4: In an emerging industry, a location’s breadth of patented innovation has a larger effect than its depth of patented innovation on its degree centrality in the industry’ technological network of locations. 2.2. Breadth, depth and the entry of locations in global technological network Why does a new industry emerge away from old industrial areas? There are two reasons as argued in our theoretical arguments. One is due to the nature of the emerging industry, which originates from the recombination of established and mature technologies. These recombinations may involve standardized knowledge rather than specialized inputs of the core industrial areas. The other reason, which Storper and Walker put forth, involves the old and established industrial center’s unwillingness to support new sectors. The old center becomes ‘dominated by the entrenched interests of the old sector’ (Storper and Walker, 1989, 361) and local labor relations become too inflexible to accommodate the requirements of the new sector. We note that the established firms and centers may participate in emerging industry innovations but they may not value the new industries as much as their (old) focal industries. Rather, they may look at emerging industry innovations opportunistically. As new industries branch out from existing industries, various locations appear in the technological network of newly emerging industries. There is likely to be significant heterogeneity in terms of the innovation activity at these locations prior to entering the emerging industry. The entering locations may be well-established innovative locations with a long history of innovation and thus a high total output of patented innovations prior to entering the emerging industry. Or, these may be emergent innovative locations with relatively lower total patented innovation output prior to entry. In addition to entry timing, the established and emergent locations may also differ in terms of the firms situated there. As firms geographically disperse their value chains (Mudambi, 2008), they are more likely to prefer locations with agglomeration externalities such as knowledge spillovers and the availability of skilled labor for their R&D activities (Storper, 1995; Almeida, 1996; Alcácer, 2006; Beugelsdijk and Mudambi, 2013). On average, established locations with a history of high innovation output are more likely to have both generic and specialized local knowledge resources and social institutions promoting knowledge externalities when compared with emergent locations (Mudambi and Santangelo, 2016). In absolute terms, established locations may have more breadth and depth of patented innovations than emergent locations. As established locations branch out into new industries, part of their technological knowledge overlaps with the new industry. They may continue to exploit this overlap for emerging industry innovation and increase depth within those technologies. They may also explore and expand their technological knowledge for emerging industry innovation and increase breadth in the industry’s technologies. However, exploration in an emerging industry is doubly uncertain. In addition to the industry uncertainty noted earlier, there is exploration uncertainty where payoffs are long term and uncertain (March, 1991). However, as we argue, for emerging industries, breadth of patented innovation associated with exploratory search has a larger effect on the location’s degree centrality than depth of patented innovation associated with exploitative search. Thus, although doubly uncertain, exploratory breadth may have benefits. On the surface, such uncertain yet rewarding exploration and breadth may appear to be the strength of established locations given their long innovation experience and high innovation output. However, exploration is an option that firms often choose not to exercise (Kogut and Zander, 1992). There is strong evidence that successful firms’ search is path dependent (Penrose, 1959; Kogut and Zander, 1992; Teece, 1994; Patel and Pavitt, 1997) which aggregates at the level of the location (Storper, 1995; Maskell and Malmberg, 2007). Firms diversify into new technologies closer to their existing technological knowledge and changes in their technological profiles occur very slowly (Jaffe, 1986). Extending this logic to locations, cluster studies noted the presence of cluster identity within which the agglomerated firms define their field of competition (Florida and Kenney, 1990; Pouder and St. John, 1996). Deviating from this identity and competing in new technologies may be costly to the clustered firms given their capabilities, resources and competitive pressures (Porter, 1980; Kogut and Zander, 1992). Further, competitiveness in existing industries can impact the evaluation of a technological opportunity (Jaffe, 1986), especially if it does not directly address existing customer needs (Christensen and Bower, 1996). Therefore, exploitative search focused on short-term outcomes may appear more lucrative than exploratory search with longer-term outcomes (March, 1991). Although a sole focus on short-term profitability is detrimental to a firm’s long-run competitiveness, an established firm often does not have the flexibility and risk tolerance necessary to succeed when moving away from its existing knowledge base (Kogut and Zander, 1992). Building on this logic, we argue that, when evaluating an emerging industry, a majority of firms at established locations may prefer exploitative within-technology search resulting in increasing depth of innovation. It is likely that only a small minority of resident firms in such locations will focus on the breadth of innovation in the emerging industry. Moreover, firms deciding to locate in established locations may be attracted by the high level of specialized resources present there (Mudambi and Santangelo, 2016). Firms’ conscious decisions to exploit these specialized resources would then result in a greater depth of innovation at the established locations. Established locations may thus participate in the emerging industry innovation network more through their depth than breadth. Emergent locations host fewer innovative firms than established locations and therefore offer relatively lower agglomerative benefits. However, the resident firms may well be very successful innovators. Since leading firms can tap into local resources more readily than lagging ones (Cantwell and Mudambi, 2011), they may be willing to consider emergent locations that have limited specialized knowledge resources but high-quality generic resources (Mudambi and Santangelo, 2016). Generic resources may support the firms’ across-technology exploratory search resulting in increasing breadth of innovation at the emergent location. The concepts of established and emergent are relative to each other and so are our arguments. In absolute terms, emergent locations may be low in both breadth and depth. Relatively however, they may focus more on breadth than depth when compared with the established locations when innovating for the emerging industry. That is, their participation in the industry innovation network may be more through their breadth than depth. We thus hypothesize the following: H5a: In an emerging industry, established innovative locations are more likely to increase technological depth rather than technological breadth in the industry’s technologies. H5b: In an emerging industry, emergent innovative locations are more likely to increase technological breadth rather than technological depth in the industry’s technologies. Combining the arguments from Hypotheses 4 and 5, we state the following: emergent locations focus more on breadth which has a stronger effect on a location’s degree centrality in the global innovation network than depth. Taken together, these statements imply that emergent locations that host successful innovative firms are likely to rapidly achieve centrality in the global innovation network underpinning emergent industries. Thus, although emergent locations have lower innovation experience, fewer innovative firms and thus lower innovation output prior to entering the emerging industry, they may still exhibit fast accretion toward centrality in the industry’s global innovation network with respect to established locations. Accretion toward centrality does not mean that emergent locations achieve parity with established locations (Awate et al., 2012, 2015). However, it is indicative of a significant narrowing of the technological gap between established and emergent locations. 3. The global wind power industry Wind power is one of the world’s fastest growing sources of energy. In 2011, the global cumulative installed wind power capacity was 237,669 megawatts (MW) (GWEC, 2012), capable of providing around 3% of global electricity consumption (WWEA, 2012). The industry is highly concentrated with a few large manufacturers together having a market share of over 50%. The top five manufacturers in 2011 were Vestas (12.7%), Sinovel (9%), Goldwind (8.7%), Gamesa (8.0%) and Enercon (7.8%), with GE (7.7%) and Suzlon (7.6%) close behind. Western (and particularly European) firms such as Vestas, NEG (Denmark) and Enercon (Germany) have traditionally dominated the industry. However, in the 1990s, it witnessed the entrance of a number of emerging economy manufacturers that offered turbines for considerably lower prices than the more established industry players (see Musgrove, 2010, especially Chapter 5). In particular, Chinese and Indian companies such as Goldwind (China), Sinovel (China) and Suzlon (India) have displayed highly impressive growth rates, both domestically and internationally, and have begun to challenge the more established players in the industry (Awate et al., 2012). 3.1. Genesis and growth Early experiments with the use of wind energy to generate electricity began in the late eighteenth century in Europe and the USA. During the two world wars, the restrictions on fossil fuel imports started wind turbine development in numerous countries, but these developments were small scale and experimental. The oil crises of the 1970s were the triggering events that resulted in the establishment and growth of the modern wind turbine industry. In the early 1980s, the industry experienced a number of product innovations primarily from two competing technological trajectories, one from California in the USA, and another from Jutland in Denmark (Garud and Karnøe, 2003; Musgrove, 2010; Nielsen, 2010). Providing further details on the design competition in the industry, Awate et al. (2012) note that ‘in California, the U.S. Department of Energy and NASA engaged a number of engineers in response to the oil crises to cooperate with companies in the aircraft industry to develop sophisticated, high-technology, large-scale and aerodynamically-optimized turbines based on aeronautical engineering principles. These turbines were particularly distinguished by their two-bladed rotor pitch regulation. In Denmark, however, different wind power enthusiasts such as farmers, carpenters, and engineers collaborated to develop robust, small-scale three-bladed turbines with reliability and ruggedness as the key concerns. While these turbines were initially small in size, a number of incremental innovations led to the wind turbines eventually being scaled up to meet broader commercial demands. During the California wind boom of the 1980s, the Danish low technology—high reliability wind turbine proved commercially superior to its American high technology—low reliability counterpart. As a result, the Danish turbine emerged in the late 1980s as the dominant industry standard with Danish firms and designs controlling the majority of the world market share (Garud and Karnøe, 2003).’ 3.2. Technological development The basic design introduced by Danish manufacturers in the late 1980s has not seen any radical changes, although wind turbines today are both larger in size and typically use blades that optimize the output of the turbine by automatically adjusting with changing wind directions. The average blade span of the larger wind turbines has increased from about 30 m in 1990 to 90 m by 2008. New generator designs also allow the rotor to operate at varying speeds. Overall, the successive generations of turbines have been developed with a particular focus on increasing scale and reducing the overall cost of energy (Musgrove, 2010). Electricity and aerodynamics are the industry’s core technologies and in addition it uses a variety of other peripheral technologies from sectors such as automobiles, construction, chemicals, information and communication technology and so on. In addition, innovations in linked industries such as energy storage systems have a large impact on industry’s growth (Gates, 2011). In the early 2000s, the industry witnessed a major change—the offshore era. Offshore wind generation was particularly attractive because wind speeds are stronger and steadier on the sea than on the land, which increases turbine ‘availability’, an industry term that roughly corresponds to capacity utilization. Higher availability leads to more electricity produced per installed MW, and therefore cheaper electricity generation. However, the installation of such large offshore turbines and their connection to the grid poses a major challenge. Moreover, the design and engineering involved in such large offshore turbines is far different from the mid-scale onshore turbines. The diverse capabilities, risks and large capital investment necessary for offshore wind turbine projects required the participation of a whole new set of industry players. These included firms from various sectors such as oil and natural gas, electricity (Markard and Petersen, 2009), shipbuilding and heavy engineering. Many of them were large multinationals with deep pockets and strong R&D skills. Some examples include General Electric (GE), Siemens, Hitachi and Mitsubishi. As a result, a major change also came in the form of a dramatic shift in the industry’s IPR regime. More and more firms and independent inventors began patenting their technology. Established wind turbine manufacturers like Vestas that had formerly relied on informal innovation networks, began rethinking their R&D strategies in the offshore era (Pedersen and Larsen, 2009). The offshore era thus represented a discontinuous change for the industry that resulted in increased entry and fiercer competition. 3.3. Policy and market activity While the patented technological innovations of the wind industry are the core of our study, we would like to point out that the establishment and growth of its market has always been dependent on the government support. The support has been in the form of production tax credits (PTCs) that make wind power prices comparable to conventional power (Cardwell, 2013). Additionally, policy instruments such as guaranteed grid access facilitated the growth of the wind industry in several countries. Among the leading market locations, Denmark was the pioneer in offering government support for the growth of the wind industry from its early years. The government policies involved long-term agreements with power companies, producers and users. The government has provided investment subsidies to individuals and cooperatives investing in wind turbines since the late 1970s. Further, investments in the sector and the sale of surplus electricity were tax deductible, which popularized investment in wind power and increased public support (Buen, 2006). In the 1980s, the government negotiated a long-term agreement with the power companies by which they agreed to guarantee grid connection for turbines operated by individuals and cooperatives, pay part of the grid connection cost and buy excess power from them at 85% of the consumer price. Danish policies served as a model for a number of European markets. The German government, for example, adopted the Danish focus on small-scale wind energy in the late 1980s by assuring demand for these turbines. These were soon ramped up in 1991 to support utility scale electricity generation. The policies included feed-in tariffs to wind power producers to feed wind power into the grid at a fixed price. Like Denmark and Germany, other European markets such as Spain, Portugal and Ireland established similar policy instruments to encourage market growth. In addition to the government policies, countries such as the UK and Italy also opted for market-based mechanisms (e.g., tradable green certificates) to reduce the cost of wind generation. Across various European countries, the policies were applied more or less consistently over the years (GWEC, 2012). In the USA, however, the governmental support was comparatively inconsistent with the main policy instrument being PTC. PTC however is a performance-based incentive with a set expiry, which has often resulted in uncertainty among wind power producers and delayed investment and short-term market slowdown. After economic reforms, the rapid economic development of emerging markets, particularly India and China, exerted excessive demand on their national energy supply. The large power deficits from the traditional energy sources made renewable sources such as wind very attractive to those governments. In the late 1980s, the Indian government began a process of creating an open economy environment and implemented favorable policies to encourage entry by foreign firms in grid-quality wind energy generation. These included 100% accelerated depreciation on the wind equipment, customs and excise duty relief, 5-year tax holiday and soft loans (Rajsekhar et al. 1999). The favorable policy environment supported the entry of European manufacturers such as Vestas and NEG that entered through joint ventures with local firms. Compared with India, the Chinese market started developing slowly. In the 1980s and the early-1990s, the projects were small-scale, demonstrational and funded through foreign grants and government loans. However, in the1990s, the government introduced a number of policy instruments that resulted in exponential increases in the installed capacity. 4. Methodology 4.1. Data We examine patents generated at each location and track them over time. We gathered all the patents filed with the USPTO that claimed wind turbine innovations. The final dataset included 1895 patents filed and granted between 1961 and March 2012. From each patent, we extracted the inventors, their locations and the technological classifications (or technology classes) assigned to them by the USPTO. We then performed manual checks to identify unique locations and inventors. We applied a 30-mile radius threshold to group nearby locations. This analysis yielded a total of 566 unique locations, 2204 unique inventors1 and 177 technology classes over the study period. Next, we created the industry’s technological network, outlined earlier, by connecting technology classes with the locations. Thus, a tie existed between the two nodes of technology and geography if there was an inventor working in that particular technology class and situated at that particular location.2 The number of such inventors appeared as the weight for that tie. Further, if the same location-inventor-technology pattern repeated on multiple patents within the same year, the number of such patents was also accounted for in the tie weight. Such networks of technology classes and locations were created for each year of the cumulative patent records. Figure 1 shows the network in 2011. The nodes are placed depending on their connections. For example, the top four classes in terms of the number of patents (classes 290, 322, 415 and 416) are at the core of the map encircled by several locations that directly connect to them. Classes 290, 322 and 415 gather a number of locations shown by a large cluster of red nodes at the center of the figure. The location nodes appearing in the middle of this cluster, for example San Francisco in California and Aalborg in Denmark, primarily connect to the three core classes. The location nodes on the perimeter connect to the core classes as well as a number of other classes. For example, the top locations in terms of the patent output, Schenectady in New York and Greenville in South Carolina, connect to a host of other classes depicted by a cluster of blue squares below these locations. The two emerging market locations, namely Shanghai in China and Bangalore in India, also appear closer to the perimeter indicating that they connect to the core as well as non-core classes. Figure 1 View largeDownload slide Technology-geography network in 2011. Notes: The figure is drawn with a spring embedding layout using UCINET’s NetDraw software (Borgatti et al., 2002). For clarity, tie weights are not shown. Figure 1 View largeDownload slide Technology-geography network in 2011. Notes: The figure is drawn with a spring embedding layout using UCINET’s NetDraw software (Borgatti et al., 2002). For clarity, tie weights are not shown. 4.2. Variables 4.2.1. Dependent variable We measured the dependent variable representing a location’s innovation activity (as measured by patent output) by focusing on the locations in the two-mode networks. We used a location node’s valued degree centrality (vdeg) to measure the patenting activity of that location. This centrality was calculated as the sum of tie weights of all ties originating from the location. Thus, the centrality score increases with an increase in the individual tie weights or in the number of ties or both. In other words, it increases in value if the inventors at a location invent more in the same technology class (increase in tie weight) as well as if they invent more in different technology classes (increase in the number of ties). Further, if the location adds new inventors, the increase is also captured in the centrality score through increase in tie weight. Thus, our dependent variable captures patenting activity along the dimensions of technologies as well as inventors. Raw patent counts treat both kinds of activities as one and underestimate the total patenting activity at a location. Our dependent variable on the other hand provides a more accurate measure of a location’s patenting activity. 4.2.2. Explanatory variables Depthi,t−1: Depth develops when the location repeatedly invents in a certain set of technologies. We thus measured the depth of patented innovation at a location by finding the number of ties the location adds in the same technology class every year. As the two-mode networks contain cumulative data, we computed the number of ties added to the existing technology classes from previous year. The variable was lagged by a year to denote the depth of patented knowledge base as of previous year. This was necessary since we study how the existing patent stock of a location impacts its patenting activity. The variable and its squared transformation were used in the regressions to test H2. Breadthi,t−1: Breadth develops when the location increases the number of technology classes in which it invents, particularly when it adds classes that are important in the wind turbine innovation network. We gauged breadth using the closeness centrality measure which denotes the reach of a node within the network (Faust, 1997). With this measure, the reach of a location increases with the number of direct connections to distinct technology classes. Further, the measure also takes into account the importance of those technology classes within the network. This is because a location connected to more central classes has a higher reach in the overall network than a location connected to less central, peripheral classes. Thus, even if two locations are directly connected to the same number of classes, the location connected to more central classes has a higher reach in the network. We calculated breadth of a location as sum of its reciprocal distances to all other technology nodes in the network. In this way, breadth is captured at the network level and captures all the changes happening in the topology of the network through small increments. This breadth variable and its squared transformation were used in the regressions to test H1. As breadth is a network-level measure, it increases when the location makes a connection to a new technological class as well as when a previously connected class becomes more central in the network. Thus, the breadth values in our dataset continue to increase in small increments whenever there is a change in the network structure and even if the location does not produce new patents or has a new depth-only patent. Pre-entry patent output: Pre-entry patent output helps to qualify established and emergent locations to test hypotheses H5a and H5b. It is measured as the location’s share of the USPTO patents one year before filing its first wind turbine patent. As the denominator (i.e., total number of USPTO patents) keeps growing each year, only those locations that continue to file patents and increase the numerator achieve higher score on this measure. This is a time invariant variable. For location i that entered in year t1, it is denoted as: Pre-entry patent outputi, t1 = total patents filed by i until (t1 − 1)/total USPTO patents filed until (t1−1). By measuring the pre-entry patent output, the variable identifies locations that may be new to the wind power industry but otherwise established innovators. We also use a dummy variable to distinguish developing economy location from advanced economies. The dummy variable is created following World Bank classification and identifies locations that are likely to be new to both, patents in general and wind patents in particular. 4.2.3. Controls Market activity: Demand is often shown to drive innovative clusters (Bresnahan et al., 2001). We used the installed wind power capacity to measure the market activity and thus demand. The data were only available at country-level (GWEC, 2012). It is likely that countries with higher installed capacity may consider wind power as an important source of electricity and may encourage innovation activity in this industry. This is a time variant variable. Global connectivity: Given the important role played by global knowledge sources, a location’s global connectivity may positively impact its innovation activity. We measure global connectivity along two dimensions—physical connectivity and virtual connectivity. Physical connectivity is measured as a ratio variable taking into account how far a location is from an airport scaled by the airport size, as follows:   Airport connectivity=(number of destinations served by the nearest international airport)/(distance to the nearest international airport). Using this variable, we capture the effect of being in the vicinity of a larger airport versus a smaller airport. Availability of air services is shown to promote information exchange between cities and a factor deciding headquarters location choice (Bel and Fageda, 2008). We thus argue that the variable airport connectivity captures a location’s physical access to global sources of information and the kind of firms that are located there, which impacts the innovation output of the location. The Google Maps program was used to find the shortest distance in miles to the nearest international airport from each of the location. This is a time invariant variable. Similarly, virtual connectivity to the world may also be necessary to access documented knowledge through the World Wide Web. We use international Internet bandwidth, which is the capacity measured as bits per person. The variable is time variant as it is based on the longitudinal data obtained from International Telecommunications Union and World Bank estimates provided by the EconStats database (ITU, 2005; EconStats, 2012). Patent impact: A location that produces more impactful innovations may be more likely to attract entry by other firms. As more firms enter, the number of inventors at a location may increase. This may increase the location’s patenting activity. The impact of a patent is often measured using the number of citations the patent receives. Accordingly, we found the total number of citations received by a location’s patents for each year. For each year, this variable captures the cumulative number of citations received. We then lagged this variable by 1 year, so that it captured the impact of a location’s patents as of previous year. Of course, patents granted in earlier years may receive more citations than newer patents. We follow Rosenkopf and Nerkar (2001), and control for this bias by including year dummies. Innovativeness of technologies: A location’s innovation activity may be influenced by its technological connections. That is, a location that operates in the most innovative technologies is likely to be more innovative than other locations. In our network model, eigenvector centralities measure how central the locations’ connections are (in terms of patenting activity) and the strengths of those connections (Faust, 1997). We computed this centrality measure, but found it to be highly correlated with the explanatory variables. The variable was thus dropped. As an alternative approach, we computed the valued degree centralities of technology classes to identify the most innovative classes in terms of the patent output. This procedure enabled us to identify the top four innovative classes (classes 290, 322, 415 and 416). These classes belonged to the core technologies of the wind power industry, namely electricity (classes 290, 322) and aerodynamics (classes 415, 416). We used dummy variables to indicate locations’ connections to these classes. Post technological discontinuity dummy: Patenting activity in the industry increased considerably with the offshore technological discontinuity post-2000. We included a dummy variable to capture the effect of this technological discontinuity on locations’ patenting activity. Country controls: As the innovativeness varies considerably across countries, we controlled for this heterogeneity using dummy variables for the top three countries in terms of patenting activity. We calculated the patenting activity of a country by summing the cumulative measure of valued degree centralities of locations in that country in 2011. Accordingly, the top three innovative countries were the USA, Denmark and Germany. Finally, we also controlled for country-level GDP per capita, using data obtained from the World Bank. Additionally, we used a dummy variable to denote developed countries. 5. Estimation and results We present univariate descriptive statistics, including correlations in Table 1. The correlations generally exhibit the expected signs and none are large enough to cause serious concern about multicollinearity as indicated by the variance inflation factor (VIF). The mean VIF for all the variables is 1.55. As can be seen, the dependent variable is an over dispersed count variable. This led us to opt for a negative binomial regression methodology for our location-year panel data as the Poisson regression model would be too restrictive given its equidispersion property. Several of our variables were time invariant. Further, although many explanatory variables were time-variant, several locations exhibited relatively low levels of intertemporal variation. In these circumstances, it is generally preferable to choose a random-effects model, as using a fixed-effects model would lead to many data points being omitted (Hsiao, 1986). Table 1 Univariate statistics and variable correlations Variable  1  2  3  4  5  6  7  8  9  (1) Vdeg                    (2) Breadth  0.4303                  (3) Depth  0.9674  0.3129                (4) ln(Installed wind capacity)  0.1936  0.4632  0.1228              (5) Pre-entry innovation output  0.2099  0.1332  0.1861  0.0105            (6) Airport connectivity  0.056  0.0802  0.0293  −0.035  0.1818          (7) ln(Bandwidth)  0.0805  0.2186  0.0332  0.3492  0.0065  0.0065        (8) Patent impact  0.6355  0.4308  0.5866  0.1959  0.3454  0.1112  0.1039      (9) ln(GDP per capita)  0.1850  0.4224  0.1227  0.6129  0.0629  0.0186  0.2784  0.1899    Mean*  2.6316  0.7814  1.2473  2.7498  0.0866  2.3555  −7.9092  19.0190  9.6260  (9.1680)  (1.0905)  (6.7858)  (8.4733)  (0.3095)  (2.2243)  (8.0707)  (54.3449)  (1.1512)  Min  0  0  0  −12.2061  0  0.0156  −12.206  0  4.3291  Max  414  4.5574  376  11.0407  3.9518  10.4667  11.2665  689  11.4636  Mean VIF  1.55  Variable  1  2  3  4  5  6  7  8  9  (1) Vdeg                    (2) Breadth  0.4303                  (3) Depth  0.9674  0.3129                (4) ln(Installed wind capacity)  0.1936  0.4632  0.1228              (5) Pre-entry innovation output  0.2099  0.1332  0.1861  0.0105            (6) Airport connectivity  0.056  0.0802  0.0293  −0.035  0.1818          (7) ln(Bandwidth)  0.0805  0.2186  0.0332  0.3492  0.0065  0.0065        (8) Patent impact  0.6355  0.4308  0.5866  0.1959  0.3454  0.1112  0.1039      (9) ln(GDP per capita)  0.1850  0.4224  0.1227  0.6129  0.0629  0.0186  0.2784  0.1899    Mean*  2.6316  0.7814  1.2473  2.7498  0.0866  2.3555  −7.9092  19.0190  9.6260  (9.1680)  (1.0905)  (6.7858)  (8.4733)  (0.3095)  (2.2243)  (8.0707)  (54.3449)  (1.1512)  Min  0  0  0  −12.2061  0  0.0156  −12.206  0  4.3291  Max  414  4.5574  376  11.0407  3.9518  10.4667  11.2665  689  11.4636  Mean VIF  1.55  *Standard deviations in parentheses. 5.1. Breadth and depth The results of the panel negative binomial regressions are shown in Table 2. The first model contains only the control variables. The chi-squared test statistic indicates that the model as a whole is significant relative to the null model. The control variables are generally statistically significant with the expected signs. Table 2 Random effects negative binomial regression results Dependent variable Vdeg  (1) Controls only  (2) Breadth and square added  (3) Depth and square added  (4) Interaction added  (5) Standardized variables  (6) Incident rate ratio  Breadth    1.9322***  1.8545***  1.9570***  1.3286***  3.7756***      (0.049)  (0.044)  (0.046)  (0.039)  (0.147)  Breadth2    −0.3392***  −0.4226***  −0.4773***  −0.5677***  0.5669***      (0.012)  (0.011)  (0.013)  (0.016)  (0.009)  Depth      0.0235***  0.0066***  0.0743***  1.0771***        (0.001)  (0.002)  (0.013)  (0.014)  Depth2      −0.00003***  −0.0001***  −0.0028***  0.9972***        (0.000002)  (0.000004)  (0.0002)  (0.0002)  Breadth×depth        0.0058***  0.0430***  1.0439***          (0.001)  (0.006)  (0.006)  Pre-entry innovation  −0.1590*  0.0794  0.1871*  0.1877*  0.0581*  1.0598*  output  (0.085)  (0.084)  (0.096)  (0.097)  (0.030)  (0.032)  ln(Installed wind)  0.0165*  0.0130  0.0070  0.0053  0.0448  1.0458    (0.009)  (0.009)  (0.009)  (0.009)  (0.073)  (0.076)  Airport connectivity  −0.0003  0.0454***  0.0357**  0.0334**  0.0743**  1.0771**    (0.015)  (0.014)  (0.014)  (0.014)  (0.031)  (0.034)  ln(Bandwidth)  0.0029  −0.0001  −0.0017  −0.0018  −0.0145  0.9856    (0.002)  (0.002)  (0.002)  (0.002)  (0.015)  (0.015)  Patent impact  0.0018***  0.0024***  0.0002  0.0004***  0.0226***  1.0228***    (0.000)  (0.000)  (0.000)  (0.000)  (0.008)  (0.008)  ln(GDP per capita)  0.4105***  0.1591***  0.1149***  0.1188***  0.1368***  1.1466***    (0.041)  (0.039)  (0.039)  (0.039)  (0.045)  (0.052)  Class 240  1.1338***  0.6480***  0.7791***  0.8120***  0.8120***  2.2525***    (0.030)  (0.030)  (0.029)  (0.029)  (0.029)  (0.066)  Class 320  0.3375***  0.3326***  0.3550***  0.3698***  0.3698***  1.4474***    (0.030)  (0.030)  (0.030)  (0.030)  (0.030)  (0.043)  Class 415  0.5075***  0.4106***  0.4100***  0.4273***  0.4273***  1.5331***    (0.027)  (0.026)  (0.026)  (0.026)  (0.026)  (0.039)  Class 416  1.5731***  0.7688***  0.8689***  0.8901***  0.8901***  2.4355***    (0.037)  (0.039)  (0.037)  (0.037)  (0.037)  (0.091)  USA  −0.2680***  0.0666  0.3072***  0.2962***  0.2962***  1.3448***    (0.087)  (0.074)  (0.075)  (0.075)  (0.075)  (0.101)  Denmark  −1.1085***  −0.3882***  −0.1033  −0.1447  −0.1447  0.8653    (0.158)  (0.141)  (0.144)  (0.145)  (0.145)  (0.125)  Germany  −0.4931***  −0.2115*  −0.0204  −0.0205  −0.0205  0.9797    (0.131)  (0.111)  (0.111)  (0.111)  (0.111)  (0.109)  Post-discontinuity  1.7033***  0.5846*  1.5605***  1.7283***  1.7283***  5.6311***    (0.308)  (0.334)  (0.312)  (0.311)  (0.311)  (1.754)  Constant  −5.0319***  −3.2699***  −2.6156***  −2.6673***  −0.1403  0.8691    (0.418)  (0.415)  (0.406)  (0.406)  (0.276)  (0.240)  Observations  16,053  16,053  16,053  16,053  16,053  16,053  Number of locations  557  557  557  557  557  557  Chi-squared  20,290.9***  18,607.9***  21,080.3***  20,956.2***  20,956.2***    Log-likelihood  −19,016.4  −17,790.8  −17,378.4  −17,349.83  −17,349.83    L-R test statistic    2451.2***  824.8***  57.17***      Wald: Breadth=Depth        2041.4***  1277.8***    Wald: Breadth2=Depth2        1316.5***  1316.5***    Dependent variable Vdeg  (1) Controls only  (2) Breadth and square added  (3) Depth and square added  (4) Interaction added  (5) Standardized variables  (6) Incident rate ratio  Breadth    1.9322***  1.8545***  1.9570***  1.3286***  3.7756***      (0.049)  (0.044)  (0.046)  (0.039)  (0.147)  Breadth2    −0.3392***  −0.4226***  −0.4773***  −0.5677***  0.5669***      (0.012)  (0.011)  (0.013)  (0.016)  (0.009)  Depth      0.0235***  0.0066***  0.0743***  1.0771***        (0.001)  (0.002)  (0.013)  (0.014)  Depth2      −0.00003***  −0.0001***  −0.0028***  0.9972***        (0.000002)  (0.000004)  (0.0002)  (0.0002)  Breadth×depth        0.0058***  0.0430***  1.0439***          (0.001)  (0.006)  (0.006)  Pre-entry innovation  −0.1590*  0.0794  0.1871*  0.1877*  0.0581*  1.0598*  output  (0.085)  (0.084)  (0.096)  (0.097)  (0.030)  (0.032)  ln(Installed wind)  0.0165*  0.0130  0.0070  0.0053  0.0448  1.0458    (0.009)  (0.009)  (0.009)  (0.009)  (0.073)  (0.076)  Airport connectivity  −0.0003  0.0454***  0.0357**  0.0334**  0.0743**  1.0771**    (0.015)  (0.014)  (0.014)  (0.014)  (0.031)  (0.034)  ln(Bandwidth)  0.0029  −0.0001  −0.0017  −0.0018  −0.0145  0.9856    (0.002)  (0.002)  (0.002)  (0.002)  (0.015)  (0.015)  Patent impact  0.0018***  0.0024***  0.0002  0.0004***  0.0226***  1.0228***    (0.000)  (0.000)  (0.000)  (0.000)  (0.008)  (0.008)  ln(GDP per capita)  0.4105***  0.1591***  0.1149***  0.1188***  0.1368***  1.1466***    (0.041)  (0.039)  (0.039)  (0.039)  (0.045)  (0.052)  Class 240  1.1338***  0.6480***  0.7791***  0.8120***  0.8120***  2.2525***    (0.030)  (0.030)  (0.029)  (0.029)  (0.029)  (0.066)  Class 320  0.3375***  0.3326***  0.3550***  0.3698***  0.3698***  1.4474***    (0.030)  (0.030)  (0.030)  (0.030)  (0.030)  (0.043)  Class 415  0.5075***  0.4106***  0.4100***  0.4273***  0.4273***  1.5331***    (0.027)  (0.026)  (0.026)  (0.026)  (0.026)  (0.039)  Class 416  1.5731***  0.7688***  0.8689***  0.8901***  0.8901***  2.4355***    (0.037)  (0.039)  (0.037)  (0.037)  (0.037)  (0.091)  USA  −0.2680***  0.0666  0.3072***  0.2962***  0.2962***  1.3448***    (0.087)  (0.074)  (0.075)  (0.075)  (0.075)  (0.101)  Denmark  −1.1085***  −0.3882***  −0.1033  −0.1447  −0.1447  0.8653    (0.158)  (0.141)  (0.144)  (0.145)  (0.145)  (0.125)  Germany  −0.4931***  −0.2115*  −0.0204  −0.0205  −0.0205  0.9797    (0.131)  (0.111)  (0.111)  (0.111)  (0.111)  (0.109)  Post-discontinuity  1.7033***  0.5846*  1.5605***  1.7283***  1.7283***  5.6311***    (0.308)  (0.334)  (0.312)  (0.311)  (0.311)  (1.754)  Constant  −5.0319***  −3.2699***  −2.6156***  −2.6673***  −0.1403  0.8691    (0.418)  (0.415)  (0.406)  (0.406)  (0.276)  (0.240)  Observations  16,053  16,053  16,053  16,053  16,053  16,053  Number of locations  557  557  557  557  557  557  Chi-squared  20,290.9***  18,607.9***  21,080.3***  20,956.2***  20,956.2***    Log-likelihood  −19,016.4  −17,790.8  −17,378.4  −17,349.83  −17,349.83    L-R test statistic    2451.2***  824.8***  57.17***      Wald: Breadth=Depth        2041.4***  1277.8***    Wald: Breadth2=Depth2        1316.5***  1316.5***    Note: Standard errors in parentheses, ***p < 0.01, **p < 0.05, and *p < 0.1. Model 2 adds the breadth variable along with its squared term. While breadth has a positive and significant coefficient, its squared term has a negative and significant effect signaling the possibility of a negative curvilinear effect. Model 3 adds the depth variable with its squared term. Both breadth and its squared term retain their effects after this addition. Depth has a positive and significant coefficient and its squared term produces a small negative but significant effect. Next, Model 4 adds the interaction term of breadth and depth, whose coefficient is positive and significant. The models are significant as a whole when compared with a null model, shown by the chi-squared statistic. Further, the log-likelihood increases with the addition of successive regressors from Model 1 to Model 4. We conduct likelihood ratio (L-R) tests to check if the nested models are significantly different from one another (UCLA, 2013). As shown by the L-R test statistic, adding explanatory variables to the control model 1 significantly improves the model fit. Further, each model offers a better fit over its predecessor. Finally, and perhaps most importantly, the coefficients of all the key variables retain their signs and significance through all models 1–4. Model 4 is the enveloping model and we use it to test our hypotheses. Both the breadth and depth variables have positive and significant coefficients. Their squared terms are negative and significant, so that both breadth and depth have negative curvilinear effects on a location’s patenting activity. This evidence supports H1 and H2. Further, their interaction term has a positive and significant effect on the dependent variable thereby supporting H3. 5.2. Breadth versus depth We test H4 by estimating Model 4 using standardized independent variables with zero mean and unit variance. These results are reported in Model 5 in Table 2. As seen, the positive significant coefficient of breadth is larger than that of depth. A Wald test indicates that this breadth coefficient is significantly larger than the corresponding depth coefficient, reported in Table 2. Further, the negative coefficient of the squared term of breadth is found to be significantly more negative than the corresponding coefficient of depth (also shown in Table 2). While breadth has a larger positive effect than depth, its squared effect is more negative. Thus, the effect of breadth rises much faster but also falls much faster than depth, indicating that the centrality of a location is more sensitive to breadth than depth. Model 6 reports the incident rate ratios (IRRs) for Model 5. It represents the percentage change in the dependent variable produced by a unit change in the independent variable. It can be seen that one unit change in breadth produces 278% (377.56-100) increase and its squared term produces 43% (56.69-100) decrease in the centrality. Whereas a unit increase in depth produces 8% increase and its squared term produces about 0.3% decrease in the centrality.3 To understand the combined (i.e., quadratic) effects of The breadth and depth coefficients, we look at the predictive margins, plotted in Figure 2. The predictions are calculated for a range of standardized breadth and depth values with other variables at their observed values. The range is selected keeping in mind the distributions of breadth and depth while at the same time keeping the two graphs comparable. It includes 90.89% of the standardized breadth values and 99.26% of the standardized depth values. As shown in the figure, the effect of breadth rises much faster and peaks much before depth. In our dataset that contains the entire population of patenting locations in the industry, the effect of depth is uniformly positive, while the effect of breadth is an inverted U-shape. Further, for most of the range, breadth produces a larger effect than depth, except at the extremities where the confidence intervals overlap. Therefore, while theoretically depth may have a higher effect than breadth at certain extreme values, those values are not common. We therefore conclude that, in general, breadth produces a larger effect than depth in this emerging industry, supporting H4. Figure 2 View largeDownload slide Effect of breadth versus depth. Figure 2 View largeDownload slide Effect of breadth versus depth. 5.3. Established versus emergent locations In order to test hypotheses H5a and H5b that study relative accumulations of breadth and depth in established and emergent locations, we perform two tests, shown in Table 3. We first estimate random effects models with breadth and depth as functions of pre-entry patent output. We control for the developed economy locations, time of entry into the sample and the top technology classes in terms of the patent output. We do not expect any other control variables to have an effect on breadth and omit them from this analysis. The results are shown in Models (1) and (2) in Table 3. The dummy variable for developed countries produces a positive significant effect on the breadth but insignificant on the depth. Additionally, an increase in pre-entry patent output is negatively associated with breadth and positively associated with depth. Thus, more established innovative locations with higher pre-entry patent output increase depth of patented innovation in the wind power industry relative to the emergent locations. Emergent locations on the other hand increase breadth of patented innovation in the wind industry relative to the more established locations. Table 3 Breadth, depth and entering locations Variables  (1) Dependent variable: standardized breadth  (2) Dependent variable: standardized depth  (3) Hazard of depth patents        Hazard ratios  Pre-entry innovation output  −0.2198**  0.3503***  1.4410***    (0.1047)  (0.0872)  (0.1185)  Developed countries  0.0946***  0.0369  1.0730    (0.0261)  (0.0406)  (0.2678)  Time since wind entry  −0.0043**  −0.0090***  0.9732***    (0.0012)  (0.0031)  (0.0056)  Class 240  0.9569***  0.2501***      (0.0556)  (0.0561)    Class 320  0.6150***  1.4904***      (0.1061)  (0.3998)    Class 415  0.4849***  0.4713***      (0.0662)  (0.1339)    Class 416  0.9350***  0.2576***      (0.0537)  (0.0537)    Top classes      3.2236***        (0.6597)  Constant  −0.5152***  −0.1726***      (0.0251)  (0.0330)    Observations  22,035  22,035  1358  Overall R2  0.5802  0.1970    Number of locations  565  565  565  Mean VIF  1.37  1.37    Proportional hazard test        Chi-squared, p-value      3.96, 0.4112  Variables  (1) Dependent variable: standardized breadth  (2) Dependent variable: standardized depth  (3) Hazard of depth patents        Hazard ratios  Pre-entry innovation output  −0.2198**  0.3503***  1.4410***    (0.1047)  (0.0872)  (0.1185)  Developed countries  0.0946***  0.0369  1.0730    (0.0261)  (0.0406)  (0.2678)  Time since wind entry  −0.0043**  −0.0090***  0.9732***    (0.0012)  (0.0031)  (0.0056)  Class 240  0.9569***  0.2501***      (0.0556)  (0.0561)    Class 320  0.6150***  1.4904***      (0.1061)  (0.3998)    Class 415  0.4849***  0.4713***      (0.0662)  (0.1339)    Class 416  0.9350***  0.2576***      (0.0537)  (0.0537)    Top classes      3.2236***        (0.6597)  Constant  −0.5152***  −0.1726***      (0.0251)  (0.0330)    Observations  22,035  22,035  1358  Overall R2  0.5802  0.1970    Number of locations  565  565  565  Mean VIF  1.37  1.37    Proportional hazard test        Chi-squared, p-value      3.96, 0.4112  Notes: Models (1) and (2): random effects regression (standard errors clustered on locations). Model (3): Cox proportional hazard model (multiple events model using time to subsequent events from entry, Efron method of ties, robust standard errors). Standard errors in parentheses; ***p < 0.01, **p < 0.05, *p < 0.1. Next, we test a location’s likelihood of filing depth-building patents (or depth patents) using the time-to-event or hazard rate model. The nuances in the model detailed below helps to capture the richness of the data and to understand locations’ patenting behavior over time. The dependent variable is the event of filing a depth patent. It is modeled as a hazard function which is the probability that a location files a depth patent at time t conditional on not filing a depth patent up to t. We estimate the effect of pre-entry patent output on depth patenting hazard using a Cox proportional hazard model. We assume that locations enter the risk set for this estimation after filing their first patent, and then measure time to subsequent depth patenting from entry. Further, we use a conditional risk set model so that a location is at risk of the kth depth patenting event only after encountering (k−1)th depth patenting event. Results are reported in Model (3) in Table 3. To maintain the Cox model’s assumption of proportional hazards, we replace separate dummy variables for the top classes in terms of patent output using a single dummy variable representing all four classes (Table 3 reports the result of the proportional hazard test showing that the null hypothesis of proportionality cannot be rejected). As seen in Model (3), a unit increase in the pre-entry patent output increases the hazard of depth patenting by 43%. Thus, established locations have a higher likelihood of continuing with depth patenting throughout the sample period than emergent locations. Combining the results of models (1), (2) and (3) in Table 3, we find emergent locations to have increases in breadth of patented innovation and lower likelihoods of depth patenting in the wind power industry and when compared with established locations, thereby providing support for Hypotheses H5a and H5b. 5.4. Additional robustness tests We tested our results with alternative model specifications and estimation methodologies to check their robustness. We tested for potential endogeneity arising from our cumulative data structure using additional time lags of 3 and 5 years on the independent variables. The goal of these models was to test for reverse causality as a source of endogeneity as well as omitted variables simultaneously affecting the network-based dependent and independent variables. We further identified an instrument to test for omitted variable bias as a cause of endogeneity. In addition, we tested the effect of outlier locations driving the hypothesized results by excluding the outliers. Further, we tested for firm effects by including dummy variables for the top firms to check if the top firms might be driving the results. Finally, we attempted to measure government support for wind power by including Renewable Energy Country Attractiveness Index developed by Ernst & Young. As argued earlier, the policy instruments have largely influenced the market activity of the wind power industry. These policies, in the long run, may affect the innovation activity at a location. Our results persisted in all the robustness tests (details in the Online Appendix). 6. Discussion, implications and limitations The global dispersion of technology and the growing geographic spread of value creation define today’s knowledge economy (Mudambi, 2008; Bathelt and Cohendet, 2014). Firms are increasingly coordinating their knowledge creation activities across geographic space, especially in knowledge-intensive industries (Lorenzen, 2005). While existing innovative locations are becoming more globally connected (Bathelt et al., 2004; Lorenzen and Mudambi, 2013), new locations are increasingly entering these industries’ global innovation networks (Mudambi, 2008; Kumaraswamy et al., 2012). In this paper we map an industry’s technological network in which the various patented technologies of the industry are connected to the respective inventor locations. We argue that such a network shows the locations’ relative positions based on their patenting activity. The locations with high patent output are more central. The industry’s evolution makes this technological network dynamic with location centralities undergoing major changes. We find that the features of a location’s existing patent stock significantly influence its centrality. The results of our analyses show that the breadth of patented innovation at a location has a larger impact on its centrality in the technological network, than the depth of patented innovation. Next, we show that, after entering the emerging industry, the established innovative locations are more likely to increase depth of patented innovation in the industry’s technologies than breadth of patented innovation. On the other hand, when the emergent locations enter this industry, they are more likely to increase breadth of patented innovation in the industry’s technologies rather than depth of patented innovation. The argument is about the relative difference in focus. In absolute terms, the established locations are likely to have more breadth and depth. We therefore argue that, given the higher effect of breadth on centrality, emergent locations’ breadth focus in the focal industry, in comparison to established locations, helps them to rapidly achieve high centrality scores. Innovation networks evolve over time so the metrics associated with centrality constantly rise (Awate et al., 2015). Thus, our results do not imply that all the emergent locations eventually overtake established locations. As for the established locations, we would again like to stress the comparative nature of the arguments. We use pre-entry patent output that is a continuous measurement scale that identifies established and emergent locations. Therefore, our results imply that the more established a location is, the more likely it is to focus on depth than breadth for innovating in the new industry. This however does not imply that the centralities of established locations are lower on average. With their mix of depth and breadth, many of these locations may still be highly central in the industry’s global innovation network. Our results however do imply that, due to their strong focus on the existing technologies, their marginal rate of accession to centrality in the global innovation network is lower. Emergent locations on the other hand may face lower barriers to exploring new technologies and therefore have a higher marginal rate of accession to such centrality. Our evidence demonstrates the successful entry and survival of emergent locations in the industry’s global patented innovation network. This may suggest that the entry barriers for the locations entering the global innovation network of emerging industries may be lower. It may be indicative of a much wider geographic dispersal of emerging industry innovations. This claim however needs direct empirical tests, which can be pursued in future research. We would like to emphasize that emergent locations do not specifically refer to ‘emerging economy’ locations. After controlling for developed versus developing countries, the emergent innovative locations include all locations with lower total patent output prior to entering the emerging industry. In fact, our rich longitudinal dataset captures the pre-entry experiences of many of the locations that are today’s leaders in patent output. These are largely advanced economy locations. The initial pattern of dispersal predicted by Storper and Walker’s model of geographic industrialization would be supported by these emergent locations.4 Our study, with considerable empirical rigor, provides strong support to this pattern of global geographic dispersal of emerging industries, particularly in the context of innovation. While the innovative locations in our dataset are largely advanced economy locations, we do see some middle-income and low-income locations such as those in China, India, Southeast Asia and the Middle East. These newer locations do not yet have as heavy a concentration of inventors as in the advanced economies; however, the very presence of these locations with significant inventor populations is indicative of the catch-up processes under way. Two examples of this rapid catch-up in patented innovations, also highlighted in Figure 1, are Bangalore and Shanghai. These locations filed their first wind turbine patents in 2004 and 2006, respectively. However, by 2011, they are highly innovative with the average degree centralities in the 95th and 99th percentiles, respectively. Bangalore is known as a major global ICT cluster (Lorenzen and Mudambi, 2013), while Shanghai is rapidly becoming one of the world’s major manufacturing hubs (Wu and Radbone, 2005). Our analyses show that, with respect to the wind power industry, Bangalore is also active in electrical technology research. Bangalore’s ICT capabilities and established high-technology infrastructure have attracted R&D in other related industries. It has also attracted some of the world’s largest MNEs to set up major R&D subsidiaries. For example, GE’s John Welch Technology Center in Bangalore is the company’s largest R&D center outside of the USA. As shown in Figure 1, Bangalore connects to not just the core classes but also a number of non-core classes. On the contrary, San Francisco and Aalborg, which were the cradles of the industry and are equally innovative, are seen to connect primarily to the core classes. The appearance of middle-income and low-income locations in the global innovation networks of many industries is a feature of today’s knowledge economy. A number of such locations may currently be involved in standardized activities. However, over time, through learning and innovation catch-up, they may attract increasingly complex activities (Mudambi, 2008; Awate et al., 2015) and enter global innovation networks. We point out that the early strategy of such new locations should focus more on increasing breadth than depth. We use a novel methodology in which we model the industry technological network by mapping locations’ patented technological connections weighted by its inventor population. Rather than using overall patents (e.g., patent counts) as a broad measure of innovation, we focus on more nuanced measure that incorporates who invents where and what. This approach is far more information-rich than the traditional approach that uses overall patents. This is because one new patent from a location may actually correspond to the addition of more than one new inventor and/or more than one new technology class, thereby improving the measurement of the patenting activity at a location. The innovations studied are patented technological innovations. Although predominantly associated with industrialization, these are only one form of innovation. Patents are only one component of overall innovation activity, and do not explicitly capture the incremental continuous processes related to learning by doing and learning by interaction. Further, apart from patents, innovations are materialized as copyrights, trademarks, new products and/or their features. In addition to these publicized forms, innovations may be undisclosed and held within the firm. Therefore, while a location’s patent-based centrality is one of the important factors associated with regional development, it is by no means the only one. Another limitation of the study is a more direct empirical proxy for governmental support to the emerging industry as it may encourage a location’s innovation activity. While we use Ernst & Young RECAI index and its sub-index for wind power, which are internally consistent, a more direct measure such as financial support by the governments or the number of targeted new regulations would be useful to test the effect. We end by noting an important caveat with regard to emergent locations, particularly for those in emerging economies. In order to enter the technological networks of emerging industries, these locations must demonstrate the availability of basic knowledge resources like skilled labor, a population of entrepreneurial local firms in related industries and supporting business services. MNE subsidiaries are often the spark that jump-starts local innovation through knowledge spillovers as well as through providing connectivity to the global innovation network. Today these MNEs have a wealth of choices in terms of R&D location, so there is an element of chance in whether a potential location actually attracts enough MNE R&D investment to begin a virtuous cycle of local innovation. However, in the words of Louis Pasteur, ‘Chance favors the prepared mind.’ Funding This research was partially funded by the Indian School of Business – Ernst & Young Initiative for Emerging Market Studies. Supplementary material Supplementary data for this paper are available at Journal of Economic Geography online. Footnotes 1 The raw data totaled to 3652 inventor names on which we ran several rounds of disambiguation resulting in 2204 unique inventors. Hence, we are quite confident that we have the real inventor in overwhelming majority of the cases. This leads to 1.7 appearances of the same inventor name in the raw data, with a minimum of 1 entry to a maximum of 20 entries per inventor. 2 The analysis also allows for the movement of inventors (Saxenian, 1994). A new patent following an inventor’s move adds either completely new location or technology class nodes, or new ties between existing location and technology class nodes. 3 It should be noted that the interpretation of the comparative magnitudes of the breadth and depth effects are possible only after variable standardization. The comparative interpretation holds as both breadth and depth variables are standardized to have zero mean and unit variance. 4 Storper and Walker’s explanation of the shift of an industry is essentially one of regional institutions, while our focus is on patented technological innovations. References Ahuja G. ( 2000) Collaboration networks, structural holes, and innovation: a longitudinal study. Administrative Science Quarterly , 45: 425– 455. Google Scholar CrossRef Search ADS   Alcácer J. ( 2006) Location choices across the value chain: how activity and capability influence collocation. Management Science , 52: 1457– 1471. Google Scholar CrossRef Search ADS   Almeida P. ( 1996) Knowledge sourcing by foreign multinationals: patent citation analysis in the U.S. semiconductor industry. Strategic Management Journal , 17: 155– 165. Google Scholar CrossRef Search ADS   Almeida P., Kogut B. ( 1999) Localization of knowledge and the mobility of engineers in regional networks. Management Science , 45: 905– 917. Google Scholar CrossRef Search ADS   Arrow K. ( 1962) The economic implications of learning by doing. Review of Economic Studies , 29: 155– 173. Google Scholar CrossRef Search ADS   Awate S., Larsen M. M., Mudambi R. ( 2012) EMNE catch-up strategies in the wind turbine industry: is there a trade-off between output and innovation capabilities? Global Strategy Journal , 2: 205– 223. Google Scholar CrossRef Search ADS   Awate S., Larsen M. M., Mudambi R. ( 2015) Accessing vs sourcing knowledge: a comparative study of RD internationalization between emerging and advanced economy firms. Journal of International Business Studies  46: 63– 86. Google Scholar CrossRef Search ADS   Banker R., Cao Z., Menon N. M., Mudambi R. ( 2013) The Red Queen in action: the longitudinal effects of capital investments in the mobile telecommunications sector. Industrial and Corporate Change , 22: 1195– 1228. Google Scholar CrossRef Search ADS   Bathelt H., Boggs J. S. ( 2003) Towards a reconceptualization of regional development paths: is Leipzig’s media cluster a continuation of or a rupture with the past? Economic Geography , 79: 265– 293. Google Scholar CrossRef Search ADS   Bathelt H., Cohendet P. ( 2014) The creation of knowledge: local building, global accessing and economic development—toward an agenda. Journal of Economic Geography , 14: 869– 882. Google Scholar CrossRef Search ADS   Bathelt H., Malmberg A., Maskell P. ( 2004) Clusters and knowledge: local buzz, global pipelines and the process of knowledge creation. Progress in Human Geography , 28: 31– 56. Google Scholar CrossRef Search ADS   Bel G., Fageda X. ( 2008) Getting there fast: globalization, intercontinental flights and location of headquarters. Journal of Economic Geography , 8: 471– 495. Google Scholar CrossRef Search ADS   Bell G. G., Zaheer A. ( 2007) Geography, networks, and knowledge flow. Organization Science , 18: 955– 972. Google Scholar CrossRef Search ADS   Beugelsdijk S., Mudambi R. ( 2013) MNEs as border-crossing multi-location enterprises: the role of discontinuities in geographic space. Journal of International Business Studies , 44: 413– 426. Google Scholar CrossRef Search ADS   Bierly P., Chakrabarti A. ( 1996) Knowledge strategies in the U.S. pharmaceutical industry. Strategic Management Journal , Winter Special Issue, 17: 123– 135. Google Scholar CrossRef Search ADS   Borgatti S., Everett M., Freeman L. ( 2002) UCINET 6 for Windows: Software for Social Network Analysis . Harvard: Analytic Technologies. Boschma R., Frenken K. ( 2011) The emerging empirics of evolutionary economic geography. Journal of Economic Geography , 11: 295– 307. Google Scholar CrossRef Search ADS   Bresnahan T., Gambardella A., Saxenian A. ( 2001) “Old economy” inputs for “new economy” outcomes: cluster formation in the New Silicon Valleys. Industrial and Corporate Change , 10: 835– 860. Google Scholar CrossRef Search ADS   Buen J. ( 2006) Danish and Norwegian wind industry: the relationship between policy instruments, innovation and diffusion. Energy Policy , 34: 3887– 3897. Google Scholar CrossRef Search ADS   Cantwell J. A., Mudambi R. ( 2011) Physical attraction and the geography of knowledge sourcing in multinational enterprises. Global Strategy Journal , 1: 206– 232. Google Scholar CrossRef Search ADS   Cantwell J., Mudambi R. ( 2005) MNE competence-creating subsidiary mandates. Strategic Management Journal , 26: 1109– 1128. Google Scholar CrossRef Search ADS   Cardwell D. ( 2013) Renewed tax credit buoys wind-power projects. The New York Times, 21 March. Christensen C. ( 1993) The rigid disk drive industry: a history of commercial and technological turbulence, Business History Review , 67: 531– 588. Google Scholar CrossRef Search ADS   Christensen C. M., Bower J. L. ( 1996) Customer power, strategic investment, and the failure of leading firms. Strategic Management Journal , 17: 197– 218. Google Scholar CrossRef Search ADS   Coe N. M., Bunnell T. ( 2003) “Spatializing” knowledge communities: towards a conceptualization of transnational innovation networks. Global Networks , 3: 437– 456. Google Scholar CrossRef Search ADS   Coe N. M., Dicken P., Hess M. ( 2008) Global production networks: realizing the potential. Journal of Economic Geography , 8: 271– 295. Google Scholar CrossRef Search ADS   Coe N. M., Hess M., Yeung H. W. C., Dicken P., Henderson J. ( 2004) ‘Globalizing’ regional development: a global production networks perspective. Transactions of the Institute of British Geographers , 29: 468– 484. Google Scholar CrossRef Search ADS   Cohen W., Levinthal D. ( 1990) Absorptive capacity: a new perspective on learning and innovation. Administrative Science Quarterly , 35: 128– 152. Google Scholar CrossRef Search ADS   Cyert R., March J. ( 1963) A Behavioral Theory of the Firm . Prentice-Hall, NJ: Englewood Cliffs. Econstats. ( 2012) International Internet bandwidth. Econstats.com, 25 October. Faust K. ( 1997) Centrality in affiliation networks. Social Networks , 19: 157– 191. Google Scholar CrossRef Search ADS   Feldman M., Lendel I. ( 2010) Under the lens: the geography of optical science as an emerging industry. Economic Geography , 86: 147– 171. Google Scholar CrossRef Search ADS   Florida R., Kenney M. ( 1990) Silicon Valley and Route 128 won’t save us. California Management Review , 33: 68– 85. Google Scholar CrossRef Search ADS   Funk R. J. ( 2014). Making the most of where you are: geography, networks, and innovation in organizations. Academy of Management Journal , 57: 193– 222. Google Scholar CrossRef Search ADS   Garud R., Karnøe P. ( 2003) Bricolage versus breakthrough: distributed and embedded agency in technology entrepreneurship Research Policy , 32: 277– 300. Google Scholar CrossRef Search ADS   Gates B. ( 2011) Taking energy storage to a higher level. The Gates Notes, 16 November. Grant R. ( 1996) Prospering in dynamically competitive environments: organizational capability as knowledge integration. Organization Science , 7: 375– 386. Google Scholar CrossRef Search ADS   Griliches Z. ( 1979) Issues in assessing the contribution of research and development to productivity growth, Bell Journal of Economics , 10: 92– 116. Google Scholar CrossRef Search ADS   Griliches Z. ( 1990) Patent statistics as economic indicators: a survey, Journal of Economic Literature , 28: 1661– 1707. GWEC. ( 2012) Global wind report: annual market update 2011. Global Wind Energy Council, 24 August. Hagedoorn J., Cloodt M. ( 2003) Measuring innovative performance: is there an advantage in using multiple indicators? Research Policy , 32: 1365– 1379. Google Scholar CrossRef Search ADS   Hamel G., Prahalad C. K. ( 1994) Competing for the Future . Boston: Harvard Business School Press. Hannigan T. J., Cano-Kollmann M., Mudambi R. ( 2015) Thriving innovation amidst manufacturing decline: the Detroit auto cluster and the resilience of local knowledge production. Industrial and Corporate Change , 24: 613– 634. Google Scholar CrossRef Search ADS   Henderson J., Kuncoro A., Turner M. ( 1995) Industrial development in cities. Journal of Political Economy , 103: 1067– 1085. Google Scholar CrossRef Search ADS   Hsiao C. ( 1986) Analysis of Panel Data . Cambridge, UK: Cambridge University Press. ITU. ( 2005) International Internet statistics. International Telecommunications Union, 25 October. Jacobs J. ( 1969) The Economy of Cities . New York, NY: Random House. Jaffe A. ( 1989) Real effect of academic research, American Economic Review , 79: 957– 970. Jaffe A. B. ( 1986) Technological opportunity and spillovers of R&D: evidence from firms’ patents, profits, and market value. American Economic Review , 76: 984– 1001. Jaffe A., Trajtenberg M., Henderson R. ( 1993) Geographic localization of knowledge spillovers as evidenced by patent citations, Quarterly Journal of Economics , 108: 577– 598. Google Scholar CrossRef Search ADS   Katila R., Ahuja G. ( 2002) Something old, something new: a longitudinal study of search behavior and new product introduction. Academy of Management Journal , 45: 1183– 1194. Google Scholar CrossRef Search ADS   Kher U. ( 2010) A call for collaboration, Nature , 466: S21– S22. Google Scholar CrossRef Search ADS   Klepper S. ( 1996) Entry, exit, growth, and innovation over the product life cycle. American Economic Review , 86: 562– 583. Kodama F. ( 1992) Technology fusion and the new R&D. Harvard Business Review , 70: 70– 78. Kogut B., Zander U. ( 1992) Knowledge of the firm, combinative capabilities, and the replication of technology. Organization Science , 3: 383– 397. Google Scholar CrossRef Search ADS   Kuhn T. ( 1962) The Structure of Scientific Revolutions . Chicago, IL: University of Chicago Press. Kumaraswamy A., Mudambi R., Saranga H., Tripathy A. ( 2012) Catch-up strategies in the Indian auto component industry: domestic firms’ responses to market liberalization. Journal of International Business Studies , 43: 368– 395. Google Scholar CrossRef Search ADS   Lahiri N. ( 2010). Geographic distribution of R&D activity: how does it affect innovation quality? Academy of Management Journal , 53: 1194– 1209. Google Scholar CrossRef Search ADS   Leonard-Barton D. ( 1995) Wellsprings of Knowledge . Boston: Harvard Business School Press. Levinthal D., March J. ( 1981) A model of adaptive organizational search. Journal of Economic Behavior and Organization , 2: 307– 333. Google Scholar CrossRef Search ADS   Lorenzen M. ( 2005) Introduction: knowledge and geography. Industry and Innovation , 12: 399– 407. Google Scholar CrossRef Search ADS   Lorenzen M., Mudambi R. ( 2013) Clusters, connectivity and catch-up: Bollywood and Bangalore in the global economy. Journal of Economic Geography , 13: 501– 534. Google Scholar CrossRef Search ADS   March J. ( 1991) Exploration and exploitation in organizational learning, Organization Science , 2: 71– 87. Google Scholar CrossRef Search ADS   Markard J, Petersen R. 2009. The offshore trend: structural changes in the wind power sector. Energy Policy , 37: 3545– 3556. Google Scholar CrossRef Search ADS   Marshall A. ( 1890) Principles of Economics . London, UK: Macmillan. Martin R., Sunley P. ( 2007) Complexity thinking and evolutionary economic geography. Journal of Economic Geography , 7: 573– 601. Google Scholar CrossRef Search ADS   Martin R., Sunley P. ( 2011) Conceptualizing cluster evolution: beyond the life cycle model? Regional Studies , 45: 1299– 1318. Google Scholar CrossRef Search ADS   Maskell P., Malmberg A. ( 2007) Myopia, knowledge development and cluster evolution. Journal of Economic Geography , 7: 603– 618. Google Scholar CrossRef Search ADS   Mudambi R. ( 2008) Location, control and innovation in knowledge-intensive industries. Journal of Economic Geography , 8: 699– 725. Google Scholar CrossRef Search ADS   Mudambi R., Hannigan T., Kline W. ( 2012) Advancing science on the knife’s edge: integration and specialization in management Ph.D. programs. Academy of Management Perspectives , 26: 83– 105. Google Scholar CrossRef Search ADS   Mudambi R., Santangelo G. D. ( 2016) From shallow resource pools to emerging clusters: the role of multinational enterprise subsidiaries in peripheral areas. Regional Studies , 50: 1965– 1979. Google Scholar CrossRef Search ADS   Mudambi R., Swift T. ( 2014) Knowing when to leap: transitioning between exploitative and explorative R&D. Strategic Management Journal , 35: 126– 145. Google Scholar CrossRef Search ADS   Musgrove P. ( 2010) Wind Power . Cambridge, UK: Cambridge University Press. Narin F., Noma E., Perry R. ( 1987) Patents as indicators of corporate technological strength. Research Policy , 16: 143– 155. Google Scholar CrossRef Search ADS   Narula R. ( 2014) Exploring the paradox of competence-creating subsidiaries: balancing bandwidth and dispersion in MNEs. Long Range Planning , 47, 4– 15. Google Scholar CrossRef Search ADS   Neffke F., Svensson Henning M., Boschma R., Lundquist K., Olander L. ( 2011) The dynamics of agglomeration externalities along the life cycle of industries. Regional Studies , 45: 49– 65. Google Scholar CrossRef Search ADS   Nelson R., Winter S. ( 1982) An Evolutionary Theory of Economic Change . Cambridge: Harvard University Press. Nielsen K. 2010. Technological trajectories in the making: two cases from the contemporary history of wind power. Centaurus , 52: 175– 205. Google Scholar CrossRef Search ADS   Paruchuri S., Awate S. ( 2016) Organizational knowledge networks and local search: the role of intra‐organizational inventor networks. Strategic Management Journal , 38: 657– 675. Google Scholar CrossRef Search ADS   Patel P., Pavitt K. ( 1997) The technological competencies of the world’s largest firms: complex and path-dependent, but not much variety. Research Policy , 26: 141– 156. Google Scholar CrossRef Search ADS   Pedersen T, Larsen M. ( 2009) Vestas Wind Systems A/S—Exploiting Global RD Synergies. Case: 9B09M079, 26 November, pp. 1–17. London, UK/Canada: Ivey Management Services. Penrose E. G. ( 1959) The Theory of the Growth of the Firm . New York: Wiley. Podolny J. M., Stuart T. E. ( 1995) A role-based ecology of technological change. American Journal of Sociology , 100: 1224– 1260. Google Scholar CrossRef Search ADS   Porter M. ( 1980) Competitive Strategy . New York (NY): The Free Press. Porter M. E., Stern S. ( 2000) Measuring the “Ideas” Production Function: Evidence from International Patent Output. Working Paper No. w7891, National Bureau of Economic Research. Pouder R., St. John C. H. ( 1996) Hot spots and blind spots: geographical clusters of firms and innovation. Academy of Management Review , 21: 1192– 1225. Rajsekhar B., Van Hulle F., Jansen J. C. ( 1999) Indian wind energy programme: performance and future directions. Energy Policy , 27: 669– 678. Google Scholar CrossRef Search ADS   Romer P. ( 1986) Increasing returns and long-run growth. Journal of Political Economy , 94: 1002– 1037. Google Scholar CrossRef Search ADS   Rosenkopf L., Nerkar A. ( 2001) Beyond local search: boundary-spanning, exploration, and impact in the optical disk industry. Strategic Management Journal , 22: 287– 306. Google Scholar CrossRef Search ADS   Saxenian A. ( 1994) Regional Advantage: Culture and Competition in Silicon Valley and Route 128 . Cambridge (MA): Harvard University Press. Scalera, V., Perri, A., Hannigan, T. J. (2018). Knowledge connectedness within and across home country borders: spatial heterogeneity and the technological scope of firm innovations. Journal of International Business Studies, forthcoming. Schumpeter J. A. ( 1934) The Theory of Economic Development: An Inquiry into Profits, Capital, Credit, Interest, and the Business Cycle . Cambridge (MA): Harvard University Press. Soda G., Zaheer A. ( 2012) A network perspective on organizational architecture: performance effects of the interplay of formal and informal organization. Strategic Management Journal , 33: 751– 771. Google Scholar CrossRef Search ADS   Storper M. ( 1995) The resurgence of regional economies, ten years later: the region as a nexus of untraded interdependencies. European Urban and Regional Studies , 2: 191– 221. Google Scholar CrossRef Search ADS   Storper M., Walker R. ( 1989) The Capitalist Imperative: Territory, Technology and Industrial Growth . New York (NY): Basil Blackwell Inc. Sturgeon T., Van Biesebroeck J., Gereffi G. ( 2008) Value chains, networks and clusters: reframing the global automotive industry. Journal of Economic Geography , 8: 297– 321. Google Scholar CrossRef Search ADS   Teece D. J. ( 1994) Information sharing, innovation, and antitrust. Antitrust Law Journal , 62: 465– 481. Tripsas M. ( 2009) Technology, identity and inertia through the lens of the ‘Digital Photography Company’. Organization Science , 20: 441– 460. Google Scholar CrossRef Search ADS   Turkina, E., Van Assche, A. Kali, R. (2016). Structure and evolution in global cluster networks: Evidence from the aerospace industry. Journal of Economic Geography, 16: 1211–1234. Tushman M. L., Murmann P. ( 1998). Dominant designs, innovation types and organizational outcomes, Research in Organizational Behavior , 20: 231– 266. UCLA. ( 2013) Stata FAQ: how can I perform the likelihood ratio, Wald, and Lagrange multiplier (score) test in Stata? UCLA: Statistical Consulting Group, 25 October. Van der Panne G. ( 2004) Agglomeration externalities: Marshall versus Jacobs. Journal of Evolutionary Economics , 14: 593– 604. Google Scholar CrossRef Search ADS   Wu J., Radbone I. ( 2005) Global integration and intra-urban determinants of foreign direct investment in Shanghai. Cities , 22: 275– 286. Google Scholar CrossRef Search ADS   WWEA. ( 2012) World Wind Energy Report 2011. World Wind Energy Association, July 1. Yeung H., Coe N. ( 2015) Toward a dynamic theory of global production networks. Economic Geography , 91: 29– 58. Google Scholar CrossRef Search ADS   Zhu S., He C., Zhou Y. ( 2017) How to jump further and catch up? Path-breaking in an uneven industry space. Journal of Economic Geography , lbw047, 1– 25. © The Author (2017). Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Economic Geography Oxford University Press

On the geography of emerging industry technological networks: the breadth and depth of patented innovations

Loading next page...
 
/lp/ou_press/on-the-geography-of-emerging-industry-technological-networks-the-fjMPgKORes
Publisher
Oxford University Press
Copyright
© The Author (2017). Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com
ISSN
1468-2702
eISSN
1468-2710
D.O.I.
10.1093/jeg/lbx032
Publisher site
See Article on Publisher Site

Abstract

Abstract We study an emerging industry’s technological network in which the patented technologies of the industry are connected to the geographic locations of the inventors. Our research context is the global wind turbine industry. The network maps the geography of the industry’s patented technologies over time. It shows the locations’ patenting activities in different technologies, centered around the core classes of electricity and aerodynamics. These activities shape the locations’ positions in the global patent network and indicate their relative importance to the industry’s innovations. We note that the network and thus the relative positions of the locations within it are constantly changing. We examine how the existing patented knowledge stock at a location affects its position in this technological network. Further, we analyze how locations that enter the global technology network attain centrality. 1. Introduction Regional development is a complex phenomenon involving a variety of actors and linkages (Coe et al., 2008). While industries influence regional development (Storper and Walker, 1989), global production networks and regional relational networks interact (Coe et al., 2004; Yeung and Coe, 2015) feeding into the process of regional growth. Growth at the level of firm, region or nation does not always imply rising innovation activity there (Awate et al., 2012). However, a high level of innovation activity positively correlates with growth. Further, going beyond the absolute levels, a rising trend of innovation activity indicates the growth focus of the relevant entity. In this paper, we study the innovation-led growth of locations. Innovation activity at the level of a location encapsulates the activities of micro entities such as firms, and within them, of people. The aggregation of the innovation activities of these entities defines the innovativeness of a location. The technologies that the local firms and their scientists and engineers work on form the technological signature of that location. As these actors search for knowledge, their search spans technological as well as geographic dimensions, which are non-orthogonal. In fact, a location’s technological linkages with other locations are exploited by firms and their inventors in their knowledge search strategies to generate innovations (Bell and Zaheer, 2007; Funk, 2014; Hannigan et al., 2015; Paruchuri and Awate, 2016). The technological network connections of a location thus facilitate its coupling to the industry’s innovation network (Hannigan et al., 2015) and the global production network, thereby driving its growth (Coe et al., 2004). An industry’s technological network of locations maps the technologies that a certain location directly connects to (i.e., works in) and the strength of those connections. Further, in its entirety, the network models the core and peripheral technologies associated with the industry, the relative position of the locations in this technological network and avenues for their nonlocal technological searches. A central position of a location in such a network is the outcome of a higher amount of innovation activity in a number of technologies relative to other locations in the network, commonly measured by degree centrality (Turkina et al., 2016). However, as the literatures on economic geography, strategy, technology management and sociology note, such a network is far from constant. Be it the ever-changing geography of innovation (Storper and Walker, 1989), the Red Queen effect in firms’ technological competition (Banker et al., 2013), industry lifecycles (Klepper, 1996) or the ecology of technological change (Podolny and Stuart, 1995), all point to the dynamics of technology-based networks. Thus, a central innovative location with a strong technological coupling to the industry’s innovation network is likely to experience its coupling weaken over time if it does not reinforce it through sustained innovation activity. Over time, it may find itself relegated to a peripheral position in relation to other locations. How might a location and the micro entities present there keep up as well as catch up with others in the changing technological network? How might a location leverage its extant innovations to define its position in the network? We empirically explore these questions using patented technological innovations created at a location. A key element of a patent innovation network is the importance given to the individual inventor. Thus, our study addresses the ‘need for approaches that recognize the significance of innovative networks that extend beyond firms and, in particular, those associated with the movement of knowledgeable individuals’ (Coe and Bunnell, 2003, 437). We situate our study in an emerging/growing industry, as the technological dynamics in such industries are stronger than in mature industries. The emerging industry studied in this paper is the global wind power industry. The industry came into existence in the early 1980s after the leading industrial nations of the world suffered the oil crises of the 1970s. Since then, the industry has witnessed the establishment of a dominant design, steady output growth, shakeouts, changing policy regimes and technological discontinuities, all of which are characteristic features of emerging industries (Awate et al., 2012). We use patented innovation data of the wind power industry from the United States Patent and Trademark Office (USPTO) database. Our unit of analysis is a location, which aggregates inventor locations, noted on patents, within a 30-mile radius. Specifically, we construct measures of breadth and depth of patented innovation at each location. Breadth measures the number of active patent technology classes in a location and their importance to the industry. Depth measures the amount of patenting in the set of local technology classes. We find that breadth and depth of patented innovations produce curvilinear effects on a location’s future degree centrality in the industry's technological network. Further, we find that breadth produces a larger effect than depth suggesting that location’s centrality in an emerging-industry technological network is far more sensitive to breadth than depth. We also find that, among the locations entering the emerging industry, the established innovative locations are more likely to increase depth of patented innovation than breadth in the focal industry’s technologies. In contrast, the emergent innovative locations are more likely to increase breadth of patented innovation than depth. We argue that the emergent locations’ breadth focus helps them to rapidly improve their centrality in the technological network of locations. The remaining paper is organized as follows. We review the literature in economic geography and technology management to build our hypotheses in Section 2. We explain the development of the global wind power industry in Section 3, followed by details of the data and methodology in Section 4. We then present the estimations and the results in Section 5 and discuss the findings, implications and limitations in Section 6. 2. Geographic industrialization and innovation In the dominant model of geographic industrialization and innovation, Storper and Walker (1989) proposed that dynamic sectors are capable of generating their own resources thereby granting the leading firms ‘substantial freedom to develop where they are, or to locate where they please … innovators in a fast growing industry can ignore the traditional locational calculus’ (p. 75). Their model predicts four stages of geographic industrialization, namely localization, clustering, dispersal and the rise of new industrial centers. While that is the general pattern, the growth of a location in a particular industry is driven by the complex interactions of multiple entities, both local and global, that are embedded in a variety of networks (Coe et al., 2008). These networks contain different relations such as vertical buyer–supplier connections, horizontal competitive ties, institutional connections, interfirm and intrafirm linkages, personal diaspora connections as well as technological linkages (Turkina et al., 2016). These various networks are responsible for a region’s strategic coupling with global production networks thereby facilitating the region’s value creation, enhancement and value capture resulting in regional growth (Coe et al., 2004; Yeung and Coe, 2015). Given that our focus is on the innovation-led growth of locations, we study the technological linkages of locations. These develop as firms and inventors conduct technological searches and generate technological recombinations. While technological relatedness can only be judged ex post by an outsider to a focal innovation, the insider firm makes constant judgments about the underlying technologies and their combinative capabilities (Kogut and Zander, 1992). Thus, when the innovative activities of all the firms situated at a location are aggregated, one sees the realized industry-specific technological relatedness, which is discovered by the firms and the inventors situated at that location. Extending this exercise to include all of the industry-specific innovations reveals a technological network of locations for the industry, which shows the core and peripheral industry technologies and the locations’ connections to them. Going a step further, these connections can be weighted by the number of micro actors such as firms and their inventors situated at a location. A location that connects to a number of technologies, and connects strongly due to a large population of firms and inventors there, has high degree centrality and is very likely to be a technological hotspot of the industry (Pouder and St. John, 1996). Such a central location is tightly coupled to the industry’s global innovation network due to its high innovation output (Hannigan et al., 2015). Further, the network also shows how two locations are related by common technological knowledge. As the geographic scope of innovation widens in the modern knowledge economy, firms are increasingly expanding their knowledge networks to tap into distant knowledge sources (Cantwell and Mudambi, 2005). As firms and their inventors expand their technological search to include distant locations and develop extra-regional linkages, knowledge flows through a variety of conduits such as intrafirm, interfirm, institutional as well as personal ties (Soda and Zaheer, 2012; Lorenzen and Mudambi, 2013; Scalera et al., 2018), with the least common denominator being technological commonality among the locations. A central location in the network is thus likely to guide the nonlocal searches of a number of industry firms and inventors, thereby shaping the industry’s technological trajectory. However, as the industries evolve, the relative importance of the technologies changes. The firms change their location strategies and the locations experience a reorganization of their local assets (Martin and Sunley, 2011). Although manufacturing activities at a location may undergo drastic structural changes due to this reorganization, knowledge-related activities are sticky and a location’s innovation capabilities are more persistent (Sturgeon et al., 2008; Hannigan et al., 2015). In addition to the local knowledge stock, a location’s technological connections to other locations bring in new knowledge that finds innovative recombinations with the local knowledge, and assimilates with the local knowledge base. The combination of local and nonlocal knowledge reinforces a location’s innovation capabilities (Hannigan et al., 2015). We focus on the features of the patented knowledge stock of a location, specifically the breadth and depth of patented innovations. We then hypothesize their effects on the location’s future centrality in the industry’s technological network of locations. The participation of locations that are currently central in an industry’s global innovation network and those that are recent entrants, differ in systematic ways. A key contribution of our study is to develop theory and examine empirically the contrasting logics that drive these differences. Our analysis is based on the rich longitudinal data on technology classes, inventor locations and co-inventor linkages contained in the comprehensive population of wind turbine patents. We use it to build the associated global innovation network and analyze the logics that drive a location’s centrality within it. 2.1. Breadth and depth of patented innovation Patents are the manifestations of the underlying innovation activity. While innovative knowledge, including commercially valuable knowledge, exists in many forms, the most common form studied by the technology management literature consists of patents (Narin et al., 1987). Patents are considered as a proxy of firm innovation capabilities (Griliches, 1990; Jaffe et al., 1993) and have been shown to track very well with other measures of innovation like R&D expenditures and new product announcements (Hagedoorn and Cloodt, 2003). The process of generating patented innovations is often modeled using a knowledge production function (Griliches, 1979; Jaffe, 1989). An important input to this function is the existing knowledge stock. The knowledge production function highlights that new patented knowledge builds on the existing stock of patented knowledge (Porter and Stern, 2000). As early as the seminal work of Thomas Kuhn (1962), knowledge is shown to advance through two distinct processes of ‘exploration of new possibilities and exploitation of old certainties’ (March, 1991, 71). Exploration involves searching across disciplines and integrating diverse bodies of knowledge. In contrast, exploitation involves execution and refinement of existing knowledge, learning-by-doing (Arrow, 1962), and specialization. Integration and specialization are thus two distinct yet correlated processes (Mudambi et al., 2012). The integration process can be measured by the diversity of the knowledge building blocks that are recombined. In a like manner, the specialization process can be gauged in terms of the extent of specific and focused prior knowledge extant and used. In this sense, integration and specialization may be operationalized by the terms ‘breadth’ and ‘depth’, respectively. Breadth captures the number of discrete knowledge elements involved in creating the patented innovation as well as their importance to the focal industry. That is, it measures the technological scope of the patented innovation (Bierly and Chakrabarti, 1996; Katila and Ahuja, 2002; Lorenzen and Mudambi, 2013). Knowledge in more fields helps in identifying related knowledge elements that may exist in other technological areas. In addition to the raw number, the importance of these knowledge elements to the industry also factors into the definition of breadth. An industry’s core technologies when recombined with other technologies may find more direct application to the focal industry (Tushman and Murmann, 1998). Such exploratory recombinations are known to generate path-breaking innovations (Schumpeter, 1934; Penrose, 1959; Nelson and Winter, 1982). Depth, on the other hand, captures how well a certain technology is known. Depth develops as R&D activities continue in certain technologies. It often involves exploitative learning by repeating organizational routines (Levinthal and March, 1981). Over time, through learning-by-doing, deep knowledge of individual technologies and their interconnections is developed. For a firm, this deep technological acumen forms the basis of its core competencies making it a specialist in these technological areas (Hamel and Prahalad, 1994). For a location, the patented innovations are embodied in the firms and their inventors situated there (Almeida and Kogut, 1999). The literature on economic geography discusses such effective aggregation of the innovation activities of the micro entities at the local and regional level. For example, the prominent model of geographic industrialization by Storper and Walker (1989) assigns regional evolution to firms’ location strategies. Likewise, the literature on regional renewal and adaptive cycles describes how the micro entities shape location-level macro structures (Martin and Sunley, 2007, 2011) and how firms’ innovation activities change the regional innovation trajectory (Zhu et al., 2017). The literature on spatial knowledge creation places firms and their inventors at the center of the evolutionary dynamic. It extends the micro-level theories of individuals and firms to explain innovations at the aggregate level of a location (Maskell and Malmberg, 2007). The local technological milieu as well as the nonlocal pipelines of a location are indeed maintained and exploited by the micro actors (Bathelt et al., 2004; Lorenzen and Mudambi, 2013). As the knowledge elements of individual firms aggregate to form the local knowledge base, agglomeration effects and synergies increase the possible permutations and combinations of these knowledge elements that an individual firm can achieve. The aggregation of breadth and depth at the location level may therefore be more effective in generating innovative recombinations than firm-specific breadth and depth alone. As the micro actors conduct their technological searches in generating patented innovations, they use the breadth and depth of patented innovation already existing at a location. Locations that contain and support a large number of technologies, including those fundamental to the industry, exhibit a high breadth of patented innovation. The breadth of such locations may help firms identify exploratory innovative recombinations across technology fields. Such economies of technological scope may enable locations ‘to reap the intangible benefits of learning’ (p. 470) from many different high value-added local activities and facilitate a broad spectrum of production and entrepreneurial endeavors (Coe et al., 2004). On the other hand, the locations that contain and support a great concentration of certain technologies exhibit high depth in those technologies. It corresponds to the economies of scale achieved through ‘highly localized concentrations of specific knowledge, skills and expertise’ (p. 470) that attract firms to use the extant technology base at the location to identify niche and exploitative advances (Coe et al., 2004). However, a sole focus on either exploration (breadth) or exploitation (depth) is detrimental to innovation activity (March, 1991). Too much depth in certain technologies may lock in the current technological trajectory. As core competencies of the firms become core rigidities (Leonard-Barton, 1995), the location may experience an inward focus, groupthink and disconnect from key external stakeholders (Kher, 2010). It causes spatial myopia as the boundedly rational innovation actors select technological solutions that are closer to their existing knowledge base. Their search tends to be local not just technologically but also spatially (Maskell and Malmberg, 2007). With local labor primarily immobile, increase in the technological depth of a location and decrease in the variety may be natural consequences of the location’s evolution unless corrected by building global pipelines to distant knowledge sources and trans-local actors to access diverse knowledge (Coe et al., 2004; Lorenzen and Mudambi, 2013; Yeung and Coe, 2015; Turkina et al., 2016). Absent knowledge diversity, the local innovative specialization may continue beyond the point where buyers consider such advances to be valuable or worth paying for (Christensen 1993; Scalera et al., 2018). Further, returns to even the most valuable existing competencies eventually begin to decline (Mudambi and Swift, 2014). As the technological homogeneity in the local environment increases, firms in the technological hotspots find themselves less competitive than the non-hotspot firms (Pouder and St. John, 1996). Excessive breadth with comparatively less depth may also pose problems. A sole focus on integrating across various technological domains creates an excessive reliance on distant knowledge and increases the uncertainty of innovative outcomes (Cyert and March, 1963; Katila and Ahuja, 2002). The limits to the new knowledge, brought in by high breadth, are shown to exist in several contexts. For example, increases in a firm’s innovation breadth through structural hole spanning in its interfirm alliance network brings in new technological knowledge, but reduces interfirm collaboration and decreases its ultimate innovation performance (Ahuja, 2000). Likewise, a firm’s breadth in geographic space has been shown to have a negative curvilinear effect on innovation quality (Lahiri, 2010). Leveraging increases in innovation breadth also requires more cognitive and organizational bandwidth (Narula, 2014). Such bandwidth builds over time through intensive knowledge sharing between actors working in particular technologies. Without this, the search for recombinations solely based on breadth is often infeasible. Thus, while breadth and depth positively affect a location’s innovation activity, an excessive amount of either of the two has a negative effect. Further, they also have an interactive effect. A location with higher breadth has more high-breadth firms working in a variety of industry technologies. As a result, a higher number of distinct knowledge elements may be available locally. However, that may not translate into more recombinations unless these firms (or other local firms) have high depth in those technologies. The sheer mass of specialist knowledge in specific technological fields generates challenges for those who attempt to integrate it (Grant, 1996). If the location has predominantly low depth firms, they may find integrating these diverse knowledge elements beyond their cognitive capabilities (Cohen and Levinthal, 1990). Knowledge integration and specialization are thus correlated processes. The knowledge bases of the innovation leaders in most industries tend to be both broad and deep (Awate et al., 2012). We thus hypothesize the following: H1: A location’s breadth of patented innovation in terms of technology classes has a negative curvilinear effect on its degree centrality in the industry’s global technological network. H2: A location’s depth of patented innovation in the technology classes currently present has a negative curvilinear effect on its degree centrality in the industry’s global technological network. H3: A location’s breadth and depth of patented innovation have a positive joint effect on its degree centrality in the industry’s global technological network. As we have argued, breadth requires a location to support several industries whereas depth requires a location to support specialist firms operating in certain industries. The trade-off between depth and breadth brings us to the well-known debate about the relative importance of Marshall–Arrow–Romer (MAR) externalities and Jacobian externalities in industrial agglomeration. According to the MAR model, knowledge is industry-specific (Marshall, 1890; Arrow, 1962; Romer, 1986; Van der Panne, 2004) so that knowledge spillovers arise between firms within the same industry. These externalities are enhanced when a location specializes or builds depth in particular technologies. Jacobs (1969), on the other hand, noted that knowledge spills over across industries, resulting in novel recombinations, so that the key to innovative strength is industry diversity or technological variety. Innovations in general are new combinations of extant ideas (Schumpeter, 1934). From a technological standpoint, it implies that new industries arise when knowledge from several industries is usefully recombined to generate revolutionary products or services (Kodama, 1992). Such industry-creating knowledge recombinations include biomedicine, nanotechnology and new media (Feldman and Lendel, 2010) as well as digital photography (Tripsas, 2009). Accordingly, emerging industries are found to benefit more from Jacobian externalities of technological variety than MAR externalities of technological specialization (Henderson et al., 1995; Boschma and Frenken, 2011; Neffke et al., 2011). Additionally, from a regional standpoint, Storper and Walker (1989) argued that ‘industries create regional resources and not the other way around … Firms and sectors generate their own input histories, and those of their chosen regions, at the same time.’ (p. 96). Criticizing classic location theory which suggested that industries develop where the best locational conditions are, Storper and Walker proposed that the conditions do not exist per se but are created first and often as an outcome of industry activities. They argued that new industries form away from the established industrial centers as they thrive not on specialist inputs but modifiable generic resources. The new locations with such generic resources may be peripheral to the existing industrial nexus and may be present in both advanced and emerging economies, but are unlikely to be in the poorest underdeveloped countries (Mudambi and Santangelo, 2016). That is, these locations must contain enough generic resources providing them with the ‘potential to engage in global innovation networks’ (Mudambi and Santangelo, 2016, 3). The generic resources may be infrastructure, institutions, universities and supplier networks, but more importantly for this paper, a skilled labor force that can undertake standardized activities in established, mature technologies. These technologies also offer a significant amount of standardized and codified knowledge through patented innovations. The standard knowledge of mature technologies when recombined may find innovative applications in the new industries. Firms may recombine such generic knowledge of mature technologies available locally with firm-specific knowledge to rebundle the existing technologies and find a new combination that generates growth (Bathelt and Boggs, 2003). Thus, knowledge about a number of existing technologies, including those core to the new industry (which we call the breadth of patented innovation) is more effective for emerging industry patented innovations than specialization in particular technologies (which we call the depth of patented innovation). We thus hypothesize the following: H4: In an emerging industry, a location’s breadth of patented innovation has a larger effect than its depth of patented innovation on its degree centrality in the industry’ technological network of locations. 2.2. Breadth, depth and the entry of locations in global technological network Why does a new industry emerge away from old industrial areas? There are two reasons as argued in our theoretical arguments. One is due to the nature of the emerging industry, which originates from the recombination of established and mature technologies. These recombinations may involve standardized knowledge rather than specialized inputs of the core industrial areas. The other reason, which Storper and Walker put forth, involves the old and established industrial center’s unwillingness to support new sectors. The old center becomes ‘dominated by the entrenched interests of the old sector’ (Storper and Walker, 1989, 361) and local labor relations become too inflexible to accommodate the requirements of the new sector. We note that the established firms and centers may participate in emerging industry innovations but they may not value the new industries as much as their (old) focal industries. Rather, they may look at emerging industry innovations opportunistically. As new industries branch out from existing industries, various locations appear in the technological network of newly emerging industries. There is likely to be significant heterogeneity in terms of the innovation activity at these locations prior to entering the emerging industry. The entering locations may be well-established innovative locations with a long history of innovation and thus a high total output of patented innovations prior to entering the emerging industry. Or, these may be emergent innovative locations with relatively lower total patented innovation output prior to entry. In addition to entry timing, the established and emergent locations may also differ in terms of the firms situated there. As firms geographically disperse their value chains (Mudambi, 2008), they are more likely to prefer locations with agglomeration externalities such as knowledge spillovers and the availability of skilled labor for their R&D activities (Storper, 1995; Almeida, 1996; Alcácer, 2006; Beugelsdijk and Mudambi, 2013). On average, established locations with a history of high innovation output are more likely to have both generic and specialized local knowledge resources and social institutions promoting knowledge externalities when compared with emergent locations (Mudambi and Santangelo, 2016). In absolute terms, established locations may have more breadth and depth of patented innovations than emergent locations. As established locations branch out into new industries, part of their technological knowledge overlaps with the new industry. They may continue to exploit this overlap for emerging industry innovation and increase depth within those technologies. They may also explore and expand their technological knowledge for emerging industry innovation and increase breadth in the industry’s technologies. However, exploration in an emerging industry is doubly uncertain. In addition to the industry uncertainty noted earlier, there is exploration uncertainty where payoffs are long term and uncertain (March, 1991). However, as we argue, for emerging industries, breadth of patented innovation associated with exploratory search has a larger effect on the location’s degree centrality than depth of patented innovation associated with exploitative search. Thus, although doubly uncertain, exploratory breadth may have benefits. On the surface, such uncertain yet rewarding exploration and breadth may appear to be the strength of established locations given their long innovation experience and high innovation output. However, exploration is an option that firms often choose not to exercise (Kogut and Zander, 1992). There is strong evidence that successful firms’ search is path dependent (Penrose, 1959; Kogut and Zander, 1992; Teece, 1994; Patel and Pavitt, 1997) which aggregates at the level of the location (Storper, 1995; Maskell and Malmberg, 2007). Firms diversify into new technologies closer to their existing technological knowledge and changes in their technological profiles occur very slowly (Jaffe, 1986). Extending this logic to locations, cluster studies noted the presence of cluster identity within which the agglomerated firms define their field of competition (Florida and Kenney, 1990; Pouder and St. John, 1996). Deviating from this identity and competing in new technologies may be costly to the clustered firms given their capabilities, resources and competitive pressures (Porter, 1980; Kogut and Zander, 1992). Further, competitiveness in existing industries can impact the evaluation of a technological opportunity (Jaffe, 1986), especially if it does not directly address existing customer needs (Christensen and Bower, 1996). Therefore, exploitative search focused on short-term outcomes may appear more lucrative than exploratory search with longer-term outcomes (March, 1991). Although a sole focus on short-term profitability is detrimental to a firm’s long-run competitiveness, an established firm often does not have the flexibility and risk tolerance necessary to succeed when moving away from its existing knowledge base (Kogut and Zander, 1992). Building on this logic, we argue that, when evaluating an emerging industry, a majority of firms at established locations may prefer exploitative within-technology search resulting in increasing depth of innovation. It is likely that only a small minority of resident firms in such locations will focus on the breadth of innovation in the emerging industry. Moreover, firms deciding to locate in established locations may be attracted by the high level of specialized resources present there (Mudambi and Santangelo, 2016). Firms’ conscious decisions to exploit these specialized resources would then result in a greater depth of innovation at the established locations. Established locations may thus participate in the emerging industry innovation network more through their depth than breadth. Emergent locations host fewer innovative firms than established locations and therefore offer relatively lower agglomerative benefits. However, the resident firms may well be very successful innovators. Since leading firms can tap into local resources more readily than lagging ones (Cantwell and Mudambi, 2011), they may be willing to consider emergent locations that have limited specialized knowledge resources but high-quality generic resources (Mudambi and Santangelo, 2016). Generic resources may support the firms’ across-technology exploratory search resulting in increasing breadth of innovation at the emergent location. The concepts of established and emergent are relative to each other and so are our arguments. In absolute terms, emergent locations may be low in both breadth and depth. Relatively however, they may focus more on breadth than depth when compared with the established locations when innovating for the emerging industry. That is, their participation in the industry innovation network may be more through their breadth than depth. We thus hypothesize the following: H5a: In an emerging industry, established innovative locations are more likely to increase technological depth rather than technological breadth in the industry’s technologies. H5b: In an emerging industry, emergent innovative locations are more likely to increase technological breadth rather than technological depth in the industry’s technologies. Combining the arguments from Hypotheses 4 and 5, we state the following: emergent locations focus more on breadth which has a stronger effect on a location’s degree centrality in the global innovation network than depth. Taken together, these statements imply that emergent locations that host successful innovative firms are likely to rapidly achieve centrality in the global innovation network underpinning emergent industries. Thus, although emergent locations have lower innovation experience, fewer innovative firms and thus lower innovation output prior to entering the emerging industry, they may still exhibit fast accretion toward centrality in the industry’s global innovation network with respect to established locations. Accretion toward centrality does not mean that emergent locations achieve parity with established locations (Awate et al., 2012, 2015). However, it is indicative of a significant narrowing of the technological gap between established and emergent locations. 3. The global wind power industry Wind power is one of the world’s fastest growing sources of energy. In 2011, the global cumulative installed wind power capacity was 237,669 megawatts (MW) (GWEC, 2012), capable of providing around 3% of global electricity consumption (WWEA, 2012). The industry is highly concentrated with a few large manufacturers together having a market share of over 50%. The top five manufacturers in 2011 were Vestas (12.7%), Sinovel (9%), Goldwind (8.7%), Gamesa (8.0%) and Enercon (7.8%), with GE (7.7%) and Suzlon (7.6%) close behind. Western (and particularly European) firms such as Vestas, NEG (Denmark) and Enercon (Germany) have traditionally dominated the industry. However, in the 1990s, it witnessed the entrance of a number of emerging economy manufacturers that offered turbines for considerably lower prices than the more established industry players (see Musgrove, 2010, especially Chapter 5). In particular, Chinese and Indian companies such as Goldwind (China), Sinovel (China) and Suzlon (India) have displayed highly impressive growth rates, both domestically and internationally, and have begun to challenge the more established players in the industry (Awate et al., 2012). 3.1. Genesis and growth Early experiments with the use of wind energy to generate electricity began in the late eighteenth century in Europe and the USA. During the two world wars, the restrictions on fossil fuel imports started wind turbine development in numerous countries, but these developments were small scale and experimental. The oil crises of the 1970s were the triggering events that resulted in the establishment and growth of the modern wind turbine industry. In the early 1980s, the industry experienced a number of product innovations primarily from two competing technological trajectories, one from California in the USA, and another from Jutland in Denmark (Garud and Karnøe, 2003; Musgrove, 2010; Nielsen, 2010). Providing further details on the design competition in the industry, Awate et al. (2012) note that ‘in California, the U.S. Department of Energy and NASA engaged a number of engineers in response to the oil crises to cooperate with companies in the aircraft industry to develop sophisticated, high-technology, large-scale and aerodynamically-optimized turbines based on aeronautical engineering principles. These turbines were particularly distinguished by their two-bladed rotor pitch regulation. In Denmark, however, different wind power enthusiasts such as farmers, carpenters, and engineers collaborated to develop robust, small-scale three-bladed turbines with reliability and ruggedness as the key concerns. While these turbines were initially small in size, a number of incremental innovations led to the wind turbines eventually being scaled up to meet broader commercial demands. During the California wind boom of the 1980s, the Danish low technology—high reliability wind turbine proved commercially superior to its American high technology—low reliability counterpart. As a result, the Danish turbine emerged in the late 1980s as the dominant industry standard with Danish firms and designs controlling the majority of the world market share (Garud and Karnøe, 2003).’ 3.2. Technological development The basic design introduced by Danish manufacturers in the late 1980s has not seen any radical changes, although wind turbines today are both larger in size and typically use blades that optimize the output of the turbine by automatically adjusting with changing wind directions. The average blade span of the larger wind turbines has increased from about 30 m in 1990 to 90 m by 2008. New generator designs also allow the rotor to operate at varying speeds. Overall, the successive generations of turbines have been developed with a particular focus on increasing scale and reducing the overall cost of energy (Musgrove, 2010). Electricity and aerodynamics are the industry’s core technologies and in addition it uses a variety of other peripheral technologies from sectors such as automobiles, construction, chemicals, information and communication technology and so on. In addition, innovations in linked industries such as energy storage systems have a large impact on industry’s growth (Gates, 2011). In the early 2000s, the industry witnessed a major change—the offshore era. Offshore wind generation was particularly attractive because wind speeds are stronger and steadier on the sea than on the land, which increases turbine ‘availability’, an industry term that roughly corresponds to capacity utilization. Higher availability leads to more electricity produced per installed MW, and therefore cheaper electricity generation. However, the installation of such large offshore turbines and their connection to the grid poses a major challenge. Moreover, the design and engineering involved in such large offshore turbines is far different from the mid-scale onshore turbines. The diverse capabilities, risks and large capital investment necessary for offshore wind turbine projects required the participation of a whole new set of industry players. These included firms from various sectors such as oil and natural gas, electricity (Markard and Petersen, 2009), shipbuilding and heavy engineering. Many of them were large multinationals with deep pockets and strong R&D skills. Some examples include General Electric (GE), Siemens, Hitachi and Mitsubishi. As a result, a major change also came in the form of a dramatic shift in the industry’s IPR regime. More and more firms and independent inventors began patenting their technology. Established wind turbine manufacturers like Vestas that had formerly relied on informal innovation networks, began rethinking their R&D strategies in the offshore era (Pedersen and Larsen, 2009). The offshore era thus represented a discontinuous change for the industry that resulted in increased entry and fiercer competition. 3.3. Policy and market activity While the patented technological innovations of the wind industry are the core of our study, we would like to point out that the establishment and growth of its market has always been dependent on the government support. The support has been in the form of production tax credits (PTCs) that make wind power prices comparable to conventional power (Cardwell, 2013). Additionally, policy instruments such as guaranteed grid access facilitated the growth of the wind industry in several countries. Among the leading market locations, Denmark was the pioneer in offering government support for the growth of the wind industry from its early years. The government policies involved long-term agreements with power companies, producers and users. The government has provided investment subsidies to individuals and cooperatives investing in wind turbines since the late 1970s. Further, investments in the sector and the sale of surplus electricity were tax deductible, which popularized investment in wind power and increased public support (Buen, 2006). In the 1980s, the government negotiated a long-term agreement with the power companies by which they agreed to guarantee grid connection for turbines operated by individuals and cooperatives, pay part of the grid connection cost and buy excess power from them at 85% of the consumer price. Danish policies served as a model for a number of European markets. The German government, for example, adopted the Danish focus on small-scale wind energy in the late 1980s by assuring demand for these turbines. These were soon ramped up in 1991 to support utility scale electricity generation. The policies included feed-in tariffs to wind power producers to feed wind power into the grid at a fixed price. Like Denmark and Germany, other European markets such as Spain, Portugal and Ireland established similar policy instruments to encourage market growth. In addition to the government policies, countries such as the UK and Italy also opted for market-based mechanisms (e.g., tradable green certificates) to reduce the cost of wind generation. Across various European countries, the policies were applied more or less consistently over the years (GWEC, 2012). In the USA, however, the governmental support was comparatively inconsistent with the main policy instrument being PTC. PTC however is a performance-based incentive with a set expiry, which has often resulted in uncertainty among wind power producers and delayed investment and short-term market slowdown. After economic reforms, the rapid economic development of emerging markets, particularly India and China, exerted excessive demand on their national energy supply. The large power deficits from the traditional energy sources made renewable sources such as wind very attractive to those governments. In the late 1980s, the Indian government began a process of creating an open economy environment and implemented favorable policies to encourage entry by foreign firms in grid-quality wind energy generation. These included 100% accelerated depreciation on the wind equipment, customs and excise duty relief, 5-year tax holiday and soft loans (Rajsekhar et al. 1999). The favorable policy environment supported the entry of European manufacturers such as Vestas and NEG that entered through joint ventures with local firms. Compared with India, the Chinese market started developing slowly. In the 1980s and the early-1990s, the projects were small-scale, demonstrational and funded through foreign grants and government loans. However, in the1990s, the government introduced a number of policy instruments that resulted in exponential increases in the installed capacity. 4. Methodology 4.1. Data We examine patents generated at each location and track them over time. We gathered all the patents filed with the USPTO that claimed wind turbine innovations. The final dataset included 1895 patents filed and granted between 1961 and March 2012. From each patent, we extracted the inventors, their locations and the technological classifications (or technology classes) assigned to them by the USPTO. We then performed manual checks to identify unique locations and inventors. We applied a 30-mile radius threshold to group nearby locations. This analysis yielded a total of 566 unique locations, 2204 unique inventors1 and 177 technology classes over the study period. Next, we created the industry’s technological network, outlined earlier, by connecting technology classes with the locations. Thus, a tie existed between the two nodes of technology and geography if there was an inventor working in that particular technology class and situated at that particular location.2 The number of such inventors appeared as the weight for that tie. Further, if the same location-inventor-technology pattern repeated on multiple patents within the same year, the number of such patents was also accounted for in the tie weight. Such networks of technology classes and locations were created for each year of the cumulative patent records. Figure 1 shows the network in 2011. The nodes are placed depending on their connections. For example, the top four classes in terms of the number of patents (classes 290, 322, 415 and 416) are at the core of the map encircled by several locations that directly connect to them. Classes 290, 322 and 415 gather a number of locations shown by a large cluster of red nodes at the center of the figure. The location nodes appearing in the middle of this cluster, for example San Francisco in California and Aalborg in Denmark, primarily connect to the three core classes. The location nodes on the perimeter connect to the core classes as well as a number of other classes. For example, the top locations in terms of the patent output, Schenectady in New York and Greenville in South Carolina, connect to a host of other classes depicted by a cluster of blue squares below these locations. The two emerging market locations, namely Shanghai in China and Bangalore in India, also appear closer to the perimeter indicating that they connect to the core as well as non-core classes. Figure 1 View largeDownload slide Technology-geography network in 2011. Notes: The figure is drawn with a spring embedding layout using UCINET’s NetDraw software (Borgatti et al., 2002). For clarity, tie weights are not shown. Figure 1 View largeDownload slide Technology-geography network in 2011. Notes: The figure is drawn with a spring embedding layout using UCINET’s NetDraw software (Borgatti et al., 2002). For clarity, tie weights are not shown. 4.2. Variables 4.2.1. Dependent variable We measured the dependent variable representing a location’s innovation activity (as measured by patent output) by focusing on the locations in the two-mode networks. We used a location node’s valued degree centrality (vdeg) to measure the patenting activity of that location. This centrality was calculated as the sum of tie weights of all ties originating from the location. Thus, the centrality score increases with an increase in the individual tie weights or in the number of ties or both. In other words, it increases in value if the inventors at a location invent more in the same technology class (increase in tie weight) as well as if they invent more in different technology classes (increase in the number of ties). Further, if the location adds new inventors, the increase is also captured in the centrality score through increase in tie weight. Thus, our dependent variable captures patenting activity along the dimensions of technologies as well as inventors. Raw patent counts treat both kinds of activities as one and underestimate the total patenting activity at a location. Our dependent variable on the other hand provides a more accurate measure of a location’s patenting activity. 4.2.2. Explanatory variables Depthi,t−1: Depth develops when the location repeatedly invents in a certain set of technologies. We thus measured the depth of patented innovation at a location by finding the number of ties the location adds in the same technology class every year. As the two-mode networks contain cumulative data, we computed the number of ties added to the existing technology classes from previous year. The variable was lagged by a year to denote the depth of patented knowledge base as of previous year. This was necessary since we study how the existing patent stock of a location impacts its patenting activity. The variable and its squared transformation were used in the regressions to test H2. Breadthi,t−1: Breadth develops when the location increases the number of technology classes in which it invents, particularly when it adds classes that are important in the wind turbine innovation network. We gauged breadth using the closeness centrality measure which denotes the reach of a node within the network (Faust, 1997). With this measure, the reach of a location increases with the number of direct connections to distinct technology classes. Further, the measure also takes into account the importance of those technology classes within the network. This is because a location connected to more central classes has a higher reach in the overall network than a location connected to less central, peripheral classes. Thus, even if two locations are directly connected to the same number of classes, the location connected to more central classes has a higher reach in the network. We calculated breadth of a location as sum of its reciprocal distances to all other technology nodes in the network. In this way, breadth is captured at the network level and captures all the changes happening in the topology of the network through small increments. This breadth variable and its squared transformation were used in the regressions to test H1. As breadth is a network-level measure, it increases when the location makes a connection to a new technological class as well as when a previously connected class becomes more central in the network. Thus, the breadth values in our dataset continue to increase in small increments whenever there is a change in the network structure and even if the location does not produce new patents or has a new depth-only patent. Pre-entry patent output: Pre-entry patent output helps to qualify established and emergent locations to test hypotheses H5a and H5b. It is measured as the location’s share of the USPTO patents one year before filing its first wind turbine patent. As the denominator (i.e., total number of USPTO patents) keeps growing each year, only those locations that continue to file patents and increase the numerator achieve higher score on this measure. This is a time invariant variable. For location i that entered in year t1, it is denoted as: Pre-entry patent outputi, t1 = total patents filed by i until (t1 − 1)/total USPTO patents filed until (t1−1). By measuring the pre-entry patent output, the variable identifies locations that may be new to the wind power industry but otherwise established innovators. We also use a dummy variable to distinguish developing economy location from advanced economies. The dummy variable is created following World Bank classification and identifies locations that are likely to be new to both, patents in general and wind patents in particular. 4.2.3. Controls Market activity: Demand is often shown to drive innovative clusters (Bresnahan et al., 2001). We used the installed wind power capacity to measure the market activity and thus demand. The data were only available at country-level (GWEC, 2012). It is likely that countries with higher installed capacity may consider wind power as an important source of electricity and may encourage innovation activity in this industry. This is a time variant variable. Global connectivity: Given the important role played by global knowledge sources, a location’s global connectivity may positively impact its innovation activity. We measure global connectivity along two dimensions—physical connectivity and virtual connectivity. Physical connectivity is measured as a ratio variable taking into account how far a location is from an airport scaled by the airport size, as follows:   Airport connectivity=(number of destinations served by the nearest international airport)/(distance to the nearest international airport). Using this variable, we capture the effect of being in the vicinity of a larger airport versus a smaller airport. Availability of air services is shown to promote information exchange between cities and a factor deciding headquarters location choice (Bel and Fageda, 2008). We thus argue that the variable airport connectivity captures a location’s physical access to global sources of information and the kind of firms that are located there, which impacts the innovation output of the location. The Google Maps program was used to find the shortest distance in miles to the nearest international airport from each of the location. This is a time invariant variable. Similarly, virtual connectivity to the world may also be necessary to access documented knowledge through the World Wide Web. We use international Internet bandwidth, which is the capacity measured as bits per person. The variable is time variant as it is based on the longitudinal data obtained from International Telecommunications Union and World Bank estimates provided by the EconStats database (ITU, 2005; EconStats, 2012). Patent impact: A location that produces more impactful innovations may be more likely to attract entry by other firms. As more firms enter, the number of inventors at a location may increase. This may increase the location’s patenting activity. The impact of a patent is often measured using the number of citations the patent receives. Accordingly, we found the total number of citations received by a location’s patents for each year. For each year, this variable captures the cumulative number of citations received. We then lagged this variable by 1 year, so that it captured the impact of a location’s patents as of previous year. Of course, patents granted in earlier years may receive more citations than newer patents. We follow Rosenkopf and Nerkar (2001), and control for this bias by including year dummies. Innovativeness of technologies: A location’s innovation activity may be influenced by its technological connections. That is, a location that operates in the most innovative technologies is likely to be more innovative than other locations. In our network model, eigenvector centralities measure how central the locations’ connections are (in terms of patenting activity) and the strengths of those connections (Faust, 1997). We computed this centrality measure, but found it to be highly correlated with the explanatory variables. The variable was thus dropped. As an alternative approach, we computed the valued degree centralities of technology classes to identify the most innovative classes in terms of the patent output. This procedure enabled us to identify the top four innovative classes (classes 290, 322, 415 and 416). These classes belonged to the core technologies of the wind power industry, namely electricity (classes 290, 322) and aerodynamics (classes 415, 416). We used dummy variables to indicate locations’ connections to these classes. Post technological discontinuity dummy: Patenting activity in the industry increased considerably with the offshore technological discontinuity post-2000. We included a dummy variable to capture the effect of this technological discontinuity on locations’ patenting activity. Country controls: As the innovativeness varies considerably across countries, we controlled for this heterogeneity using dummy variables for the top three countries in terms of patenting activity. We calculated the patenting activity of a country by summing the cumulative measure of valued degree centralities of locations in that country in 2011. Accordingly, the top three innovative countries were the USA, Denmark and Germany. Finally, we also controlled for country-level GDP per capita, using data obtained from the World Bank. Additionally, we used a dummy variable to denote developed countries. 5. Estimation and results We present univariate descriptive statistics, including correlations in Table 1. The correlations generally exhibit the expected signs and none are large enough to cause serious concern about multicollinearity as indicated by the variance inflation factor (VIF). The mean VIF for all the variables is 1.55. As can be seen, the dependent variable is an over dispersed count variable. This led us to opt for a negative binomial regression methodology for our location-year panel data as the Poisson regression model would be too restrictive given its equidispersion property. Several of our variables were time invariant. Further, although many explanatory variables were time-variant, several locations exhibited relatively low levels of intertemporal variation. In these circumstances, it is generally preferable to choose a random-effects model, as using a fixed-effects model would lead to many data points being omitted (Hsiao, 1986). Table 1 Univariate statistics and variable correlations Variable  1  2  3  4  5  6  7  8  9  (1) Vdeg                    (2) Breadth  0.4303                  (3) Depth  0.9674  0.3129                (4) ln(Installed wind capacity)  0.1936  0.4632  0.1228              (5) Pre-entry innovation output  0.2099  0.1332  0.1861  0.0105            (6) Airport connectivity  0.056  0.0802  0.0293  −0.035  0.1818          (7) ln(Bandwidth)  0.0805  0.2186  0.0332  0.3492  0.0065  0.0065        (8) Patent impact  0.6355  0.4308  0.5866  0.1959  0.3454  0.1112  0.1039      (9) ln(GDP per capita)  0.1850  0.4224  0.1227  0.6129  0.0629  0.0186  0.2784  0.1899    Mean*  2.6316  0.7814  1.2473  2.7498  0.0866  2.3555  −7.9092  19.0190  9.6260  (9.1680)  (1.0905)  (6.7858)  (8.4733)  (0.3095)  (2.2243)  (8.0707)  (54.3449)  (1.1512)  Min  0  0  0  −12.2061  0  0.0156  −12.206  0  4.3291  Max  414  4.5574  376  11.0407  3.9518  10.4667  11.2665  689  11.4636  Mean VIF  1.55  Variable  1  2  3  4  5  6  7  8  9  (1) Vdeg                    (2) Breadth  0.4303                  (3) Depth  0.9674  0.3129                (4) ln(Installed wind capacity)  0.1936  0.4632  0.1228              (5) Pre-entry innovation output  0.2099  0.1332  0.1861  0.0105            (6) Airport connectivity  0.056  0.0802  0.0293  −0.035  0.1818          (7) ln(Bandwidth)  0.0805  0.2186  0.0332  0.3492  0.0065  0.0065        (8) Patent impact  0.6355  0.4308  0.5866  0.1959  0.3454  0.1112  0.1039      (9) ln(GDP per capita)  0.1850  0.4224  0.1227  0.6129  0.0629  0.0186  0.2784  0.1899    Mean*  2.6316  0.7814  1.2473  2.7498  0.0866  2.3555  −7.9092  19.0190  9.6260  (9.1680)  (1.0905)  (6.7858)  (8.4733)  (0.3095)  (2.2243)  (8.0707)  (54.3449)  (1.1512)  Min  0  0  0  −12.2061  0  0.0156  −12.206  0  4.3291  Max  414  4.5574  376  11.0407  3.9518  10.4667  11.2665  689  11.4636  Mean VIF  1.55  *Standard deviations in parentheses. 5.1. Breadth and depth The results of the panel negative binomial regressions are shown in Table 2. The first model contains only the control variables. The chi-squared test statistic indicates that the model as a whole is significant relative to the null model. The control variables are generally statistically significant with the expected signs. Table 2 Random effects negative binomial regression results Dependent variable Vdeg  (1) Controls only  (2) Breadth and square added  (3) Depth and square added  (4) Interaction added  (5) Standardized variables  (6) Incident rate ratio  Breadth    1.9322***  1.8545***  1.9570***  1.3286***  3.7756***      (0.049)  (0.044)  (0.046)  (0.039)  (0.147)  Breadth2    −0.3392***  −0.4226***  −0.4773***  −0.5677***  0.5669***      (0.012)  (0.011)  (0.013)  (0.016)  (0.009)  Depth      0.0235***  0.0066***  0.0743***  1.0771***        (0.001)  (0.002)  (0.013)  (0.014)  Depth2      −0.00003***  −0.0001***  −0.0028***  0.9972***        (0.000002)  (0.000004)  (0.0002)  (0.0002)  Breadth×depth        0.0058***  0.0430***  1.0439***          (0.001)  (0.006)  (0.006)  Pre-entry innovation  −0.1590*  0.0794  0.1871*  0.1877*  0.0581*  1.0598*  output  (0.085)  (0.084)  (0.096)  (0.097)  (0.030)  (0.032)  ln(Installed wind)  0.0165*  0.0130  0.0070  0.0053  0.0448  1.0458    (0.009)  (0.009)  (0.009)  (0.009)  (0.073)  (0.076)  Airport connectivity  −0.0003  0.0454***  0.0357**  0.0334**  0.0743**  1.0771**    (0.015)  (0.014)  (0.014)  (0.014)  (0.031)  (0.034)  ln(Bandwidth)  0.0029  −0.0001  −0.0017  −0.0018  −0.0145  0.9856    (0.002)  (0.002)  (0.002)  (0.002)  (0.015)  (0.015)  Patent impact  0.0018***  0.0024***  0.0002  0.0004***  0.0226***  1.0228***    (0.000)  (0.000)  (0.000)  (0.000)  (0.008)  (0.008)  ln(GDP per capita)  0.4105***  0.1591***  0.1149***  0.1188***  0.1368***  1.1466***    (0.041)  (0.039)  (0.039)  (0.039)  (0.045)  (0.052)  Class 240  1.1338***  0.6480***  0.7791***  0.8120***  0.8120***  2.2525***    (0.030)  (0.030)  (0.029)  (0.029)  (0.029)  (0.066)  Class 320  0.3375***  0.3326***  0.3550***  0.3698***  0.3698***  1.4474***    (0.030)  (0.030)  (0.030)  (0.030)  (0.030)  (0.043)  Class 415  0.5075***  0.4106***  0.4100***  0.4273***  0.4273***  1.5331***    (0.027)  (0.026)  (0.026)  (0.026)  (0.026)  (0.039)  Class 416  1.5731***  0.7688***  0.8689***  0.8901***  0.8901***  2.4355***    (0.037)  (0.039)  (0.037)  (0.037)  (0.037)  (0.091)  USA  −0.2680***  0.0666  0.3072***  0.2962***  0.2962***  1.3448***    (0.087)  (0.074)  (0.075)  (0.075)  (0.075)  (0.101)  Denmark  −1.1085***  −0.3882***  −0.1033  −0.1447  −0.1447  0.8653    (0.158)  (0.141)  (0.144)  (0.145)  (0.145)  (0.125)  Germany  −0.4931***  −0.2115*  −0.0204  −0.0205  −0.0205  0.9797    (0.131)  (0.111)  (0.111)  (0.111)  (0.111)  (0.109)  Post-discontinuity  1.7033***  0.5846*  1.5605***  1.7283***  1.7283***  5.6311***    (0.308)  (0.334)  (0.312)  (0.311)  (0.311)  (1.754)  Constant  −5.0319***  −3.2699***  −2.6156***  −2.6673***  −0.1403  0.8691    (0.418)  (0.415)  (0.406)  (0.406)  (0.276)  (0.240)  Observations  16,053  16,053  16,053  16,053  16,053  16,053  Number of locations  557  557  557  557  557  557  Chi-squared  20,290.9***  18,607.9***  21,080.3***  20,956.2***  20,956.2***    Log-likelihood  −19,016.4  −17,790.8  −17,378.4  −17,349.83  −17,349.83    L-R test statistic    2451.2***  824.8***  57.17***      Wald: Breadth=Depth        2041.4***  1277.8***    Wald: Breadth2=Depth2        1316.5***  1316.5***    Dependent variable Vdeg  (1) Controls only  (2) Breadth and square added  (3) Depth and square added  (4) Interaction added  (5) Standardized variables  (6) Incident rate ratio  Breadth    1.9322***  1.8545***  1.9570***  1.3286***  3.7756***      (0.049)  (0.044)  (0.046)  (0.039)  (0.147)  Breadth2    −0.3392***  −0.4226***  −0.4773***  −0.5677***  0.5669***      (0.012)  (0.011)  (0.013)  (0.016)  (0.009)  Depth      0.0235***  0.0066***  0.0743***  1.0771***        (0.001)  (0.002)  (0.013)  (0.014)  Depth2      −0.00003***  −0.0001***  −0.0028***  0.9972***        (0.000002)  (0.000004)  (0.0002)  (0.0002)  Breadth×depth        0.0058***  0.0430***  1.0439***          (0.001)  (0.006)  (0.006)  Pre-entry innovation  −0.1590*  0.0794  0.1871*  0.1877*  0.0581*  1.0598*  output  (0.085)  (0.084)  (0.096)  (0.097)  (0.030)  (0.032)  ln(Installed wind)  0.0165*  0.0130  0.0070  0.0053  0.0448  1.0458    (0.009)  (0.009)  (0.009)  (0.009)  (0.073)  (0.076)  Airport connectivity  −0.0003  0.0454***  0.0357**  0.0334**  0.0743**  1.0771**    (0.015)  (0.014)  (0.014)  (0.014)  (0.031)  (0.034)  ln(Bandwidth)  0.0029  −0.0001  −0.0017  −0.0018  −0.0145  0.9856    (0.002)  (0.002)  (0.002)  (0.002)  (0.015)  (0.015)  Patent impact  0.0018***  0.0024***  0.0002  0.0004***  0.0226***  1.0228***    (0.000)  (0.000)  (0.000)  (0.000)  (0.008)  (0.008)  ln(GDP per capita)  0.4105***  0.1591***  0.1149***  0.1188***  0.1368***  1.1466***    (0.041)  (0.039)  (0.039)  (0.039)  (0.045)  (0.052)  Class 240  1.1338***  0.6480***  0.7791***  0.8120***  0.8120***  2.2525***    (0.030)  (0.030)  (0.029)  (0.029)  (0.029)  (0.066)  Class 320  0.3375***  0.3326***  0.3550***  0.3698***  0.3698***  1.4474***    (0.030)  (0.030)  (0.030)  (0.030)  (0.030)  (0.043)  Class 415  0.5075***  0.4106***  0.4100***  0.4273***  0.4273***  1.5331***    (0.027)  (0.026)  (0.026)  (0.026)  (0.026)  (0.039)  Class 416  1.5731***  0.7688***  0.8689***  0.8901***  0.8901***  2.4355***    (0.037)  (0.039)  (0.037)  (0.037)  (0.037)  (0.091)  USA  −0.2680***  0.0666  0.3072***  0.2962***  0.2962***  1.3448***    (0.087)  (0.074)  (0.075)  (0.075)  (0.075)  (0.101)  Denmark  −1.1085***  −0.3882***  −0.1033  −0.1447  −0.1447  0.8653    (0.158)  (0.141)  (0.144)  (0.145)  (0.145)  (0.125)  Germany  −0.4931***  −0.2115*  −0.0204  −0.0205  −0.0205  0.9797    (0.131)  (0.111)  (0.111)  (0.111)  (0.111)  (0.109)  Post-discontinuity  1.7033***  0.5846*  1.5605***  1.7283***  1.7283***  5.6311***    (0.308)  (0.334)  (0.312)  (0.311)  (0.311)  (1.754)  Constant  −5.0319***  −3.2699***  −2.6156***  −2.6673***  −0.1403  0.8691    (0.418)  (0.415)  (0.406)  (0.406)  (0.276)  (0.240)  Observations  16,053  16,053  16,053  16,053  16,053  16,053  Number of locations  557  557  557  557  557  557  Chi-squared  20,290.9***  18,607.9***  21,080.3***  20,956.2***  20,956.2***    Log-likelihood  −19,016.4  −17,790.8  −17,378.4  −17,349.83  −17,349.83    L-R test statistic    2451.2***  824.8***  57.17***      Wald: Breadth=Depth        2041.4***  1277.8***    Wald: Breadth2=Depth2        1316.5***  1316.5***    Note: Standard errors in parentheses, ***p < 0.01, **p < 0.05, and *p < 0.1. Model 2 adds the breadth variable along with its squared term. While breadth has a positive and significant coefficient, its squared term has a negative and significant effect signaling the possibility of a negative curvilinear effect. Model 3 adds the depth variable with its squared term. Both breadth and its squared term retain their effects after this addition. Depth has a positive and significant coefficient and its squared term produces a small negative but significant effect. Next, Model 4 adds the interaction term of breadth and depth, whose coefficient is positive and significant. The models are significant as a whole when compared with a null model, shown by the chi-squared statistic. Further, the log-likelihood increases with the addition of successive regressors from Model 1 to Model 4. We conduct likelihood ratio (L-R) tests to check if the nested models are significantly different from one another (UCLA, 2013). As shown by the L-R test statistic, adding explanatory variables to the control model 1 significantly improves the model fit. Further, each model offers a better fit over its predecessor. Finally, and perhaps most importantly, the coefficients of all the key variables retain their signs and significance through all models 1–4. Model 4 is the enveloping model and we use it to test our hypotheses. Both the breadth and depth variables have positive and significant coefficients. Their squared terms are negative and significant, so that both breadth and depth have negative curvilinear effects on a location’s patenting activity. This evidence supports H1 and H2. Further, their interaction term has a positive and significant effect on the dependent variable thereby supporting H3. 5.2. Breadth versus depth We test H4 by estimating Model 4 using standardized independent variables with zero mean and unit variance. These results are reported in Model 5 in Table 2. As seen, the positive significant coefficient of breadth is larger than that of depth. A Wald test indicates that this breadth coefficient is significantly larger than the corresponding depth coefficient, reported in Table 2. Further, the negative coefficient of the squared term of breadth is found to be significantly more negative than the corresponding coefficient of depth (also shown in Table 2). While breadth has a larger positive effect than depth, its squared effect is more negative. Thus, the effect of breadth rises much faster but also falls much faster than depth, indicating that the centrality of a location is more sensitive to breadth than depth. Model 6 reports the incident rate ratios (IRRs) for Model 5. It represents the percentage change in the dependent variable produced by a unit change in the independent variable. It can be seen that one unit change in breadth produces 278% (377.56-100) increase and its squared term produces 43% (56.69-100) decrease in the centrality. Whereas a unit increase in depth produces 8% increase and its squared term produces about 0.3% decrease in the centrality.3 To understand the combined (i.e., quadratic) effects of The breadth and depth coefficients, we look at the predictive margins, plotted in Figure 2. The predictions are calculated for a range of standardized breadth and depth values with other variables at their observed values. The range is selected keeping in mind the distributions of breadth and depth while at the same time keeping the two graphs comparable. It includes 90.89% of the standardized breadth values and 99.26% of the standardized depth values. As shown in the figure, the effect of breadth rises much faster and peaks much before depth. In our dataset that contains the entire population of patenting locations in the industry, the effect of depth is uniformly positive, while the effect of breadth is an inverted U-shape. Further, for most of the range, breadth produces a larger effect than depth, except at the extremities where the confidence intervals overlap. Therefore, while theoretically depth may have a higher effect than breadth at certain extreme values, those values are not common. We therefore conclude that, in general, breadth produces a larger effect than depth in this emerging industry, supporting H4. Figure 2 View largeDownload slide Effect of breadth versus depth. Figure 2 View largeDownload slide Effect of breadth versus depth. 5.3. Established versus emergent locations In order to test hypotheses H5a and H5b that study relative accumulations of breadth and depth in established and emergent locations, we perform two tests, shown in Table 3. We first estimate random effects models with breadth and depth as functions of pre-entry patent output. We control for the developed economy locations, time of entry into the sample and the top technology classes in terms of the patent output. We do not expect any other control variables to have an effect on breadth and omit them from this analysis. The results are shown in Models (1) and (2) in Table 3. The dummy variable for developed countries produces a positive significant effect on the breadth but insignificant on the depth. Additionally, an increase in pre-entry patent output is negatively associated with breadth and positively associated with depth. Thus, more established innovative locations with higher pre-entry patent output increase depth of patented innovation in the wind power industry relative to the emergent locations. Emergent locations on the other hand increase breadth of patented innovation in the wind industry relative to the more established locations. Table 3 Breadth, depth and entering locations Variables  (1) Dependent variable: standardized breadth  (2) Dependent variable: standardized depth  (3) Hazard of depth patents        Hazard ratios  Pre-entry innovation output  −0.2198**  0.3503***  1.4410***    (0.1047)  (0.0872)  (0.1185)  Developed countries  0.0946***  0.0369  1.0730    (0.0261)  (0.0406)  (0.2678)  Time since wind entry  −0.0043**  −0.0090***  0.9732***    (0.0012)  (0.0031)  (0.0056)  Class 240  0.9569***  0.2501***      (0.0556)  (0.0561)    Class 320  0.6150***  1.4904***      (0.1061)  (0.3998)    Class 415  0.4849***  0.4713***      (0.0662)  (0.1339)    Class 416  0.9350***  0.2576***      (0.0537)  (0.0537)    Top classes      3.2236***        (0.6597)  Constant  −0.5152***  −0.1726***      (0.0251)  (0.0330)    Observations  22,035  22,035  1358  Overall R2  0.5802  0.1970    Number of locations  565  565  565  Mean VIF  1.37  1.37    Proportional hazard test        Chi-squared, p-value      3.96, 0.4112  Variables  (1) Dependent variable: standardized breadth  (2) Dependent variable: standardized depth  (3) Hazard of depth patents        Hazard ratios  Pre-entry innovation output  −0.2198**  0.3503***  1.4410***    (0.1047)  (0.0872)  (0.1185)  Developed countries  0.0946***  0.0369  1.0730    (0.0261)  (0.0406)  (0.2678)  Time since wind entry  −0.0043**  −0.0090***  0.9732***    (0.0012)  (0.0031)  (0.0056)  Class 240  0.9569***  0.2501***      (0.0556)  (0.0561)    Class 320  0.6150***  1.4904***      (0.1061)  (0.3998)    Class 415  0.4849***  0.4713***      (0.0662)  (0.1339)    Class 416  0.9350***  0.2576***      (0.0537)  (0.0537)    Top classes      3.2236***        (0.6597)  Constant  −0.5152***  −0.1726***      (0.0251)  (0.0330)    Observations  22,035  22,035  1358  Overall R2  0.5802  0.1970    Number of locations  565  565  565  Mean VIF  1.37  1.37    Proportional hazard test        Chi-squared, p-value      3.96, 0.4112  Notes: Models (1) and (2): random effects regression (standard errors clustered on locations). Model (3): Cox proportional hazard model (multiple events model using time to subsequent events from entry, Efron method of ties, robust standard errors). Standard errors in parentheses; ***p < 0.01, **p < 0.05, *p < 0.1. Next, we test a location’s likelihood of filing depth-building patents (or depth patents) using the time-to-event or hazard rate model. The nuances in the model detailed below helps to capture the richness of the data and to understand locations’ patenting behavior over time. The dependent variable is the event of filing a depth patent. It is modeled as a hazard function which is the probability that a location files a depth patent at time t conditional on not filing a depth patent up to t. We estimate the effect of pre-entry patent output on depth patenting hazard using a Cox proportional hazard model. We assume that locations enter the risk set for this estimation after filing their first patent, and then measure time to subsequent depth patenting from entry. Further, we use a conditional risk set model so that a location is at risk of the kth depth patenting event only after encountering (k−1)th depth patenting event. Results are reported in Model (3) in Table 3. To maintain the Cox model’s assumption of proportional hazards, we replace separate dummy variables for the top classes in terms of patent output using a single dummy variable representing all four classes (Table 3 reports the result of the proportional hazard test showing that the null hypothesis of proportionality cannot be rejected). As seen in Model (3), a unit increase in the pre-entry patent output increases the hazard of depth patenting by 43%. Thus, established locations have a higher likelihood of continuing with depth patenting throughout the sample period than emergent locations. Combining the results of models (1), (2) and (3) in Table 3, we find emergent locations to have increases in breadth of patented innovation and lower likelihoods of depth patenting in the wind power industry and when compared with established locations, thereby providing support for Hypotheses H5a and H5b. 5.4. Additional robustness tests We tested our results with alternative model specifications and estimation methodologies to check their robustness. We tested for potential endogeneity arising from our cumulative data structure using additional time lags of 3 and 5 years on the independent variables. The goal of these models was to test for reverse causality as a source of endogeneity as well as omitted variables simultaneously affecting the network-based dependent and independent variables. We further identified an instrument to test for omitted variable bias as a cause of endogeneity. In addition, we tested the effect of outlier locations driving the hypothesized results by excluding the outliers. Further, we tested for firm effects by including dummy variables for the top firms to check if the top firms might be driving the results. Finally, we attempted to measure government support for wind power by including Renewable Energy Country Attractiveness Index developed by Ernst & Young. As argued earlier, the policy instruments have largely influenced the market activity of the wind power industry. These policies, in the long run, may affect the innovation activity at a location. Our results persisted in all the robustness tests (details in the Online Appendix). 6. Discussion, implications and limitations The global dispersion of technology and the growing geographic spread of value creation define today’s knowledge economy (Mudambi, 2008; Bathelt and Cohendet, 2014). Firms are increasingly coordinating their knowledge creation activities across geographic space, especially in knowledge-intensive industries (Lorenzen, 2005). While existing innovative locations are becoming more globally connected (Bathelt et al., 2004; Lorenzen and Mudambi, 2013), new locations are increasingly entering these industries’ global innovation networks (Mudambi, 2008; Kumaraswamy et al., 2012). In this paper we map an industry’s technological network in which the various patented technologies of the industry are connected to the respective inventor locations. We argue that such a network shows the locations’ relative positions based on their patenting activity. The locations with high patent output are more central. The industry’s evolution makes this technological network dynamic with location centralities undergoing major changes. We find that the features of a location’s existing patent stock significantly influence its centrality. The results of our analyses show that the breadth of patented innovation at a location has a larger impact on its centrality in the technological network, than the depth of patented innovation. Next, we show that, after entering the emerging industry, the established innovative locations are more likely to increase depth of patented innovation in the industry’s technologies than breadth of patented innovation. On the other hand, when the emergent locations enter this industry, they are more likely to increase breadth of patented innovation in the industry’s technologies rather than depth of patented innovation. The argument is about the relative difference in focus. In absolute terms, the established locations are likely to have more breadth and depth. We therefore argue that, given the higher effect of breadth on centrality, emergent locations’ breadth focus in the focal industry, in comparison to established locations, helps them to rapidly achieve high centrality scores. Innovation networks evolve over time so the metrics associated with centrality constantly rise (Awate et al., 2015). Thus, our results do not imply that all the emergent locations eventually overtake established locations. As for the established locations, we would again like to stress the comparative nature of the arguments. We use pre-entry patent output that is a continuous measurement scale that identifies established and emergent locations. Therefore, our results imply that the more established a location is, the more likely it is to focus on depth than breadth for innovating in the new industry. This however does not imply that the centralities of established locations are lower on average. With their mix of depth and breadth, many of these locations may still be highly central in the industry’s global innovation network. Our results however do imply that, due to their strong focus on the existing technologies, their marginal rate of accession to centrality in the global innovation network is lower. Emergent locations on the other hand may face lower barriers to exploring new technologies and therefore have a higher marginal rate of accession to such centrality. Our evidence demonstrates the successful entry and survival of emergent locations in the industry’s global patented innovation network. This may suggest that the entry barriers for the locations entering the global innovation network of emerging industries may be lower. It may be indicative of a much wider geographic dispersal of emerging industry innovations. This claim however needs direct empirical tests, which can be pursued in future research. We would like to emphasize that emergent locations do not specifically refer to ‘emerging economy’ locations. After controlling for developed versus developing countries, the emergent innovative locations include all locations with lower total patent output prior to entering the emerging industry. In fact, our rich longitudinal dataset captures the pre-entry experiences of many of the locations that are today’s leaders in patent output. These are largely advanced economy locations. The initial pattern of dispersal predicted by Storper and Walker’s model of geographic industrialization would be supported by these emergent locations.4 Our study, with considerable empirical rigor, provides strong support to this pattern of global geographic dispersal of emerging industries, particularly in the context of innovation. While the innovative locations in our dataset are largely advanced economy locations, we do see some middle-income and low-income locations such as those in China, India, Southeast Asia and the Middle East. These newer locations do not yet have as heavy a concentration of inventors as in the advanced economies; however, the very presence of these locations with significant inventor populations is indicative of the catch-up processes under way. Two examples of this rapid catch-up in patented innovations, also highlighted in Figure 1, are Bangalore and Shanghai. These locations filed their first wind turbine patents in 2004 and 2006, respectively. However, by 2011, they are highly innovative with the average degree centralities in the 95th and 99th percentiles, respectively. Bangalore is known as a major global ICT cluster (Lorenzen and Mudambi, 2013), while Shanghai is rapidly becoming one of the world’s major manufacturing hubs (Wu and Radbone, 2005). Our analyses show that, with respect to the wind power industry, Bangalore is also active in electrical technology research. Bangalore’s ICT capabilities and established high-technology infrastructure have attracted R&D in other related industries. It has also attracted some of the world’s largest MNEs to set up major R&D subsidiaries. For example, GE’s John Welch Technology Center in Bangalore is the company’s largest R&D center outside of the USA. As shown in Figure 1, Bangalore connects to not just the core classes but also a number of non-core classes. On the contrary, San Francisco and Aalborg, which were the cradles of the industry and are equally innovative, are seen to connect primarily to the core classes. The appearance of middle-income and low-income locations in the global innovation networks of many industries is a feature of today’s knowledge economy. A number of such locations may currently be involved in standardized activities. However, over time, through learning and innovation catch-up, they may attract increasingly complex activities (Mudambi, 2008; Awate et al., 2015) and enter global innovation networks. We point out that the early strategy of such new locations should focus more on increasing breadth than depth. We use a novel methodology in which we model the industry technological network by mapping locations’ patented technological connections weighted by its inventor population. Rather than using overall patents (e.g., patent counts) as a broad measure of innovation, we focus on more nuanced measure that incorporates who invents where and what. This approach is far more information-rich than the traditional approach that uses overall patents. This is because one new patent from a location may actually correspond to the addition of more than one new inventor and/or more than one new technology class, thereby improving the measurement of the patenting activity at a location. The innovations studied are patented technological innovations. Although predominantly associated with industrialization, these are only one form of innovation. Patents are only one component of overall innovation activity, and do not explicitly capture the incremental continuous processes related to learning by doing and learning by interaction. Further, apart from patents, innovations are materialized as copyrights, trademarks, new products and/or their features. In addition to these publicized forms, innovations may be undisclosed and held within the firm. Therefore, while a location’s patent-based centrality is one of the important factors associated with regional development, it is by no means the only one. Another limitation of the study is a more direct empirical proxy for governmental support to the emerging industry as it may encourage a location’s innovation activity. While we use Ernst & Young RECAI index and its sub-index for wind power, which are internally consistent, a more direct measure such as financial support by the governments or the number of targeted new regulations would be useful to test the effect. We end by noting an important caveat with regard to emergent locations, particularly for those in emerging economies. In order to enter the technological networks of emerging industries, these locations must demonstrate the availability of basic knowledge resources like skilled labor, a population of entrepreneurial local firms in related industries and supporting business services. MNE subsidiaries are often the spark that jump-starts local innovation through knowledge spillovers as well as through providing connectivity to the global innovation network. Today these MNEs have a wealth of choices in terms of R&D location, so there is an element of chance in whether a potential location actually attracts enough MNE R&D investment to begin a virtuous cycle of local innovation. However, in the words of Louis Pasteur, ‘Chance favors the prepared mind.’ Funding This research was partially funded by the Indian School of Business – Ernst & Young Initiative for Emerging Market Studies. Supplementary material Supplementary data for this paper are available at Journal of Economic Geography online. Footnotes 1 The raw data totaled to 3652 inventor names on which we ran several rounds of disambiguation resulting in 2204 unique inventors. Hence, we are quite confident that we have the real inventor in overwhelming majority of the cases. This leads to 1.7 appearances of the same inventor name in the raw data, with a minimum of 1 entry to a maximum of 20 entries per inventor. 2 The analysis also allows for the movement of inventors (Saxenian, 1994). A new patent following an inventor’s move adds either completely new location or technology class nodes, or new ties between existing location and technology class nodes. 3 It should be noted that the interpretation of the comparative magnitudes of the breadth and depth effects are possible only after variable standardization. The comparative interpretation holds as both breadth and depth variables are standardized to have zero mean and unit variance. 4 Storper and Walker’s explanation of the shift of an industry is essentially one of regional institutions, while our focus is on patented technological innovations. References Ahuja G. ( 2000) Collaboration networks, structural holes, and innovation: a longitudinal study. Administrative Science Quarterly , 45: 425– 455. Google Scholar CrossRef Search ADS   Alcácer J. ( 2006) Location choices across the value chain: how activity and capability influence collocation. Management Science , 52: 1457– 1471. Google Scholar CrossRef Search ADS   Almeida P. ( 1996) Knowledge sourcing by foreign multinationals: patent citation analysis in the U.S. semiconductor industry. Strategic Management Journal , 17: 155– 165. Google Scholar CrossRef Search ADS   Almeida P., Kogut B. ( 1999) Localization of knowledge and the mobility of engineers in regional networks. Management Science , 45: 905– 917. Google Scholar CrossRef Search ADS   Arrow K. ( 1962) The economic implications of learning by doing. Review of Economic Studies , 29: 155– 173. Google Scholar CrossRef Search ADS   Awate S., Larsen M. M., Mudambi R. ( 2012) EMNE catch-up strategies in the wind turbine industry: is there a trade-off between output and innovation capabilities? Global Strategy Journal , 2: 205– 223. Google Scholar CrossRef Search ADS   Awate S., Larsen M. M., Mudambi R. ( 2015) Accessing vs sourcing knowledge: a comparative study of RD internationalization between emerging and advanced economy firms. Journal of International Business Studies  46: 63– 86. Google Scholar CrossRef Search ADS   Banker R., Cao Z., Menon N. M., Mudambi R. ( 2013) The Red Queen in action: the longitudinal effects of capital investments in the mobile telecommunications sector. Industrial and Corporate Change , 22: 1195– 1228. Google Scholar CrossRef Search ADS   Bathelt H., Boggs J. S. ( 2003) Towards a reconceptualization of regional development paths: is Leipzig’s media cluster a continuation of or a rupture with the past? Economic Geography , 79: 265– 293. Google Scholar CrossRef Search ADS   Bathelt H., Cohendet P. ( 2014) The creation of knowledge: local building, global accessing and economic development—toward an agenda. Journal of Economic Geography , 14: 869– 882. Google Scholar CrossRef Search ADS   Bathelt H., Malmberg A., Maskell P. ( 2004) Clusters and knowledge: local buzz, global pipelines and the process of knowledge creation. Progress in Human Geography , 28: 31– 56. Google Scholar CrossRef Search ADS   Bel G., Fageda X. ( 2008) Getting there fast: globalization, intercontinental flights and location of headquarters. Journal of Economic Geography , 8: 471– 495. Google Scholar CrossRef Search ADS   Bell G. G., Zaheer A. ( 2007) Geography, networks, and knowledge flow. Organization Science , 18: 955– 972. Google Scholar CrossRef Search ADS   Beugelsdijk S., Mudambi R. ( 2013) MNEs as border-crossing multi-location enterprises: the role of discontinuities in geographic space. Journal of International Business Studies , 44: 413– 426. Google Scholar CrossRef Search ADS   Bierly P., Chakrabarti A. ( 1996) Knowledge strategies in the U.S. pharmaceutical industry. Strategic Management Journal , Winter Special Issue, 17: 123– 135. Google Scholar CrossRef Search ADS   Borgatti S., Everett M., Freeman L. ( 2002) UCINET 6 for Windows: Software for Social Network Analysis . Harvard: Analytic Technologies. Boschma R., Frenken K. ( 2011) The emerging empirics of evolutionary economic geography. Journal of Economic Geography , 11: 295– 307. Google Scholar CrossRef Search ADS   Bresnahan T., Gambardella A., Saxenian A. ( 2001) “Old economy” inputs for “new economy” outcomes: cluster formation in the New Silicon Valleys. Industrial and Corporate Change , 10: 835– 860. Google Scholar CrossRef Search ADS   Buen J. ( 2006) Danish and Norwegian wind industry: the relationship between policy instruments, innovation and diffusion. Energy Policy , 34: 3887– 3897. Google Scholar CrossRef Search ADS   Cantwell J. A., Mudambi R. ( 2011) Physical attraction and the geography of knowledge sourcing in multinational enterprises. Global Strategy Journal , 1: 206– 232. Google Scholar CrossRef Search ADS   Cantwell J., Mudambi R. ( 2005) MNE competence-creating subsidiary mandates. Strategic Management Journal , 26: 1109– 1128. Google Scholar CrossRef Search ADS   Cardwell D. ( 2013) Renewed tax credit buoys wind-power projects. The New York Times, 21 March. Christensen C. ( 1993) The rigid disk drive industry: a history of commercial and technological turbulence, Business History Review , 67: 531– 588. Google Scholar CrossRef Search ADS   Christensen C. M., Bower J. L. ( 1996) Customer power, strategic investment, and the failure of leading firms. Strategic Management Journal , 17: 197– 218. Google Scholar CrossRef Search ADS   Coe N. M., Bunnell T. ( 2003) “Spatializing” knowledge communities: towards a conceptualization of transnational innovation networks. Global Networks , 3: 437– 456. Google Scholar CrossRef Search ADS   Coe N. M., Dicken P., Hess M. ( 2008) Global production networks: realizing the potential. Journal of Economic Geography , 8: 271– 295. Google Scholar CrossRef Search ADS   Coe N. M., Hess M., Yeung H. W. C., Dicken P., Henderson J. ( 2004) ‘Globalizing’ regional development: a global production networks perspective. Transactions of the Institute of British Geographers , 29: 468– 484. Google Scholar CrossRef Search ADS   Cohen W., Levinthal D. ( 1990) Absorptive capacity: a new perspective on learning and innovation. Administrative Science Quarterly , 35: 128– 152. Google Scholar CrossRef Search ADS   Cyert R., March J. ( 1963) A Behavioral Theory of the Firm . Prentice-Hall, NJ: Englewood Cliffs. Econstats. ( 2012) International Internet bandwidth. Econstats.com, 25 October. Faust K. ( 1997) Centrality in affiliation networks. Social Networks , 19: 157– 191. Google Scholar CrossRef Search ADS   Feldman M., Lendel I. ( 2010) Under the lens: the geography of optical science as an emerging industry. Economic Geography , 86: 147– 171. Google Scholar CrossRef Search ADS   Florida R., Kenney M. ( 1990) Silicon Valley and Route 128 won’t save us. California Management Review , 33: 68– 85. Google Scholar CrossRef Search ADS   Funk R. J. ( 2014). Making the most of where you are: geography, networks, and innovation in organizations. Academy of Management Journal , 57: 193– 222. Google Scholar CrossRef Search ADS   Garud R., Karnøe P. ( 2003) Bricolage versus breakthrough: distributed and embedded agency in technology entrepreneurship Research Policy , 32: 277– 300. Google Scholar CrossRef Search ADS   Gates B. ( 2011) Taking energy storage to a higher level. The Gates Notes, 16 November. Grant R. ( 1996) Prospering in dynamically competitive environments: organizational capability as knowledge integration. Organization Science , 7: 375– 386. Google Scholar CrossRef Search ADS   Griliches Z. ( 1979) Issues in assessing the contribution of research and development to productivity growth, Bell Journal of Economics , 10: 92– 116. Google Scholar CrossRef Search ADS   Griliches Z. ( 1990) Patent statistics as economic indicators: a survey, Journal of Economic Literature , 28: 1661– 1707. GWEC. ( 2012) Global wind report: annual market update 2011. Global Wind Energy Council, 24 August. Hagedoorn J., Cloodt M. ( 2003) Measuring innovative performance: is there an advantage in using multiple indicators? Research Policy , 32: 1365– 1379. Google Scholar CrossRef Search ADS   Hamel G., Prahalad C. K. ( 1994) Competing for the Future . Boston: Harvard Business School Press. Hannigan T. J., Cano-Kollmann M., Mudambi R. ( 2015) Thriving innovation amidst manufacturing decline: the Detroit auto cluster and the resilience of local knowledge production. Industrial and Corporate Change , 24: 613– 634. Google Scholar CrossRef Search ADS   Henderson J., Kuncoro A., Turner M. ( 1995) Industrial development in cities. Journal of Political Economy , 103: 1067– 1085. Google Scholar CrossRef Search ADS   Hsiao C. ( 1986) Analysis of Panel Data . Cambridge, UK: Cambridge University Press. ITU. ( 2005) International Internet statistics. International Telecommunications Union, 25 October. Jacobs J. ( 1969) The Economy of Cities . New York, NY: Random House. Jaffe A. ( 1989) Real effect of academic research, American Economic Review , 79: 957– 970. Jaffe A. B. ( 1986) Technological opportunity and spillovers of R&D: evidence from firms’ patents, profits, and market value. American Economic Review , 76: 984– 1001. Jaffe A., Trajtenberg M., Henderson R. ( 1993) Geographic localization of knowledge spillovers as evidenced by patent citations, Quarterly Journal of Economics , 108: 577– 598. Google Scholar CrossRef Search ADS   Katila R., Ahuja G. ( 2002) Something old, something new: a longitudinal study of search behavior and new product introduction. Academy of Management Journal , 45: 1183– 1194. Google Scholar CrossRef Search ADS   Kher U. ( 2010) A call for collaboration, Nature , 466: S21– S22. Google Scholar CrossRef Search ADS   Klepper S. ( 1996) Entry, exit, growth, and innovation over the product life cycle. American Economic Review , 86: 562– 583. Kodama F. ( 1992) Technology fusion and the new R&D. Harvard Business Review , 70: 70– 78. Kogut B., Zander U. ( 1992) Knowledge of the firm, combinative capabilities, and the replication of technology. Organization Science , 3: 383– 397. Google Scholar CrossRef Search ADS   Kuhn T. ( 1962) The Structure of Scientific Revolutions . Chicago, IL: University of Chicago Press. Kumaraswamy A., Mudambi R., Saranga H., Tripathy A. ( 2012) Catch-up strategies in the Indian auto component industry: domestic firms’ responses to market liberalization. Journal of International Business Studies , 43: 368– 395. Google Scholar CrossRef Search ADS   Lahiri N. ( 2010). Geographic distribution of R&D activity: how does it affect innovation quality? Academy of Management Journal , 53: 1194– 1209. Google Scholar CrossRef Search ADS   Leonard-Barton D. ( 1995) Wellsprings of Knowledge . Boston: Harvard Business School Press. Levinthal D., March J. ( 1981) A model of adaptive organizational search. Journal of Economic Behavior and Organization , 2: 307– 333. Google Scholar CrossRef Search ADS   Lorenzen M. ( 2005) Introduction: knowledge and geography. Industry and Innovation , 12: 399– 407. Google Scholar CrossRef Search ADS   Lorenzen M., Mudambi R. ( 2013) Clusters, connectivity and catch-up: Bollywood and Bangalore in the global economy. Journal of Economic Geography , 13: 501– 534. Google Scholar CrossRef Search ADS   March J. ( 1991) Exploration and exploitation in organizational learning, Organization Science , 2: 71– 87. Google Scholar CrossRef Search ADS   Markard J, Petersen R. 2009. The offshore trend: structural changes in the wind power sector. Energy Policy , 37: 3545– 3556. Google Scholar CrossRef Search ADS   Marshall A. ( 1890) Principles of Economics . London, UK: Macmillan. Martin R., Sunley P. ( 2007) Complexity thinking and evolutionary economic geography. Journal of Economic Geography , 7: 573– 601. Google Scholar CrossRef Search ADS   Martin R., Sunley P. ( 2011) Conceptualizing cluster evolution: beyond the life cycle model? Regional Studies , 45: 1299– 1318. Google Scholar CrossRef Search ADS   Maskell P., Malmberg A. ( 2007) Myopia, knowledge development and cluster evolution. Journal of Economic Geography , 7: 603– 618. Google Scholar CrossRef Search ADS   Mudambi R. ( 2008) Location, control and innovation in knowledge-intensive industries. Journal of Economic Geography , 8: 699– 725. Google Scholar CrossRef Search ADS   Mudambi R., Hannigan T., Kline W. ( 2012) Advancing science on the knife’s edge: integration and specialization in management Ph.D. programs. Academy of Management Perspectives , 26: 83– 105. Google Scholar CrossRef Search ADS   Mudambi R., Santangelo G. D. ( 2016) From shallow resource pools to emerging clusters: the role of multinational enterprise subsidiaries in peripheral areas. Regional Studies , 50: 1965– 1979. Google Scholar CrossRef Search ADS   Mudambi R., Swift T. ( 2014) Knowing when to leap: transitioning between exploitative and explorative R&D. Strategic Management Journal , 35: 126– 145. Google Scholar CrossRef Search ADS   Musgrove P. ( 2010) Wind Power . Cambridge, UK: Cambridge University Press. Narin F., Noma E., Perry R. ( 1987) Patents as indicators of corporate technological strength. Research Policy , 16: 143– 155. Google Scholar CrossRef Search ADS   Narula R. ( 2014) Exploring the paradox of competence-creating subsidiaries: balancing bandwidth and dispersion in MNEs. Long Range Planning , 47, 4– 15. Google Scholar CrossRef Search ADS   Neffke F., Svensson Henning M., Boschma R., Lundquist K., Olander L. ( 2011) The dynamics of agglomeration externalities along the life cycle of industries. Regional Studies , 45: 49– 65. Google Scholar CrossRef Search ADS   Nelson R., Winter S. ( 1982) An Evolutionary Theory of Economic Change . Cambridge: Harvard University Press. Nielsen K. 2010. Technological trajectories in the making: two cases from the contemporary history of wind power. Centaurus , 52: 175– 205. Google Scholar CrossRef Search ADS   Paruchuri S., Awate S. ( 2016) Organizational knowledge networks and local search: the role of intra‐organizational inventor networks. Strategic Management Journal , 38: 657– 675. Google Scholar CrossRef Search ADS   Patel P., Pavitt K. ( 1997) The technological competencies of the world’s largest firms: complex and path-dependent, but not much variety. Research Policy , 26: 141– 156. Google Scholar CrossRef Search ADS   Pedersen T, Larsen M. ( 2009) Vestas Wind Systems A/S—Exploiting Global RD Synergies. Case: 9B09M079, 26 November, pp. 1–17. London, UK/Canada: Ivey Management Services. Penrose E. G. ( 1959) The Theory of the Growth of the Firm . New York: Wiley. Podolny J. M., Stuart T. E. ( 1995) A role-based ecology of technological change. American Journal of Sociology , 100: 1224– 1260. Google Scholar CrossRef Search ADS   Porter M. ( 1980) Competitive Strategy . New York (NY): The Free Press. Porter M. E., Stern S. ( 2000) Measuring the “Ideas” Production Function: Evidence from International Patent Output. Working Paper No. w7891, National Bureau of Economic Research. Pouder R., St. John C. H. ( 1996) Hot spots and blind spots: geographical clusters of firms and innovation. Academy of Management Review , 21: 1192– 1225. Rajsekhar B., Van Hulle F., Jansen J. C. ( 1999) Indian wind energy programme: performance and future directions. Energy Policy , 27: 669– 678. Google Scholar CrossRef Search ADS   Romer P. ( 1986) Increasing returns and long-run growth. Journal of Political Economy , 94: 1002– 1037. Google Scholar CrossRef Search ADS   Rosenkopf L., Nerkar A. ( 2001) Beyond local search: boundary-spanning, exploration, and impact in the optical disk industry. Strategic Management Journal , 22: 287– 306. Google Scholar CrossRef Search ADS   Saxenian A. ( 1994) Regional Advantage: Culture and Competition in Silicon Valley and Route 128 . Cambridge (MA): Harvard University Press. Scalera, V., Perri, A., Hannigan, T. J. (2018). Knowledge connectedness within and across home country borders: spatial heterogeneity and the technological scope of firm innovations. Journal of International Business Studies, forthcoming. Schumpeter J. A. ( 1934) The Theory of Economic Development: An Inquiry into Profits, Capital, Credit, Interest, and the Business Cycle . Cambridge (MA): Harvard University Press. Soda G., Zaheer A. ( 2012) A network perspective on organizational architecture: performance effects of the interplay of formal and informal organization. Strategic Management Journal , 33: 751– 771. Google Scholar CrossRef Search ADS   Storper M. ( 1995) The resurgence of regional economies, ten years later: the region as a nexus of untraded interdependencies. European Urban and Regional Studies , 2: 191– 221. Google Scholar CrossRef Search ADS   Storper M., Walker R. ( 1989) The Capitalist Imperative: Territory, Technology and Industrial Growth . New York (NY): Basil Blackwell Inc. Sturgeon T., Van Biesebroeck J., Gereffi G. ( 2008) Value chains, networks and clusters: reframing the global automotive industry. Journal of Economic Geography , 8: 297– 321. Google Scholar CrossRef Search ADS   Teece D. J. ( 1994) Information sharing, innovation, and antitrust. Antitrust Law Journal , 62: 465– 481. Tripsas M. ( 2009) Technology, identity and inertia through the lens of the ‘Digital Photography Company’. Organization Science , 20: 441– 460. Google Scholar CrossRef Search ADS   Turkina, E., Van Assche, A. Kali, R. (2016). Structure and evolution in global cluster networks: Evidence from the aerospace industry. Journal of Economic Geography, 16: 1211–1234. Tushman M. L., Murmann P. ( 1998). Dominant designs, innovation types and organizational outcomes, Research in Organizational Behavior , 20: 231– 266. UCLA. ( 2013) Stata FAQ: how can I perform the likelihood ratio, Wald, and Lagrange multiplier (score) test in Stata? UCLA: Statistical Consulting Group, 25 October. Van der Panne G. ( 2004) Agglomeration externalities: Marshall versus Jacobs. Journal of Evolutionary Economics , 14: 593– 604. Google Scholar CrossRef Search ADS   Wu J., Radbone I. ( 2005) Global integration and intra-urban determinants of foreign direct investment in Shanghai. Cities , 22: 275– 286. Google Scholar CrossRef Search ADS   WWEA. ( 2012) World Wind Energy Report 2011. World Wind Energy Association, July 1. Yeung H., Coe N. ( 2015) Toward a dynamic theory of global production networks. Economic Geography , 91: 29– 58. Google Scholar CrossRef Search ADS   Zhu S., He C., Zhou Y. ( 2017) How to jump further and catch up? Path-breaking in an uneven industry space. Journal of Economic Geography , lbw047, 1– 25. © The Author (2017). Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

Journal of Economic GeographyOxford University Press

Published: Mar 1, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 12 million articles from more than
10,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Unlimited reading

Read as many articles as you need. Full articles with original layout, charts and figures. Read online, from anywhere.

Stay up to date

Keep up with your field with Personalized Recommendations and Follow Journals to get automatic updates.

Organize your research

It’s easy to organize your research with our built-in tools.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve Freelancer

DeepDyve Pro

Price
FREE
$49/month

$360/year
Save searches from Google Scholar, PubMed
Create lists to organize your research
Export lists, citations
Access to DeepDyve database
Abstract access only
Unlimited access to over
18 million full-text articles
Print
20 pages/month
PDF Discount
20% off